Membrane fusion by the GTPase atlastin requires
a conserved C-terminal cytoplasmic tail and
dimerization through the middle domain
Tyler J. Mossa, Camilla Andreazzab, Avani Vermaa, Andrea Dagab, and James A. McNewa,1
aDepartment of Biochemistry and Cell Biology, Rice University, Houston, TX 77005; andbE. Medea Scientific Institute, 31015 Conegliano, Italy
Edited* by William T. Wickner, Dartmouth Medical School, Hanover, NH, and approved May 25, 2011 (received for review March 30, 2011)
The biogenesis and maintenance of the endoplasmic reticulum (ER)
requires membrane fusion. ER homotypic fusion is driven by the
large GTPase atlastin. Domain analysis of atlastin shows that a con-
served region of the C-terminal cytoplasmic tail is absolutely re-
quired for fusion activity. Atlastin in adjacent membranes must
associate to bring the ER membranes into molecular contact. Dro-
sophila atlastin dimerizes in the presence of GTPγS but is mono-
meric with GDP or without nucleotide. Oligomerization requires
the juxtamembrane middle domain three-helix bundle, as does
efficient GTPase activity. A soluble version of the N-terminal cyto-
plasmic domain that contains the GTPase domain and the middle
domain three-helix bundle serves as a potent, concentration-
dependent inhibitor of membrane fusion both in vitro and in vivo.
However, atlastin domains lacking the middle domain are without
effect. GTP-dependent dimerization of atlastin generates an enzy-
matically active protein that drives membrane fusion after nucle-
otide hydrolysis and conformational reorganization.
endoplasmic reticulum biogenesis|organelle fusion|hereditary spastic
paraplegia|spastic paraplegia gene 3A
traffic within the secretory pathway as well as the biogenesis and
maintenance of the entire endomembrane system. SNARE pro-
teins are responsible for membrane fusion within the secretory
pathway as well as homotypic fusion of endosomes and lysosomes
(1, 2). The fusion of other organelles such as mitochondria and
the endoplasmic reticulum (ER) are less well characterized.
Many observations suggest that mitochondrial fusion is driven by
proteins called mitofusins (3–5). Although ER-resident SNARE
proteins are required forvesicular transport back tothe ER (6, 7),
the protein(s) responsible for the generation and maintenance
of the ER has only recently been discovered. We have recently
shown that Atlastin is a GTP-dependent membrane fusion pro-
tein that is responsible for ER homotypic fusion (8). Membrane
fusion provided by atlastin helps shape and maintain the dynamic
nature of the ER membrane tubule network (9–13).
Atlastin is the product of the human SPG3A locus (14). SPG3A
(Atl1) is a member of a larger family of genes that are responsible
for a group of inherited neurological disorders called hereditary
spastic paraplegia (HSP) (15, 16). This disease is characterized by
progressive lower-extremity weakness and spasticity. The neuro-
pathological basis for compromised motor function in HSP is
likely length-dependent axonopathy of the corticospinal tract
(16). Twenty HSP gene products have been identified, and their
molecular analysis has suggested that three general categories of
proteins may be responsible for HSP. These gene functions fall
into three broad groups, including intracellular trafficking, mi-
tochondrial function, and axonal pathfinding and myelination
(17). More than half of all HSP cases are caused by mutation in
ER-resident or ER-associated proteins (18–20). Atlastin and the
ER tubule-forming protein receptor expression-enhancing pro-
tein 1 (REEP1), as well as other reticulons, are all ER-resident
proteins (20, 21). Spastin, a microtubule-severing protein, has
embrane fusion reactions are vitally important for many
aspects of eukaryotic cell biology, including vesicular
been shown to associate with REEP1 and atlastin in some con-
texts (20, 22, 23), and recent work has also demonstrated that
atlastin interacts with all of the ER tubule-forming proteins in the
Reticulon and REEP/Yop/DP1 family (20). These observations
place the ER at the nexus of HSP pathology.
Recently, atlastin function has been examined in model
organisms (24), specifically in the fruit fly Drosophila mela-
nogaster. The Drosophila genome produces a single atlastin pro-
tein. Drosophila atlastin (atl) is 541 aa long and has a predicted
molecular mass of 61 kDa (Fig. 1A). The atl sequence is highly
homologous with all three human isoforms, ranging between
44% and 49% identical (61% and 68% similar) over the entire
length of the protein. Overall homology is significantly improved
(55–59% identity) when the highly variable N-terminal (17–45
residues) and C-terminal (29–50 residues) regions are excluded.
We now examine the mechanistic basis for nucleotide-dependent
membrane fusion by atl using a structure–function analysis of trun-
cated proteins and a detailed kinetic analysis of nucleotide hydro-
lysis. Finally, we determine the oligomeric state of the N-terminal
cytoplasmic domain and isolated GTPase domain in the presence
and absence of nucleotide.
atl is a 61-kDa multidomain protein (Fig. 1A). It has a short
N-terminal variable domain, followed by a well-conserved
GTPase domain, a middle domain of undefined function, two
tandem transmembrane domains, and a C-terminal cytoplasmic
domain. We previously documented that atlastin requires GTPase
activity for membrane fusion in vitro; however, the function of the
other domains remains to be explored. In this article, we extend
our previous work by conducting a systematic structure–function
analysis of atlastin with the D. melanogaster homolog.
Atlastin C-Terminal Cytoplasmic Tail Is Required for Membrane Fusion.
We began our analysis by determining the region of atlastin re-
quired for membrane fusion by producing C-terminal truncations.
We have previously shown that GTPase activity is absolutely re-
quired for membrane fusion (8), so we reasoned that significant
N-terminal deletions would abolish fusion activity by impairing
GTPase activity given that atl contains only a 14-residue N-ter-
minal extension beyond the boundary of the GTPase domain. We
produced a truncation that removed the entire C-terminal cyto-
plasmic domain (residues 472–542) and a truncation that elimi-
nated the second transmembrane domain as well as the C-
terminal cytoplasmic domain (residues 451–542); both atl(1–471)
Author contributions: T.J.M., C.A., A.D., and J.A.M. designed research; T.J.M., C.A., and
A.V. performed research; T.J.M., A.D., and J.A.M. analyzed data; and T.J.M., A.D., and
J.A.M. wrote the paper.
The authors declare no conflict of interest.
*This Direct Submission article had a prearranged editor.
1To whom correspondence should be addressed. E-mail: firstname.lastname@example.org.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
| July 5, 2011
| vol. 108
| no. 27
and atl(1–450) retain an intact GTPase domain and at least one
transmembrane anchor. We expressed these proteins in Escherichia
coli, reconstituted them into synthetic phosphatidylcholine:phos-
phatidylserine (PCPS) liposome, and measured membrane fusion
by lipid mixing (8). When either of these two proteins was recon-
stituted into proteoliposomes, no fusion was observed (Fig. 1B).
The functionality of these enzymes was confirmed by examining
GTPase activity (Fig. 1C). Both atl(1–450) and atl(1–471) were able
to cleave GTP at rates that were within ∼30% of WT rates, yet they
were incapable of driving membrane fusion. The small reduction in
GTPase activity for the C-terminal deletion mutants may reflect
a subtle effect of the C-terminal tail on atlastin oligomerization.
The inability of these proteins to produce a functional atlastin
was further explored in a tissue culture model in vivo. When
functional atl is overexpressed in mammalian cells, it correctly
sorts to the ER, causes the formation of an enlarged ER com-
partment, and redistributes Golgi-resident proteins to the ER in
much the same way as overexpression in intact animals (8). Fig. 2
A–C shows the result of expressing WT atlastin in Cos7 cells. atl
colocalizes with the ER marker Sec61β and produces significant
changes in ER (Fig. 2B) morphology. When functional atl is
expressed, the normally reticular ER (Fig. S1) is lost, and var-
iably sized puncta are now stained with ER markers. However,
overexpressing ER-localized atl(1–471) minimally disrupts ER
morphology, demonstrating that it is a less or nonfunctional
protein in vivo (Fig. 2 D–F).
These results suggest that the C-terminal cytoplasmic domain
of atlastin is critical for membrane fusion. However, this region of
atlastin is among the most divergent between atlastin homologs
and paralogs. Closer inspection of the sequences that comprise
the C-terminal cytoplasmic domain revealed an ∼25-aa stretch
of more highly conserved residues membrane-proximal to the
second transmembrane domain followed by a very divergent
extreme C terminus (Fig. S2). We further subdivided the C-
terminal cytoplasmic domain to add back the more conserved
juxtamembrane region (residues 471–497). The resulting con-
struct, atl(1–497), was also expressed, reconstituted, and assayed
for membrane fusion. Fig. 1 B and C demonstrate that the re-
introduction of these 27 residues restores ∼70% of WT fusion
activity. Additionally, overexpression of atl(1–497) in Cos7 cells
results in a phenotype similar to the overexpression of WT atlas-
tin, confirming that atl(1–497) is functional in vivo (Fig. 2 G–I).
Soluble N-Terminal Cytoplasmic Domain of Atlastin Is a Concentration-
Dependent Inhibitor of ER Membrane Fusion. All membrane fusion
proteins characterized to date require membrane integration.
However, liberated soluble domains that retain the ability to
productively interact with membrane integral components are
often inhibitors of fusion. This is definitely the case for SNARE
proteins, and inclusion of a soluble fragment of either v-SNARE
or t-SNARE component will inhibit fusion (25, 26). We tested the
functionality of C-terminal truncations that lack a membrane-
spanning domain by determining their capacity to inhibit fusion
between proteoliposomes that contain WT atlastin. We tested atl
(1–422), which encompassed the entire N-terminal cytoplasmic
domain; atl(1–303), which contains the isolated GTPase domain;
and atl(471–542), the soluble C-terminal fragment (Fig. 3A).
fusion in vitro. (A) Domain structure of atl and C-terminal truncations. The
GTPase domain is represented as a cylinder, the middle domain 3HB is shown
as a helix, and the transmembrane domains are cylinders labeled “TM.”
Relevant domain boundaries are indicated by residue numbers above the
diagram. (B) Kinetic fusion graph of unlabeled atl acceptor proteoliposomes
fused with equimolar amounts of fluorescently labeled atl donor proteoli-
posomes. NBD fluorescence was measured at 1-min intervals, and detergent
was added at 60 min to determine maximum fluorescence. WT GST-atl-His8,
●; GST-atl(1–450)-His8, ■; GST-atl(1–471)-His8, ◆; GST-atl(1–497)-His8, ○;
control reaction in the absence of GTP, gray line. In all cases, the same atl
mutant is reconstituted into both liposome populations. (Inset) The average
extent of fusion at 60 min is represented in tabular form as percentage of
WT fusion for all mutants. WT GST-atl-His8(n = 28) was set to 100 ± 0.42%.
GST-atl(1–450)-His8(n = 17) was 4.85 ± 0.53%, GST-atl(1–471)-His8(n = 17)
was 4.86 ± 1.52%, and GST-atl(1–497)-His8(n = 8) was 68.8 ± 2.22%. (C)
GTPase activity is represented as a histogram for all mutants. All activity is
reported as μM Pi/min per μM enzyme. WT GST-atl-His8(n = 47) was 0.71 ±
0.03, GST-atl(1–450)-His8(n = 12) was 0.61 ± 0.03, GST-atl(1–471)-His8(n = 19)
was 0.48 ± 0.01, and GST-atl(1–497)-His8(n = 6) was 0.88 ± 0.11. (Error bars:
SEM. Numbers in histograms = number of replicates.)
The atlastin C-terminal cytoplasmic tail is required for membrane
ER (Sec61 )
ER (Sec61 )
functional protein in vivo. Cos7 cells were cotransfected with WT atlastin-
myc (A–C), atl(1–471)-myc (D–F), or atl(1–497)-myc (G–I) and GFP-Sec61β. WT
atlastin and atl(1–497) show an abnormal, punctate ER, indicating a func-
tional atl protein, whereas atl(1–471) localizes to a normally reticular ER,
suggesting that it is a nonfunctional protein. (Scale bar: 10 μm.)
Deletion of the atlastin C-terminal cytoplasmic tail results in a non-
| www.pnas.org/cgi/doi/10.1073/pnas.1105056108 Moss et al.
When atl(1–422) was titrated into a fusion reaction containing WT
atlastin in the membrane, progressive inhibition of both the rate
and extent of membrane fusion was seen (Fig. 3B). Inhibition was
concentration-dependent, and fusion was completely inhibited
with an eightfold molar excess of the atl(1–422) soluble domain
(Fig. 3C). Although the full-length N-terminal cytoplasmic domain
was a potent inhibitor, neither the isolated GTPase domain [atl(1–
303); Fig. 3C, gray column] nor the soluble C-terminal tail [atl
(471–542); Fig. 3C, white column] inhibited fusion.
The ability of the isolated N-terminal cytoplasmic domain to
inhibit fusion was further explored by co-overexpressing atl(1–422)
in Cos7 cells with WT atlastin. As shown in Fig. 2A, overexpressing
WT atlastin in Cos7 cells disrupts ER morphology, presumably
because of inappropriate and excessive ER fusion. However, si-
multaneous overexpression of the soluble N-terminal cytoplasmic
domain of atlastin, atl(1–422), reduces the severity or largely
prevents the aberrant ER structures generated by WT atlastin
(Fig. 4). A spectrum of inhibition is seen with atl(1–422), which
likely reflects the relative degree of overexpression. When cells
were subjectively binned into three groups, defined as normal ER
morphology (Fig. 4, empty arrowheads), fused ER (Fig. 4, white
arrowheads), and partially fused ER (Fig. 4, gray arrowhead),
more than half (56%) of the cells expressing atl(1–422) exhibited
a normal ER morphology, whereas 43% displayed a partially fused
ER (Fig. S3). We interpret these results to suggest that atl(1–422)
productively and preferentially interacts with full-length WT
atlastin to inhibit its fusion activity in vivo.
Atlastin Dimerization Is Required for Enzymatic Activity. Although
we know that atlastin GTPase activity is required for membrane
fusion, we know very little regarding the enzymatic properties of
atlastin. To address this issue, we measured Michaelis–Menten
kinetics for full-length GST-atl in detergent and reconstituted
into proteoliposomes by using a coupled enzymatic assay that
quantifies inorganic phosphate production. We ensured that
quantifying inorganic phosphate production was a reliable
measure of GTPase activity by examining all of the guanine
nucleotide reaction products by HPLC because guanylate bind-
ing protein 1 (GBP1), another large GTPase family member
similar in sequence to atlastin, is capable of sequentially cleaving
GTP to GMP (Fig. S4). Fig. 5A shows a plot of initial velocity
versus GTP concentration. Surprisingly, we observed that re-
constitution significantly improved the maximum velocity of this
enzyme even though the amount of enzyme assayed in proteo-
liposomes is half (0.5 μM) of that in detergent (1 μM) (Table 1).
These data demonstrate that atlastin is a more active enzyme in
a phospholipid bilayer compared with a detergent micelle. The
most straightforward interpretation of this activity increase is
that atlastin has an improved ability to dimerize in the plane of
a membrane after reconstitution (a cis dimer) or that trans di-
merization between liposomes stimulates GTPase activity.
Next, we examined the activity of the N-terminal cytoplasmic
domain, atl(1–422). This protein was produced as a GST-SUMO
fusion protein, and the N-terminal tags were removed before
analysis. In this case, the liberated N-terminal cytoplasmic domain
displayed comparable activity to the reconstituted full-length
responsible for oligomerization and enzymatic activation reside in
this domain. Given that other large GTPases such as dynamin and
GBP1 dimerize through contacts within the G domain, we pro-
duced an additional construct that expressed the isolated GTPase
domain, atl(1–328). However, we found that the GTPase domain
alone was not sufficient to produce an active enzyme at the same
concentration (0.5 μM) as the other protein examined (Fig. 5B).
The atl(1–328) protein is not simply a misfolded protein because it
shows a concentration-dependent increase in activity at higher
GTP concentrations (up to 1 mM; Fig. S5). These results suggest
that the GTPase domain alone is not sufficient to provide a stable
interaction platform for GTP hydrolysis.
Nucleotide-Dependent Atlastin Oligomerization Is Driven by the
Middle Domain. GTPase activity measurements and soluble do-
main inhibitor results suggest that the middle domain of atlastin
may be important for oligomerization. To address the question of
422) and atl(1–328) by analytical ultracentrifugation. Fig. 5 C and
proteins: a van Holde–Weischet analysis (27) produced diffusion-
corrected sedimentation coefficients for atl(1–422) in the absence
of nucleotide (solid black trace) or in the presence of GDP (large
dashed trace), or the nonhydrolyzable GTP analog GTPγS (small
dashed trace) revealed a dimer only in the presence of GTPγS.
atl(1–422) migrated similarly to the apo form (3.21 S), and GTPγS
inhibition of fusion by the N-terminal cytoplasmic do-
main of atl. (A) Domain structure of atl and the soluble
atl fragments added acutely to fusion reactions be-
tween WT atlastin proteoliposomes. (B) Kinetic fusion
graph of unlabeled atl acceptor proteoliposomes fused
with equimolar amounts of fluorescently labeled atl
donor proteoliposomes (black) with increasing amounts
of GST-atl(1–422)-His8 (shades of gray) added to the
reaction. (C) The extent of fusion at 60 min is repre-
sented in a histogram as the percentage of fusion
without any soluble domains added to the reaction.
The middle domain is required for the in vitro
Moss et al. PNAS
| July 5, 2011
| vol. 108
| no. 27
sedimented faster at 4.95 S. These sedimentation values corre-
spond to molecular masses of 48.7 (Apo), 45.3 (GDP), and 86.3
(GTPγS) kDa, respectively (Fig. S6). The predicted monomer
molecular mass ofatl(1–422)is 49,317 Da, stronglysuggesting that
both the apo and GDP-bound protein is monomeric, whereas the
GTPγS-bound protein forms a dimer. A similar analysis with atl
(1–328) produced a different result. In this case, all three species
(apo, GDP, and GTPγS) migrated at a size most consistent with a
monomer molecular mass (36.6, 43.0, and 41.0 kDa, respectively).
The predicted monomer molecular mass of atl(1–328) is 37,243
Da. Together, GTPase activity measurements and molecular mass
analysis by analytical ultracentrifugation suggests that the N-
terminal cytoplasmic domain of atl can dimerize in the presence
of GTP and that nucleotide binding promotes a conformation
that requires the middle domain for stable oligomerization.
Traditional membrane fusion proteins such as viral fusion pro-
teins and SNAREs use energy derived from metastable protein
folding intermediates to drive fusion. The use of chemical energy
in the form of nucleotide hydrolysis at the point of membrane
fusion is unique to atlastin, and perhaps mitofusins, and defines
this class of membrane fusion proteins. Detailed analysis of this
type of mechanism has only recently begun to be explored. We
have determined that a short region of ∼27 aa located within the
cytoplasmic C-terminal tail is critically required for atlastin-
dependent membrane fusion in vitro (Fig. 1) and in vivo (Fig. 2).
Interestingly, three frameshift mutations in human Atlastin-1
(Atl1) [A492fsX522 (28), E502fsX522 (29), and I507fsX522 (30)]
known to cause HSP prematurely truncate Atl1 and scramble the
C-terminal juxtamembrane region that we have shown is required
for membrane fusion.
The N-terminal cytoplasmic domain (residues 1–422), when
freed from membrane attachment, functions as a fusion inhibitor
both in vitro (Fig. 3) and in vivo (Fig. 4). Quantitative analysis of
the enzymatic properties of atlastin revealed that the N-terminal
cytoplasmic domain is a more active enzyme than the full-length
protein in detergent solution. However, reconstitution of the full-
length protein into a phospholipid bilayer improved activity but
not to the level of the soluble protein (Fig. 5 A and B). Addi-
tionally, GTPase activity required protein sequences outside the
GTPase domain, suggesting that this domain alone is not sufficient
main dominantly inhibits WT atlastin function in vivo. Cos7 cells over-
full-length atl and atl(1–422) show a relatively normal ER phenotype (empty
arrowheads). Asterisk indicates a cell expressing full-length atl only and
shows an abnormal, punctate ER phenotype. (D–F) Cos7 cells expressing full-
length atl and varying levels of atl(1–422) show a mix of a normal ER phe-
notype (empty arrowheads) and a fused ER phenotype (white arrowheads).
(G–I) Cos7 cells coexpressing atl and atl(1–422) representing a fused ER phe-
notype (white arrowhead) and apartiallyfused ER (gray arrowhead). Insets in
A and D give a magnified view and increased color contrast of the outlined
region of a cell showing the normal ER phenotype. Empty arrowheads in-
dicate WT ER, where WT atlastin is inhibited by coexpression of atl(1–422).
Gray arrowheads show the partially fused ER phenotype, where WT atlastin
activity is partially inhibited by atl(1–422). White arrowheads illustrate fully
fused ER, where there is minimal or no effect of atl(1–422). (Scale bar: 10 μm.)
Variable overexpression of the atlastin N-terminal cytoplasmic do-
0.5µM atl (1-422)
0.5µM atl (1-328)
0 10 2030 40 5060
GTPase activity (µM Pi / min)
0.5 µM GST-atl in liposomes
1.0 µM GST-atl in detergent
Sedimentation Coefficient (10-13 seconds)
Sedimentation Coefficient (10-13 seconds)
atlastin dimerization and normal GTP activity. (A) Graph of the initial ve-
locities of GTP hydrolysis versus GTP concentration of WT atlastin solubilized
in detergent or reconstituted into synthetic liposomes. The turnover number
(Kcat) is increased about threefold when atl is reconstituted into liposomes,
whereas the nucleotide affinity is largely unaffected. (B) Initial velocities of
GTP hydrolysis by the N-terminal cytoplasmic domain, atl(1–422), and the
isolated GTPase domain, atl(1–328). The maximum velocity of atl(1–422) is
greater than full-length atlastin in detergent. The GTPase activity is almost
completely eliminated in atl(1–328), which lacks the middle domain. The
data are fitted with the Michaelis–Menten equation. (Error bars: SEM; n = 3.)
(C and D) Distributions of sedimentation coefficients of atl(1–422) and atl
(1–328) determined by analytical ultracentrifugation data analyzed by van
Holde–Weischet analysis. (C) atl(1–422) in buffer or with GDP is monomeric,
whereas it forms a dimer when incubated with GTPγS. (D) atl(1–328) is
monomeric under all conditions. Molecular masses were determined by us-
ing the genetic algorithm analysis in Ultrascan II (34) (Figs. S6 and S7).
The atlastin middle domain is required for nucleotide-dependent
| www.pnas.org/cgi/doi/10.1073/pnas.1105056108 Moss et al.
to a stable dimer. Analytical ultracentrifugation revealed that the
N-terminal cytoplasmic domain of atl could dimerize in the
presence of GTPγS but not GDP (Fig. 5C and Fig. S6). Fur-
thermore, GTP-dependent dimerization requires the middle do-
main because the isolated GTPase domain (residues 1–328) was
monomer under all nucleotide conditions (Fig. 5D and Fig. S7).
Very recently, two reports detailing the X-ray structure of the
N-terminal cytoplasmic domain of human Atl1 were published
(31, 32). These data provide important insights into both the
biochemical properties and potential fusion mechanisms of
atlastin. The large GTPase domain shares significant structural
similarity to human GBP1 (33), and the middle domain folds into
an antiparallel three-helix bundle (3HB) that would connect the
GTPase domain to the tandem transmembrane segments. The
GTPase domain is connected to the 3HB by a flexible linker. Both
position of the 3HB relative to the GTPase domain. Crystal form
1 (32) would likely position the transmembrane domains in
a “postfusion” conformation (31), whereas crystal form 2 proba-
bly represents a “prefusion” structure. All of the current struc-
tures contained bound GDP, although attempts were made to
bind nonhydrolyzable GTP analogs such as GTPγS or GMPPNP.
Biochemical analysis of the Atl1 (1–446) N-terminal cyto-
plasmic domain showed that GTPase activity of the WT human
protein was comparable to our Drosophila protein [5.33 μM Pi/
min per μM enzyme for Atl1 (1–446) vs. 4.4 μM Pi/min per μm
enzyme for atl(1–422)] and that GTP promoted dimerization
measured by small-angle X-ray scattering (32) or analytical ul-
tracentrifugation (31). Additionally, both groups measured di-
merization by gel filtration in the presence of GDP that yielded
different outcomes. Bian et al. (31) identified atlastin dimers in the
presence of GDP, whereas Byrnes and Sondermann (32) found
only monomers with GDP. Small-angle X-ray scattering analysis
also supported a monomer in the presence of GDP. This apparent
discrepancy may be explained by different experimental conditions
such as protein or nucleotide concentration. We detected only
monomeric atl(1–422) with GDP (Fig. 5D and Figs. S6 and S8).
Membrane fusion by atlastin likely involves many steps. The
schematic shown in Fig. 6 describes our working model for
atlastin function. Atlastin almost entirely faces the cytoplasm
with as few as 5 residues in an ER luminal loop. We suggest that
the fusion cycle begins with nucleotide-free atlastin monomers
(Fig. 6A) in opposing ER membranes. These monomer forms are
cartooned as the prefusion (31) or crystal form 2 (32) structures.
Then we suggest that nucleotide binding (Fig. 6B) results in
a permissive state for association between the GTPase domains
(Fig. 6C). The N-terminal cytoplasmic domain of atl, atl(1–422),
forms a dimer when bound to nonhydrolyzable GTPγS but not
with GDP (Fig. 5C and Fig. S6). We assume that this association
also occurs in the context of full-length atlastin. Although this
interaction between the GTPase domains is presumably a re-
quired intermediate, its stability may be limited. We propose that
the interaction between GTPase domains matures to a more
stable dimer facilitated primarily by an interaction between the
middle domain 3HB segments (Fig. 6D). This conformational
change is achieved, perhaps driven by nucleotide hydrolysis, by
rotating the GTPase domain dimer 180°, which forces the 3HBs
into close proximity. The new association between adjacent
3HBs liberates the C-terminal tail domain to perform its re-
quired role. We have localized this required function to a region
of 27 aa immediately adjacent to the second transmembrane
span (Fig. 1A). The activity of this C-terminal domain may be
accomplished by forming a new association with the dimeric 3HB
or by direct interaction with lipid (Fig. 6D, shown in cyan). We
currently favor a direct interaction with lipid based on the
amphipathic nature of this protein sequence (Fig. S2). An
interaction between the membrane surface and the amphipathic
C-terminal tail could destabilize the bilayer and provide the
driving force for outer leaflet mixing, resulting in a hemifusion
intermediate (Fig. 6E) that resolves by inner leaflet mixing to full
fusion (Fig. 6F). The resulting cis dimer resembles the postfusion
(31) or crystal form 1 (32) structure. Finally, we hypothesize that
GDP release could then promote dissociation.
Although the precise location and timing of GTP hydrolysis
during the cycle is speculative, some biochemical experiments
can be used to place constraints on the model. We suggest that
the large conformational change postulated to occur after
docking (Fig. 6 C and D) uses the energy provided by GTP hy-
drolysis. However, our oligomerization data suggest that GTPγS,
Table 1. Kinetic parameters for atlastin
ProteinKm, μMVmax, μM Pi/min
4.5 ± 0.5
7.5 ± 0.5
9.3 ± 0.4
218 ± 145
1.7 ± 0.03
1.1 ± 0.02
2.2 ± 0.02
0.3 ± 0.11
the transmembrane domains are illustrated as gray cylinders, and the C-terminal tails are shown as thick cyan lines. (A) Bilayer containing nucleotide-free
prefusion (form2) monomers. (B) GTP-bound prefusion monomers. (C) Initial, unstable docking intermediate between GTP-bound monomers through sur-
faces on the GTPase domain. (D) Stabilized dimer formed by domain rotation and 3HB interaction resulting from GTP hydrolysis. (E) Putative hemifusion
intermediate. (F) Postfusion bilayer with the form 1 (postfusion) dimer.
Model for atlastin-mediated fusion. The GTPase domains are cartooned as surface representations, the middle domains are shown as red cylinders,
Moss et al.PNAS
| July 5, 2011
| vol. 108
| no. 27
but not GDP, results in the stable dimeric conformation car-
tooned in Fig. 6D when soluble fragments are examined,
indicating that this intermediate may contain bound GTP.
Functional analysis revealed that GTP hydrolysis is required to
fuse proteoliposomes in vitro because GTPγS or GMPPNP do
not support fusion (8). These lipid-mixing results demonstrate
that nucleotide hydrolysis is required before hemifusion (Fig. 6E);
however, we cannot exclude the possibility that events down-
stream of GTP hydrolysis (for example, inorganic phosphate re-
lease) play an important role (31). Ongoing efforts are directed at
experimentally validating the key features of our model.
Materials and Methods
Recombinant Protein Purification. GST-tagged proteins were produced as
previously described (8) with the exception that the truncation mutants
lacking transmembrane domains were purified in the absence of detergent.
atl(1–422) (pJM781) and atl(1–328) (pJM841) were expressed as GST-SUMO
fusion proteins. The N-terminal GST-SUMO–tagged fusion proteins were
bound to glutathione resin and purified in Tris buffer with 0.5% Triton X-
100 by cleaving off the tag with SENP2 protease. (See SI Materials and
Methods for SENP2 production.) The cleaved proteins were further purified
by anion-exchange chromatography in Tris buffer [20 mM Tris (pH 7.5), 200
mM KCl, 5% glycerol, and 0.5 mM tris(2-carboxyethyl)phosphine (TCEP)].
Purified proteins were stored as aliquots at −80 °C.
Atlastin Reconstitution and in Vitro Fusion Assays. WT atlastin, atl(1–450), atl
(1–471), and atl(1–497) were reconstituted into preformed 100-nm lip-
osomes as previously described (8). In vitro fusion assays were performed by
mixing labeled and unlabeled atl proteoliposomes (0.3 mM lipid each) in 96-
well FluoroNunc PolySorp plates (Nunc) and measuring nitrobenzoxadiazole
(NBD) fluorescence over time after the addition of 0.5 mM GTP and Mg2+.
Inhibition studies with soluble domains of atl were performed by adding the
soluble proteins to wells containing atl proteoliposomes before the addition
of 1–2 mM GTP and Mg2+.
GTPase Assays. GTPase activities were measured as previously described (8) by
measuring the release of inorganic phosphate from GTP as suggested in the
EnzChek Phosphate Assay Kit (Molecular Probes).
Analytical Ultracentrifugation. atl(1–422) and atl(1–328) were sized by sedi-
mentation velocity experiments carried out in a Beckman Optima XLA an-
alytical ultracentrifuge with 3.5 μM protein and 20 μM nucleotide in cells
assembled with 12-mm, two-channel aluminum centerpieces. Sedimentation
was carried out at 50,000 or 55,000 rpm in an AN60 rotor at 20 °C. The in-
tensity at 235 nm versus radial position was measured over 4–5 h at 4-min
intervals. All data analysis was done with Ultrascan II (34).
Cell Culture, Transfection, and Immunocytochemistry. Expression plasmids for
HA-tagged or Myc-tagged WT atlastin, atl(1–471), atl(1–497), and atl(1–422)
were transfected into Cos7 cells and immunostained by using standard pro-
tocols. Cells cotransfected with WT atlastin and GFP or WT atlastin and atl(1–
422) were scored for ER morphology and binned in three categories: fused
ER, partially fused ER, normal ER. Seven independent transfection experi-
ments were performed, and ∼100 cells were scored in each experiment.
The following antibodies were used: mouse anti-Myc (1:1,000; Sigma) and
rabbit anti-HA (1:500; Santa Cruz Biotechnology). Secondary antibodies for
immunofluorescence (Cy3 conjugates from Jackson Laboratories and Alexa
Fluor 488 conjugates from Invitrogen) were used at 1:1,000. Confocal images
were acquired with a Nikon C1 confocal microscope.
Detailed methods are available in SI Materials and Methods.
ACKNOWLEDGMENTS. We thank Borries Demeler for advice regarding
analytical ultracentrifugation and data analysis with Ultrascan II, as well as
Mike Stern and Joseph Faust for comments on the manuscript. Work in J.A.M.’s
laboratory is supported by funds from the National Institutes of Health
(Grant GM071832) and the G. Harold and Leila Y. Mathers Foundation. Work
in A.D.’s laboratory is supported by grants from the Italian Ministry of
Health, the Association Française contre les Myopathies, and the Fondazione
1. Wickner W, Schekman R (2008) Membrane fusion. Nat Struct Mol Biol 15:658–664.
2. McNew JA (2008) Regulation of SNARE-mediated membrane fusion during exocytosis.
Chem Rev 108:1669–1686.
3. Westermann B (2010) Mitochondrial fusion and fission in cell life and death. Nat Rev
Mol Cell Biol 11:872–884.
4. Griffin EE, Detmer SA, Chan DC (2006) Molecular mechanism of mitochondrial
membrane fusion. Biochim Biophys Acta 1763:482–489.
5. Okamoto K, Shaw JM (2005) Mitochondrial morphology and dynamics in yeast and
multicellular eukaryotes. Annu Rev Genet 39:503–536.
6. Burri L, et al. (2003) A SNARE required for retrograde transport to the endoplasmic
reticulum. Proc Natl Acad Sci USA 100:9873–9877.
7. Dilcher M, et al. (2003) Use1p is a yeast SNARE protein required for retrograde traffic
to the ER. EMBO J 22:3664–3674.
8. Orso G, et al. (2009) Homotypic fusion of ER membranes requires the dynamin-like
GTPase Atlastin. Nature 460:978–983.
9. Farhan H, Hauri HP (2009) Membrane biogenesis: Networking at the ER with atlastin.
Curr Biol 19:R906–R908.
10. Barlowe C (2009) Atlasin GTPases shape up ER networks. Dev Cell 17:157–158.
11. English AR, Zurek N, Voeltz GK (2009) Peripheral ER structure and function. Curr Opin
Cell Biol 21:596–602.
12. Moss TJ, Daga A, McNew JA (May 6, 2011) Fusing a lasting relationship between ER
tubules. Trends Cell Biol, 10.1016/j.tcb.2011.03.009.
13. Pendin D, McNew JA, Daga A (2011) Balancing ER dynamics: Shaping, bending,
severing, and mending membranes. Curr Opin Cell Biol, 10.1016/j.ceb.2011.04.007.
14. Zhao X, et al. (2001) Mutations in a newly identified GTPase gene cause autosomal
dominant hereditary spastic paraplegia. Nat Genet 29:326–331.
15. Fink JK (2003) Advances in the hereditary spastic paraplegias. Exp Neurol 184(Suppl 1):
16. Salinas S, Proukakis C, Crosby A, Warner TT (2008) Hereditary spastic paraplegia:
Clinical features and pathogenetic mechanisms. Lancet Neurol 7:1127–1138.
17. Soderblom C, Blackstone C (2006) Traffic accidents: Molecular genetic insights into
the pathogenesis of the hereditary spastic paraplegias. Pharmacol Ther 109:42–56.
18. Hazan J, et al. (1999) Spastin, a new AAA protein, is altered in the most frequent form
of autosomal dominant spastic paraplegia. Nat Genet 23:296–303.
19. Namekawa M, et al. (2006) SPG3A is the most frequent cause of hereditary spastic
paraplegia with onset before age 10 years. Neurology 66:112–114.
20. Park SH, Zhu PP, Parker RL, Blackstone C (2010) Hereditary spastic paraplegia proteins
REEP1, spastin, and atlastin-1 coordinate microtubule interactions with the tubular ER
network. J Clin Invest 120:1097–1110.
21. Voeltz GK, Prinz WA, Shibata Y, Rist JM, Rapoport TA (2006) A class of membrane
proteins shaping the tubular endoplasmic reticulum. Cell 124:573–586.
22. Evans K, et al. (2006) Interaction of two hereditary spastic paraplegia gene products,
spastin and atlastin, suggests a common pathway for axonal maintenance. Proc Natl
Acad Sci USA 103:10666–10671.
23. Sanderson CM, et al. (2006) Spastin and atlastin, two proteins mutated in autosomal-
dominant hereditary spastic paraplegia, are binding partners. Hum Mol Genet 15:
24. Fassier C, et al. (2010) Zebrafish atlastin controls motility and spinal motor axon
architecture via inhibition of the BMP pathway. Nat Neurosci 13:1380–1387.
25. Weber T, et al. (1998) SNAREpins: Minimal machinery for membrane fusion. Cell 92:
26. McNew JA, et al. (2000) Compartmental specificity of cellular membrane fusion
encoded in SNARE proteins. Nature 407:153–159.
27. Demeler B, van Holde KE (2004) Sedimentation velocity analysis of highly
heterogeneous systems. Anal Biochem 335:279–288.
28. Ivanova N, et al. (2007) Hereditary spastic paraplegia 3A associated with axonal
neuropathy. Arch Neurol 64:706–713.
29. Loureiro JL, et al. (2009) Novel SPG3A and SPG4 mutations in dominant spastic
paraplegia families. Acta Neurol Scand 119:113–118.
30. Tessa A, et al. (2002) SPG3A: An additional family carrying a new atlastin mutation.
31. Bian X, et al. (2011) Structures of the atlastin GTPase provide insight into
homotypic fusion of endoplasmic reticulum membranes. Proc Natl Acad Sci USA
32. Byrnes LJ, Sondermann H (2011) Structural basis for the nucleotide-dependent
dimerization of the large G protein atlastin-1/SPG3A. Proc Natl Acad Sci USA 108:
33. Ghosh A, Praefcke GJ, Renault L, Wittinghofer A, Herrmann C (2006) How guanylate-
binding proteins achieve assembly-stimulated processive cleavage of GTP to GMP.
34. Demeler B (2005) Ultrascan: A comprehensive data analysis software package for
analytical ultracentrifugation experiments. Modern Analytical Ultracentrifugation:
Techniques and Methods, eds Scott DJ, Harding SE, Rowe AJ (RSC Publishing,
Cambridge, UK), pp 210–230.
| www.pnas.org/cgi/doi/10.1073/pnas.1105056108 Moss et al.