Structures of the atlastin GTPase provide insight
into homotypic fusion of endoplasmic reticulum
Xin Biana,b,1, Robin W. Klemmc,1, Tina Y. Liuc,1, Miao Zhanga,b, Sha Suna,b, Xuewu Suia,b, Xinqi Liub,d,2, Tom A. Rapoportc,2,
and Junjie Hua,b,2
Departments ofaGenetics and Cell Biology anddBiochemistry and Molecular Biology, College of Life Sciences, andbTianjin Key Laboratory of Protein Sciences,
Nankai University, Tianjin 300071, China; andcHoward Hughes Medical Institute, Department of Cell Biology, Harvard Medical School, Boston, MA 02115
Contributed by Tom A. Rapoport, January 31, 2011 (sent for review January 26, 2011)
The generation of the tubular network of the endoplasmic
reticulum (ER) requires homotypic membrane fusion that is
mediated by the dynamin-like, membrane-bound GTPase atlastin
(ATL). Here, we have determined crystal structures of the cytosolic
segment of human ATL1, which give insight into the mechanism
of membrane fusion. The structures reveal a GTPase domain and
athree-helix bundle, connected by a linker region. One structure
corresponds to a prefusion state, in which ATL molecules in
apposing membranes interact through their GTPase domains to
form a dimer with the nucleotides bound at the interface. The
other structure corresponds to a postfusion state generated after
GTP hydrolysis and phosphate release. Compared with the pre-
fusion structure, the three-helix bundles of the two ATL molecules
undergo a major conformational change relative to the GTPase
domains, which could pull the membranes together. The proposed
fusion mechanism is supported by biochemical experiments and
fusion assays with wild-type and mutant full-length Drosophila
ATL. These experiments also show that membrane fusion is facil-
itated by the C-terminal cytosolic tails following the two trans-
membrane segments. Finally, our results show that mutations in
ATL1 causing hereditary spastic paraplegia compromise homotypic
protein structure|membrane remodeling|organelle shaping|spastic
paraplegia gene 3A|endoplasmic reticulum network formation
and a peripheral network of tubules and sheets (1, 2). The ER
network is very dynamic, with tubules continuously forming and
collapsing (3–5). The linkage of ER tubules into a network
requires the fusion of identical membranes, but how such
homotypic fusion occurs is poorly understood. By contrast, het-
erotypic fusion of viral and cellular membranes and of in-
tracellular transport vesicles with target membranes has been
intensively studied. In viral fusion, the membranes are pulled
together by an irreversible conformational change of a single
protein (6). In intracellular fusion, three t-SNARE proteins in
one membrane and a v-SNARE partner in the other zipper up to
form a four-helix bundle in the fused lipid bilayer, a process
that is facilitated by additional proteins (7–10). The ATPase N-
ethylmaleimide-sensitive fusion protein (NSF) and its cofactors
then act together to disassemble the SNARE complex for the
next round of fusion (11, 12).
Recent work has implicated the membrane-bound atlastin
(ATL) GTPases in the homotypic fusion of ER tubules (13, 14).
These proteins belong to the dynamin family of GTPases (15)
and appear to exist in all metazoans, with many organisms
expressing several isoforms (16). The ATLs are anchored in the
membrane by two closely spaced transmembrane (TM) seg-
ments, exposing both the GTPase domain and the C-terminal tail
to the cytosol (Fig. 1A). They localize to ER tubules and interact
with the reticulons and DP1/Yop1p, proteins implicated in
n all eukaryotes, the endoplasmic reticulum (ER) is a contin-
uous membrane system that comprises the nuclear envelope
shaping ER tubules (13). A role for the ATLs in fusion is sug-
gested by the observation that their depletion or the overexpression
of dominant-negative forms reduces the interconnection of ER
tubules in tissue culture cells (13). In addition, antibodies to ATL
inhibit ER network formation in vitro (13). Finally, and most
convincingly, proteoliposomes containing purified ATL undergo
GTP-dependent fusion (14). Plant and yeast cells do not possess
ATLs, but a similar GTPase, called Sey1p in Saccharomyces cer-
evisiae and RHD3 in Arabidopsis thaliana, may have an analogous
Mutations in ATL1 cause the most common type of early-
onset hereditary spastic paraplegia (HSP; ref. 17). The disease is
characterized by progressive spasticity and weakness of the lower
limbs due to a length-dependent abnormality of corticospinal
axons. The expression of ATL1 with disease mutations in tissue
culture cells often results in abnormally long, nonbranched ER
tubules, consistent with ER fusion defects (13, 16).
Another member of the dynamin family, called mitofusin in
mammals and Fzo1p in yeast, also mediates homotypic mem-
brane fusion (18, 19); it has the same membrane topology as the
ATLs and fuses outer mitochondrial membranes. Other family
members, including dynamin-1, mediate the fission of mem-
branes (20–24). However, other members, including the guany-
late binding protein 1 (GBP1; refs. 25–27), the myxovirus
resistance protein 1 (MxA or MX1; ref. 28), and the bacterial
dynamin-like protein (BDLP; refs. 29 and 30), have poorly
Here, we elucidate how ATL mediates homotypic membrane
fusion. The proposed mechanism involves drastic conformational
changes of interacting ATL molecules and explains the effect of
mutations in ATL1 that cause HSP.
Crystal Structures of the Cytosolic Domain of ATL1. To investigate
the mechanism of ATL1-mediated membrane fusion, we de-
termined crystal structures of its cytosolic domain (cytATL1).
Residues 18–447 (truncation before the first TM region; Fig. 1A)
were expressed in Escherichia coli, purified, and crystallized in
the presence of GDP. A hexagonal crystal form was used to
determine a structure at 2.8-Å resolution by multi-wavelength
Author contributions: X.B., R.W.K., T.Y.L., X.L., T.A.R., and J.H. designed research; X.B.,
R.W.K., T.Y.L., M.Z., S.S., X.S., and X.L. performed research; X.B., R.W.K., T.Y.L., M.Z., S.S.,
X.L., T.A.R., and J.H. analyzed data; and T.A.R. and J.H. wrote the paper.
The authors declare no conflict of interest.
Data deposition: The atomic coordinates and structure factors have been deposited in the
Protein Data Bank, www.pdb.org (PDB ID codes 3QNU and 3QOF).
1X.B., R.W.K., and T.Y.L. contributed equally to this work.
2To whom correspondence may be addressed. E-mail: firstname.lastname@example.org, tom_rapoport@
hms.harvard.edu, or email@example.com.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
| March 8, 2011
| vol. 108
| no. 10www.pnas.org/cgi/doi/10.1073/pnas.1101643108
anomalous diffraction (MAD; Table S1). Only one ATL1 mol-
ecule was present in the asymmetric unit, but a crystallographic
dimer was seen along one of the twofold symmetry axes (crystal
packing shown in Fig. S1A). The dimer is likely of physiological
relevance, as discussed below.
CytATL1 consists of an N-terminal GTPase domain and a
three-helix bundle, which are connected by a linker region (Fig.
1B). In the dimer, the GTPase domains interact with one another,
such that the nucleotide binding sites face each other. The linker
top of the three-helix bundle of one ATL1 molecule (3HB-1) to
bind to the bottom of the GTPase domain of the other molecule
(GTPase-2). The C termini of the two ATL1 molecules are only
∼10 Å apart, indicating that, in the full-length protein, the fol-
lowing TM segments would have to sit in the same membrane.
A different conformation of the cytATL1 was obtained when
the protein was crystallized with GDP and high concentrations of
inorganic phosphate (Pi; Fig. 1C). This structure was determined
at 2.8-Å resolution from primitive orthorhombic crystals by
molecular replacement, using the GTPase domain of the first
structure as a search model (Table S1). A crystallographic and a
noncrystallographic dimer are seen (crystal packing shown in
Fig. S1B), in which the two GTPase domains interact as in the
structure derived from the hexagonal crystals. In contrast to
the first structure, the three-helix bundles are associated with
the GTPase domain of the same molecule (3HB-1 interacts with
GTPase-1; Fig. 1C) and point in opposite directions. This result
implies that, with the full-length protein, the two ATL1 mole-
cules would sit in apposing membranes. Based on the difference
between the two structures, we hypothesize that the second
structure corresponds to a membrane-tethered, prefusion state,
and the first structure corresponds to a postfusion state in which
the membranes have already fused. With this assumption, we will
use the terms “prefusion” and “postfusion” structures. Although
no density for Piwas observed in the prefusion structure, the
conformational change between the two states is likely triggered
by Pirelease during the nucleotide hydrolysis cycle (see below).
Comparison with Other Dynamin-Like Proteins and GTP Hydrolysis.
The ATL1 dimer is likely of physiological significance because
other dynamin-like proteins also form dimers of their GTPase
domains with the nucleotides bound at the interface (Fig. S2A).
ATL1 is most similar to GBP1 (Fig. S2 A and B), notably in the
active sites (Fig. S3C), although dimer formation involves dif-
ferent interface residues. The prefusion structure of ATL1
resembles GBP1 crystallized with the nonhydrolyzable GTP an-
alog (25), in that the helical domains point in opposite directions
(Fig. S2B). This similarity suggests that the prefusion state of
ATL1 is close to the GTP-bound conformation.
The GTPase domain is composed of a central β-sheet sur-
rounded by six α-helices, and the GDP molecule is coordinated
mainly by four conserved elements (Fig. S3A): the P-loop (G1
motif), the switch 1 region (G2 motif), the switch 2 region (G3
motif), and the G4 motif. The two residues of ATL1’s G4 motif
(R217 and D218) are unique to the subclass of the dynamin
family that includes ATLs, Sey1p, RHD3, and GBPs (13, 31). As
expected, a mutation in the P loop (K80A) affects the enzymatic
activity of the GTPase (Table S2). However, some uncommon
residues may also be involved in GTP hydrolysis, including res-
idue E117 in switch 1—which in human GBP1 may correspond
to the catalytically important S73 residue—and Q148 in switch 2.
Their mutations to Ala result in increased binding of GTPγS and
reduced GTPase activity (Fig. S4A and Tables S2 and S3). R77 in
ATL1 corresponds to the critical “Arg-finger” in GBP1, but it
points away from GDP in our structure (Fig. S3B), suggesting
that it may reorient upon GTP binding. Thus, ATL1 may un-
dergo at least two conformational changes during the GTP hy-
drolysis cycle—one in the active site during hydrolysis of the
γ-phosphate bond and one involving the reorientation of the
three-helix bundle during Pirelease.
Dimer Formation and Nucleotide Binding. The nucleotides are fairly
well buried in the structures, suggesting that nucleotide binding
is linked to dimer formation. Indeed, purified cytATL1 behaved
as a monomer in the absence of nucleotide, as judged by ana-
lytical ultracentrifugation (AUC; Fig. 2A, Upper) and gel filtra-
tion (Fig. 2B), whereas it formed a dimer in the presence of GDP
or GTPγS. The dimer may be somewhat less stable in GDP, as
indicated by the broadened peak in AUC (Fig. 2A). Mutating
R77 at the interface between the protomers (Fig. S5A) to Glu
abolished dimer formation, as judged by AUC (Fig. 2A, Lower),
and it reduced GTPγS affinity, as measured by isothermal titra-
tion calorimetry (ITC; Fig. S6). Mutation of residue L274 at the
dimer interface (Fig. S5A) to Ala reduced the association of the
protomers more moderately (Fig. S7); the GTPase activity was
particularly affected at low protein concentrations (Fig. S4B). A
similar, concentration-dependent effect on the GTPase activity
was also seen with another interface mutant (Q183; Figs. S5B
and S4B). These results suggest that GTP-dependent dimer
formation may link ATL molecules in apposing membranes
during the first step in membrane fusion.
1338 347 432
showing the domains of ATL1. 3HB, three-helix bundle; TMs, TM segments;
CT, cytosolic tail. (B) Structure of the GDP-bound form of ATL1, corre-
sponding to the postfusion state. The protomers in the dimer are shown in
green and purple cartoon representation, superimposed on a space-filling
model. The linkers (L-1 and L-2) between the GTPase domains and 3HBs (in
pale colors) are highlighted. GDP is shown in orange stick representation,
and magnesium ion is shown as a yellow sphere. The α-helices of the 3HBs
are numbered. Right shows a different view of the dimer. (C) Structure of
ATL1 crystallized with GDP and Pi, corresponding to the prefusion state.
Lower shows a different view of the dimer.
Structures of the cytosolic domain of human ATL1. (A) Scheme
Bian et al. PNAS
| March 8, 2011
| vol. 108
| no. 10
Conformational Change Induced by PiRelease. The structures sug-
gest how Pirelease may cause a conformational change of ATL1.
In the prefusion structure, the GTPase domain binds to the
three-helix bundle of the same ATL1 molecule. The α6-, α2-, and
α3-helices of the GTPase domain surround and interact with the
top of the α7-helix of the bundle (Fig. 3 A and B). Major contacts
include residues L192, Y196, and L199 in α3 and M347 and L348
in α7. Residue M347 of the α7-helix also associates with M169 of
the α2-helix, close to the site where α2 interacts with a kink in the
α3-helix (L192). In the postfusion structure, the α3-helix is no
longer kinked; it now interacts with the α7-helix in the bundle of
the other protomer, using in part the same residues as in the
prefusion state (L192 in α2 and M347 in α7; Fig. 3 A–C). In
addition, the GTPase domain interacts with the loop between α8
and α9 of the three-helix bundle (K406 and M408). Pirelease
from the active sites could trigger this conformational change via
movements of switch 2, which could affect the interaction of the
adjacent α2-helix with the α3-helix and dislodge the α7-helix of
the three-helix bundle, allowing interactions with the other
A trypsin-protection assay was used to confirm the confor-
mational change inferred from the structures. In the absence of
nucleotide, wild-type cytATL1 was readily proteolyzed (Fig. 3D,
lane 2 vs. 1; see Fig. S8A for controls). In the presence of GDP,
a protected fragment was generated (lane 3). The fragment
corresponds to the N terminus of the protein, with trypsin
cleavage occurring at K345 in the linker region (confirmed by
mass spectrometry analysis; data not shown); when K345 was
mutated to Glu, no cleavage was observed (Fig. S8B). In the
postfusion structure, the loop residue K345 is partially exposed
(Fig. 3 A and C), explaining why it is accessible to trypsin.
In the presence of GTPγS, which is expected to lock ATL1 in
the prefusion state, linker cleavage was largely suppressed, and
instead the full cytosolic domain became resistant to trypsin di-
gestion (Fig. 3D, lane 4). In this state, the cleavage site K345 is
embedded in the α7-helix of the three-helix bundle (Fig. 3 A and
B) and becomes inaccessible to trypsin. As expected, the mo-
nomeric, nucleotide-binding-deficient R77E mutant was readily
degraded by trypsin, even in the presence of GDP or GTPγS
(Fig. 3D, lanes 11 and 12). A similar effect was seen with a mu-
tant, in which residues M347 and L348 were converted to Ala
(Fig. 3D, lanes 15 and 16), consistent with a weakened in-
teraction of the three-helix bundle with the GTPase domain (Fig.
3 B and C). Linker cleavage was prevented not only with GTPγS,
but also when the transition state of GTP hydrolysis was mim-
icked by addition of GDP and AlF3(Fig. S8C) or when the
posthydrolysis state was generated by addition of GDP and Pi
(Fig. 3D, lane 7). Thus, Pirelease during the nucleotide hydro-
lysis cycle appears to trigger the transition from the prefusion to
the postfusion conformation.
The same conformational change was observed with full-
length Drosophila ATL in intact membranes. The full-length
protein was expressed as a GST fusion in E. coli, purified in
Triton X-100, and—after cleavage and removal of the GST tag—
reconstituted into proteoliposomes. As with the cytosolic do-
main, trypsin completely degraded the protein in the absence of
nucleotide (Fig. 3E, lane 2 vs. 1), but generated a fragment in the
presence of GDP (lane 3). In the presence of GTPγS or GDP
plus high concentrations of Pi, the protein became resistant to
proteolysis (lanes 4 and 5).
structures, we performed cross-linking experiments. The three-
in the presence of GDP, but not in GMPPNP or the absence of
nucleotide, as demonstrated by the formation of a cross-linked
dimer when the ATL mutant L357C was treated with the bi-
functional cysteine cross-linker bis-maleimidoethane (BMOE;
see Fig. S9). Furthermore, as expected from the GDP structure in
which residue L192 of one ATL molecule came close to L348 of
the other ATL molecule (Fig. 3C), cysteines introduced at these
positions can form a disulfide bridge (Fig. 3G, lane 3). Again, no
cross-linking was seenwithGMPPNP (lane 4)or inthe absence of
nucleotide (lane 2). These data strongly support the postulated
Testing ATL Mutants in a Fusion Assay. Next, we tested the effect of
mutations on the membrane fusion activity of full-length Dro-
sophila ATL. The proteins were reconstituted at equal concen-
trations into donor and acceptor proteoliposomes (Fig. S10). The
donor vesicles contained lipids labeled with two fluorophores at
quenching concentrations; the fusion with unlabeled acceptor
vesicles resulted in dilution and dequenching. Wild-type Dro-
(Fig. 4A), as reported for a GST-fusion of ATL (14), although
without the GST tag, the protein was significantly more active. No
fusion was seen with GDP or the absence of Mg2+(Fig. 4A).
and GTP binding observed with the cytosolic domain. The same
(L249A, corresponding to L274A in hATL1; Fig. 4A). Further-
more, the addition of the purified cytosolic domain of Drosophila
ATL effectively inhibited fusion (Fig. 4B). These data support the
ideathatdimer formationfrom ATLmolecules sittingindifferent
membranes is required for membrane fusion. GTPase activity of
the ATL molecules is also needed.
Mutations in switch 1 (E92A) or switch 2 (Q123A), corre-
sponding to E117A and Q148A in hATL1, reduced the GTPase
activity of the cytosolic domain (Fig. S4A and Table S2) and the
fusion activity of the full-length protein (Fig. 4C). As observed
wt + GDP
wt + GTPγS
sed coefficient [S]
R77E + GDP
R77E + GTPγS
sed coefficient [S]
wt + GDP
wt + GTPγS
type (wt) cytATL1 (theoretical molecular mass 49.8 kDa) and of the R77E
mutant (both at 0.05 mM) were determined by analytical ultracentrifugation
in the presence of the indicated nucleotides. The estimated molecular
masses are given above the peaks (in kDa). (B) Wild-type CytATL1 was sub-
jected to gel filtration on a Superdex-200 column in the presence of the
Nucleotide-dependent dimerization of ATL1. (A) The size of wild-
| www.pnas.org/cgi/doi/10.1073/pnas.1101643108 Bian et al.
(14), even mutants with moderately reduced GTPase activity
showed drastically reduced fusion activity. Finally, mutant
M322A/L323A (M347A/L348A in hATL1), in which the in-
teraction between the three-helix bundle and the GTPase do-
main is disturbed, and mutant K320E (K345E in hATL1), which
has high GTPase activity (Fig. S4A) but is expected to disrupt
a kink between the linker and the three-helix bundle (Fig. 3C),
were also defective in fusion (Fig. 4D). Mutation E328K (E353K
in hATL1), which is also located in the kink, inhibited fusion less
strongly (Fig. 4D). Together, these results suggest that the con-
clusions derived from cytATL1 are functionally significant and
relevant to the fusion reaction observed with the full-length
protein. We also found that the deletion of the C-terminal tail
following the two TM segments (truncation at residue 476)
drastically reduced, but did not abolish, fusion (Fig. 4D). Thus,
the C-terminal tail also plays an important role, but whether it is
involved in membrane tethering or fusion remains to be tested by
more direct assays.
Disease-Causing ATL1 Mutations. Most of the published ATL1
mutations causing HSP (see refs. 32 and 33) localize to two “hot
spots” (Fig. 4 E and F): in the GTPase domain near the dimer
interface and in the three-helix bundle, mostly in a region close
to the GTPase domain. Several of the changes are expected to
affect dimer formation and GTPase activity, or the conforma-
tional change of ATL during fusion. Disease mutations in
cytATL1 had GTPase activity that varied from almost wild-type
(H247P) to extremely low levels (F151S; Fig. S4C). The corre-
sponding Drosophila ATL mutants (H222P and F126S) showed
reduced fusion activity, roughly correlated with their GTPase
activity (Fig. 4C). Mutation I315S (I290S in Drosophila) localizes
to the periphery of the postfusion structure, but comes close to
the three-helix bundle in the prefusion structure (Fig. 4 E and F,
shown in red; Fig. 3B). The equivalent Drosophila mutation
completely abolished fusion (Fig. 4C). Finally, several frame-
shift mutations in the C-terminal tail also caused HSP (34–36),
consistent with a role of the tail in fusion. Collectively, these
results suggest that ATL mutations cause HSP by compromising
homotypic ER fusion.
Our results suggest how ATL causes homotypic fusion of ER
membranes (Fig. 5). First, two ATL molecules sitting in different
membranes bind GTP and form a dimer, thereby tethering the
arepointing in different directions.Following GTPhydrolysis and
probably triggered by changes in the switch 2 regions that are
propagated to the interaction surface between the GTPase do-
main and the three-helix bundle. The three-helix bundles would
then be released, and the linkers may allow the dimer of GTPase
domains to freely rotate. We speculate that the GTPase domain
of one ATL molecule captures the three-helix bundle of the
opposing ATL molecule; the resulting conformational change
wt + Pi
10 11 12
13 14 15 16
Switch 2Switch 2
Switch 2Switch 2
bundles (3HB) of the two ATL1 molecules are shown in gray and purple, respectively. Helices of the GTPase domain interacting with the 3HBs are shown in
green. The bound nucleotide (in orange sticks) and the switch 2 region are indicated. Residue K345 is located in the α7-helix in the prefusion state and
becomes accessible to trypsin in the postfusion state (arrowhead). (B) Stereoview of the prefusion interface between the GTPase domain and the 3HB.
Interacting segments of the helices of the GTPase domain and of the α7-helix of the 3HB are shown in green, and relevant amino acids are indicated. The
backbone of K345 is in red. The linker (L-1) is indicated. (C) As in B, but for the postfusion structure. (D) Confirmation of the conformational change by
a trypsin-protection assay. Purified cytATL1 of wild-type (wt) or mutant human ATL1 was incubated with the indicated nucleotides (2 mM) and treated with
protease. The samples in lanes 5–8 also contained 100 mM Pi. All samples were analyzed by SDS/PAGE and stained with Coomassie blue. The arrows indicate
the full-length form and a protected fragment, respectively. (E) As in D, but with full-length Drosophila ATL, reconstituted into proteoliposomes. The vesicles
were floated in a sucrose gradient before analysis by trypsin digestion. Piwas added at 500 mM. (F) Purified cytATL1 with a cysteine at position 357 of the 3HB
was treated with the bifunctional cross-linker BMOE in the presence of the indicated nucleotides. Non-cross-linked protein (*) and cross-linked dimer (**) are
indicated. (G) Purified cytATL1 with cysteines at positions 192 and 348 was incubated with the oxidant diamide in the presence of the indicated nucleotides.
Where indicated, the disulfide bridge was reduced with β-mercaptoethanol (BME) before nonreducing SDS/PAGE.
Nucleotide-dependent conformational changes of ATL1. (A) The two crystal structures were superimposed with their GTPase domains. The three-helix
Bian et al.PNAS
| March 8, 2011
| vol. 108
| no. 10
would pull the apposing membranes together, so that after fusion
(Fig.5). This transition appears to befacilitated by the C-terminal
cytosolic tailsoftheATLmolecules.Onepossibility isthatthetail
of one ATL molecule interacts with the three-helix bundle of the
other ATL molecule. Alternatively, the tails of ATL molecules
could interact with one another, either in a parallel orientation or
in an anti-parallel one, as reported for the tails of mitofusins (37).
Efficient membrane fusion may require several ATL dimers to
undergo the conformational change depicted in Fig. 5, perhaps
by forming higher oligomers like other dynamin-like proteins,
although such oligomers were not observed with cytATL. Sub-
sequent rounds of fusion likely require the dissociation of the
ATL dimers and exchange of GDP for GTP and could involve
additional protein factors. ATL-mediated homotypic fusion dif-
fers from heterotypic viral fusion and SNARE-mediated fusion,
in that nucleotide triphosphate hydrolysis directly drives fusion.
In SNARE-mediated fusion, nucleotide hydrolysis is used to
reset the fusion machinery by NSF-mediated disassembly of
SNARE complexes (12), and viral fusion does not require nu-
cleotide hydrolysis at all. A common feature of all fusion reac-
tions is a conformational change of proteins that ultimately leads
to the merging of the lipid bilayers. The mechanism proposed for
ATL could also apply to Sey1p and its homologs in the fusion of
yeast and plant ER membranes and to mitofusin/Fzo1p in the
fusion of mitochondrial outer membranes.
While this paper was under consideration by a different
journal, the same ATL structures were published (38), but the
authors did not have evidence that the conformational change
suggested by the structures occurred during fusion. They also did
not observe ATL dimers with GDP, likely because the nucleotide
was omitted from the gel filtration buffer. Our results provide
strong support for a GDP-bound ATL dimer as an intermediate
during the fusion reaction.
Materials and Methods
Protein Purification. Residues 18–447 of human atlastin-1 with an N-terminal,
thrombin-cleavable His6-tag were expressed in E. coli. The protein was iso-
lated by Ni-NTA chromatography, cleaved with thrombin, and further pu-
rified by ion-exchange chromatography and gel filtration.
Crystallization. For crystallization, 0.4 mM cytATL1 was incubated with 2 mM
GDPfor2hat roomtemperature.Crystallization trialswere performedbythe
% Total Fluorescence 19
% Total Fluorescence
10 20 304050
% Total Fluorescence
20 30 40 50 60
% Total Fluorescence15
1020 30 405060
were reconstituted at equal concentrations into donor and acceptor vesicles. Reconstitution of the proteins was efficient as shown by flotation (Fig. S10) and
resulted in all ATL molecules having their cytosolic domains exposed (Fig. 3E). GTP-dependent fusion of donor and acceptor vesicles was followed by the
dequenching of a fluorescent lipid present in the donor vesicles. Control experiments were performed in the absence of Mg2+or presence of GDP instead of
GTP. (B) The fusion of vesicles containing full-length wt ATL was determined in the presence of increasing concentrations of the cytosolic domain (cytATL).
The molar ratios are indicated. GST was used as a control. (C) As in A, but with mutants that affect GTPase activity or cause HSP (F126S, H222P, I290S). For
comparison, the data in A for wt ATL are replotted. (D) As in C, but with mutants that affect the interaction between the GTPase domain and the three-helix
bundle. ATL lacking the cytosolic tail (1–476) was also tested. (E) Mutations causing HSP were mapped into the postfusion structure (blue). The effect of
mutation I315S (red) can only be explained by the prefusion structure. (F) The disease mutations were mapped into the prefusion structure.
Membrane fusion with wild-type (wt) and mutant ATL. (A) Full-length wt Drosophila ATL or mutants that affect dimer formation of human cytATL1
GTP-induced dimerization &
GTP hydrolysis and Pi release
Conformational changes &
cussion for details. GTP and GDP molecules are indicated as magenta and
yellow spheres, respectively. G, GTPase domain, 3HB, three-helix bundle,
TMs, TM domains, L, linker.
A model for ATL-mediated homotypic membrane fusion. See Dis-
| www.pnas.org/cgi/doi/10.1073/pnas.1101643108Bian et al.
hanging-drop vapor-diffusion method at 20 °C. The hexagonal form was
obtained with GDP and native or Se-Met-substituted in 0.1 M Tris·HCl (pH
8.5), 25% (wt/vol) PEG 1500, and 20% (vol/vol) glycerol. The orthorhombic
form was obtained in 0.1 M Tris·HCl (pH 8.3), 25% (wt/vol) PEG 3350, and
0.1 M (NH4)2HPO4.
Data Collection and Structure Determination. Diffraction data for both crystal
forms were collected at 100 K on beamline BL17U at Shanghai Synchrotron
Radiation Facility, Shanghai, China. Hexagonal crystals were used directly and
orthorhombic crystals were soaked in mother liquor plus 20% (vol/vol)
glycerol for a few seconds. All data sets were processed with HKL2000 and
converted to CCP4 format. The structure of the Se-Met substituted protein
was determined by MAD, and the structure of the protein in primitive or-
thorhombic crystals was determined by molecular replacement. Models were
built with the Phenix program and refined and manually rebuilt against the
native data set. The CNS suite and the Phenix refinement program (phenix.
refine) were combined during refinement.
Analytical Ultracentrifugation and Gel Filtration. Sedimentation velocity data
were collected at a speed of 142,249 × g in an An-60 Ti rotor at 12 °C. The
proteins were prepared in 50 mM Tris (pH 8.0), 5 mM MgCl2, and 1 mM DTT
at 0.1–3 mg/mL Data were analyzed by the program SEDFIT (Version 11.8).
Sedimentation equilibrium experiments are described in SI Materials and
Methods. For gel filtration analysis, 100 μL of purified cytATL1 at 0.2 mM
was incubated with or without 5 mM nucleotide and loaded onto a Super-
dex 200 column in buffer containing 50 mM Tris (pH 8.0), 100 mM NaCl,
5 mM MgCl2, and 1 mM nucleotide, if present.
Trypsin-Protection Assay. Ten micrograms of cytATL1 with the indicated
nucleotides was incubated with 0.6 μg of trypsin at 37 °C for 0.5 h. Trypsin di-
gestion of full-length Drosophila ATL was done similarly with proteoliposomes
(protein to lipid ratio, 1:1,000) floated in a discontinuous sucrose gradient.
Cross-Linking Assays. For cross-linking with BMOE (Thermo Scientific), 1 μM
cytATL1 was incubated with or without GDP or GMPPNP with 2 μM BMOE at
room temperature for 1 h in 50 mM Hepes (pH 7.0), 100 mM NaCl, and 5 mM
MgCl2. The reactions were quenched with 50 mM DTT. For cross-linking with
diamide (Sigma), 5 μM cytATL1 was incubated with 50 μM diamide at room
temperature for 30 min, and the reaction was quenched with 250 μM N-
Fusion Assays. Full-length, codon-optimized Drosophila ATL was expressed in
E. coli as a GST fusion and purified on glutathione agarose, as described (14).
The protein was passed through a PD-10 column, and the GST moiety was
cleaved off by thrombin and removed with glutathione agarose. The cyto-
solic domain of Drosophila ATL (residues 1–422) was purified similarly in the
absence of detergent. The generation of reconstituted liposomes at a
1:2,000 molar ratio of protein:lipid and the fusion assay were done as
Further details on methods are provided in SI Materials and Methods.
ACKNOWLEDGMENTS. We thank P. Xue, F. Yang, and W. Shui for help with
mass spectrometry; X. Yang and Z. Wang for help with ITC and AUC,
respectively; A. Stein, S. Hubbard, and Q. Wang for critical reading of the
manuscript; and the staff at beamline BL17U of Shanghai synchrotron ra-
diation facility. X.L. is supported by the National Basic Research Program of
China (973 Program, Grant 2010CB911800) and the National S&T Major Pro-
ject on Major Infectious Diseases (Grant 2008ZX10001-010) from the Ministry
of Science and Technology of the People’s Republic of China. J.H. is sup-
ported by the National Basic Research Program of China (973 Program, Grant
2010CB833702) and the National Natural Science Foundation of China
(Grants 30971440 and 90919009). T.A.R. and J.H. received support from
the 111 Project of China (B08011). R.W.K. is supported by a European Mo-
lecular Biology Organization long-term fellowship. T.Y.L. is supported by
a National Science Foundation graduate research fellowship. T.A.R. is a
Howard Hughes Medical Institute investigator.
1. Baumann O, Walz B (2001) Endoplasmic reticulum of animal cells and its organization
into structural and functional domains. Int Rev Cytol 205:149–214.
2. Shibata Y, Voeltz GK, Rapoport TA (2006) Rough sheets and smooth tubules. Cell 126:
3. Lee C, Chen LB (1988) Dynamic behavior of endoplasmic reticulum in living cells. Cell
4. Du Y, Ferro-Novick S, Novick P (2004) Dynamics and inheritance of the endoplasmic
reticulum. J Cell Sci 117:2871–2878.
5. Prinz WA, et al. (2000) Mutants affecting the structure of the cortical endoplasmic
reticulum in Saccharomyces cerevisiae. J Cell Biol 150:461–474.
6. Harrison SC (2008) Viral membrane fusion. Nat Struct Mol Biol 15:690–698.
7. Jahn R, Scheller RH (2006) SNAREs—engines for membrane fusion. Nat Rev Mol Cell
8. Wickner W, Schekman R (2008) Membrane fusion. Nat Struct Mol Biol 15:658–664.
9. Martens S, McMahon HT (2008) Mechanisms of membrane fusion: Disparate players
and common principles. Nat Rev Mol Cell Biol 9:543–556.
10. Südhof TC, Rothman JE (2009) Membrane fusion: Grappling with SNARE and SM
proteins. Science 323:474–477.
11. Söllner T, et al. (1993) SNAP receptors implicated in vesicle targeting and fusion.
12. Mayer A, Wickner W, Haas A (1996) Sec18p (NSF)-driven release of Sec17p (alpha-
SNAP) can precede docking and fusion of yeast vacuoles. Cell 85:83–94.
13. Hu J, et al. (2009) A class of dynamin-like GTPases involved in the generation of the
tubular ER network. Cell 138:549–561.
14. Orso G, et al. (2009) Homotypic fusion of ER membranes requires the dynamin-like
GTPase atlastin. Nature 460:978–983.
15. Zhao X, et al. (2001) Mutations in a newly identified GTPase gene cause autosomal
dominant hereditary spastic paraplegia. Nat Genet 29:326–331.
16. Rismanchi N, Soderblom C, Stadler J, Zhu PP, Blackstone C (2008) Atlastin GTPases are
required for Golgi apparatus and ER morphogenesis. Hum Mol Genet 17:1591–1604.
17. Salinas S, Proukakis C, Crosby A, Warner TT (2008) Hereditary spastic paraplegia:
Clinical features and pathogenetic mechanisms. Lancet Neurol 7:1127–1138.
18. Hermann GJ, et al. (1998) Mitochondrial fusion in yeast requires the transmembrane
GTPase Fzo1p. J Cell Biol 143:359–373.
19. Chen H, et al. (2003) Mitofusins Mfn1 and Mfn2 coordinately regulate mitochondrial
fusion and are essential for embryonic development. J Cell Biol 160:189–200.
20. Takei K, McPherson PS, Schmid SL, De Camilli P (1995) Tubular membrane
invaginations coated by dynamin rings are induced by GTP-gamma S in nerve
terminals. Nature 374:186–190.
21. Sweitzer SM, Hinshaw JE (1998) Dynamin undergoes a GTP-dependent conforma-
tional change causing vesiculation. Cell 93:1021–1029.
22. Marks B, et al. (2001) GTPase activity of dynamin and resulting conformation change
are essential for endocytosis. Nature 410:231–235.
23. Pucadyil TJ, Schmid SL (2008) Real-time visualization of dynamin-catalyzed membrane
fission and vesicle release. Cell 135:1263–1275.
24. Bashkirov PV, et al. (2008) GTPase cycle of dynamin is coupled to membrane squeeze
and release, leading to spontaneous fission. Cell 135:1276–1286.
25. Prakash B, Renault L, Praefcke GJ, Herrmann C, Wittinghofer A (2000) Triphosphate
structure of guanylate-binding protein 1 and implications for nucleotide binding and
GTPase mechanism. EMBO J 19:4555–4564.
26. Prakash B, Praefcke GJ, Renault L, Wittinghofer A, Herrmann C (2000) Structure of
human guanylate-binding protein 1 representing a unique class of GTP-binding
proteins. Nature 403:567–571.
27. Ghosh A, Praefcke GJ, Renault L, Wittinghofer A, Herrmann C (2006) How guanylate-
binding proteins achieve assembly-stimulated processive cleavage of GTP to GMP.
28. Gao S, et al. (2010) Structural basis of oligomerization in the stalk region of dynamin-
like MxA. Nature 465:502–506.
29. Low HH, Löwe J (2006) A bacterial dynamin-like protein. Nature 444:766–769.
30. Low HH, Sachse C, Amos LA, Löwe J (2009) Structure of a bacterial dynamin-like
protein lipid tube provides a mechanism for assembly and membrane curving. Cell
31. Praefcke GJ, McMahon HT (2004) The dynamin superfamily: Universal membrane
tubulation and fission molecules? Nat Rev Mol Cell Biol 5:133–147.
32. Smith BN, et al. (2009) Four novel SPG3A/atlastin mutations identified in autosomal
dominant hereditary spastic paraplegia kindreds with intra-familial variability in age
of onset and complex phenotype. Clin Genet 75:485–489.
33. Guelly C, et al. (2011) Targeted high-throughput sequencing identifies mutations in
atlastin-1 as a cause of hereditary sensory neuropathy type I. Am J Hum Genet 88:
34. Tessa A, et al. (2002) SPG3A: An additional family carrying a new atlastin mutation.
35. Ivanova N, et al. (2007) Hereditary spastic paraplegia 3A associated with axonal
neuropathy. Arch Neurol 64:706–713.
36. Loureiro JL, et al. (2009) Novel SPG3A and SPG4 mutations in dominant spastic
paraplegia families. Acta Neurol Scand 119:113–118.
37. Koshiba T, et al. (2004) Structural basis of mitochondrial tethering by mitofusin
complexes. Science 305:858–862.
38. Byrnes LJ, Sondermann H (2011) Structural basis for the nucleotide-dependent
dimerization of the large G protein atlastin-1/SPG3A. Proc Natl Acad Sci U S A 108:
Bian et al.PNAS
| March 8, 2011
| vol. 108
| no. 10