A model for a partnership of lipid transfer proteinsand scramblases in membrane expansion andorganelle biogenesisAlireza Ghanbarpoura,1, Diana P. Valverdea,1, Thomas J. Meliaa,2,3, and Karin M. Reinischa,2,3

aDepartment of Cell Biology, Yale University School of Medicine, New Haven, CT 06520

Edited by Axel T. Brunger, Stanford University, Stanford, CA, and approved March 10, 2021 (received for review January 25, 2021)

The autophagy protein ATG2, proposed to transfer bulk lipid fromthe endoplasmic reticulum (ER) during autophagosome biogenesis,interacts with ER residents TMEM41B and VMP1 and with ATG9, inGolgi-derived vesicles that initiate autophagosome formation. Invitro assays reveal TMEM41B, VMP1, and ATG9 as scramblases. Wepropose a model wherein membrane expansion results from thepartnership of a lipid transfer protein, moving lipids between thecytosolic leaflets of apposed organelles, and scramblases that reequi-librate the leaflets of donor and acceptor organelle membranes aslipids are depleted or augmented. TMEM41B and VMP1 are implicatedbroadly in lipid homeostasis and membrane dynamics processes inwhich their scrambling activities likely are key.

TMEM41B | VMP1 | ATG9A | scramblase

Along-standing fundamental question in cell biology is howorganelles, such as the autophagosome, can form de novo.

The recent discovery that ATG2, required for early steps in auto-phagosome formation, is a member of a class of lipid transportprotein proposed to function in bulk lipid transfer suggests a modelof membrane growth (1–3). Namely, ATG2 could mediate lipidtransfer from the endoplasmic reticulum (ER), where most cel-lular lipid synthesis takes place, to the expanding isolation mem-brane. ATG2 localizes to contact sites where the ER and thenascent autophagosome are in close apposition (3). Based onstructural studies of ATG2 and related proteins, ATG2 forms abridge between the ER and the autophagosome with a hydro-phobic channel along which lipids could flow (4). However,transfer of lipids will occur only between the cytosolic leaflets ofthe apposed bilayers. Left unchecked, such a process would leadto bilayer asymmetry both in the ER, where lipids are depleted,and in the autophagosome, where lipids are augmented, andultimately to membrane destabilization. Thus, the model inwhich organelle expansion is supported by protein-mediated lipidtransfer predicts the existence of mechanisms, such as scramblases,to reequilibrate lipids between leaflets both in the lipid donor andlipid acceptor membranes. To test this model, we biochemicallycharacterized integral membrane proteins with reported roles inautophagosome biogenesis, finding that consistent with the modelthey are scramblases and, further, they interact physically withATG2. Although unanticipated, their direct interaction withATG2 builds confidence that these scramblases could partnerwith ATG2 in bulk lipid transfer.

ResultsWe identified three scramblase candidates, each implicated inearly autophagosome biogenesis events. ATG9A is present onGolgi-derived vesicles required to initiate autophagosome for-mation (5) while TMEM41B and VMP1 reside in a complex onthe ER and are necessary during autophagosome expansion (6–8).Intriguingly, TMEM41B and VMP1 share a six-helix transmem-brane domain also present in the DedA family of bacterial pro-teins, predicted half transporters with poorly understood roles inmembrane homeostasis (8).

We used a well-established fluorescence-based in vitro scram-bling assay (9) to assess whether TMEM41B, VMP1, or a 1:1mixture of the two proteins scrambles phospholipids (Fig. 1A andB). For this assay, the proteins were overexpressed in mammalian(Expi293F) cells and purified in n-dodecyl-β-D-maltoside deter-gent, then reconstituted into liposomes containing a small per-centage of nitrobenzoxadiazole (NBD)-labeled lipid. Dithionitereduces solvent-exposed NBD and quenches its fluorescence.Thus, in the absence of scrambling, a 50% reduction in fluo-rescence is expected as diothionite cannot access NBD in theliposome lumen. In the presence of a scramblase, NBD lipids arecontinuously exchanged between the leaflets of the bilayer, mak-ing all NBD accessible, so that the fluorescence reduction shouldbe complete (>>50%) in the ideal reconstitution scenario when100% of the liposomes incorporate the scrambling activity. Usingthis assay, we found that TMEM41B, VMP1, or a 1:1 mixture canscramble NBD-phosphatidylethanolamine (PE) (Fig. 1C). Inparallel, we used the same protocols to purify and reconstitutethe well-characterized intramembranous protease GlpG into lipo-somes, finding that GlpG does not scramble NBD-PE under ourconditions. We confirmed that the fluorescence reduction observedin the presence of TMEM41B/VMP1, TMEM41B, or VMP1, isnot due to leaky liposomes that have bilayer defects, for exampledue to incomplete detergent removal, so that dithionite couldpenetrate into the liposome lumen. To this end, we preparedproteoliposomes as before, but lacking NBD lipids, and in thepresence of NBD glucose. We found that the NBD glucose re-mains within the liposome lumen even when the liposomes areextensively dialyzed against NBD glucose-free buffer and that NBDglucose in the lumen is not affected by the addition of dithionite.We also show, using both the dithionite scrambling assay and asimilar “back extraction assay” (Fig. 1 and SI Methods), that lipidscrambling by TMEM41B/VMP1, TMEM41B, and VMP1 arenot specific to a particular glycerolipid (Fig. 1D) as both NBD-PEand NBD-phosphatidylcholine (PC) are substrates.We used the dithionite scrambling assay to show that ATG9A also

scrambles lipids, including PC, PE, and phosphatidylserine (Fig. 1E–G). Lipid scrambling by ATG9A was also recently reported byothers, who additionally showed by structure-based mutational

Author contributions: A.G., D.P.V., T.J.M., and K.M.R. designed research; A.G. and D.P.V.performed research; A.G., D.P.V., T.J.M., and K.M.R. analyzed data; and A.G., D.P.V.,T.J.M., and K.M.R. wrote the paper.

The authors declare no competing interest.

This open access article is distributed under Creative Commons Attribution License 4.0(CC BY).1A.G. and D.P.V. contributed equally to this work.2T.J.M. and K.M.R. contributed equally to this work.3To whom correspondence may be addressed. Email: thomas.melia@yale.edu or Karin.Reinisch@Yale.edu.

This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.2101562118/-/DCSupplemental.

Published April 13, 2021.

PNAS 2021 Vol. 118 No. 16 e2101562118 https://doi.org/10.1073/pnas.2101562118 | 1 of 4

CELL

BIOLO

GY

BRIEFRE

PORT

Dow

nloa

ded

by g

uest

on

Nov

embe

r 25

, 202

1

A B C

D

E F G

Fig. 1. TMEM41B, VMP1, and ATG9A scramble lipids in vitro. (A and B) Schematics for the scrambling and “leakiness” assays. (C) 1:1 mixtures of TMEM41Band VMP1 or TMEM41B or VMP1 alone, can scramble NBD-PE. Scrambling is not observed in liposomes reconstituted with the control protein GlpG or inempty liposomes. Reconstitution into liposomes is more efficient when the proteins are added at higher concentrations, resulting in near total reduction offluorescence. (D) TMEM41B/VMP1, TMEM41B, and VMP1 also scramble NBD-PC. Additionally, a bovine serum albumin (BSA) back extraction assay in whichBSA instead of dithionite was added to liposomes/proteoliposomes, shows that TMEM41B/VMP or TMEM41B or VMP1 alone scramble PC. In this assay, theNBD-PC located in the outer leaflet of the liposome is extracted and quenched by fatty-acid-free BSA instead of being reduced by dithionite. BSA bindingreduces NBD-PC fluorescence by half versus NBD-PC fluorescence in liposomes. Thus, fluorescence would be reduced to 75% of initial levels in the absence of ascramblase and to ∼50% if there is scrambling. All experiments were repeated at least three times; SD indicated. (E–G) Scrambling assays for Atg9. Thereduction in fluorescence for ATG9 (to ∼35% of initial levels) is less than for TMEM41B or VMP1; this does not imply less scrambling activity for ATG9 butrather less efficient reconstitution of functional ATG9 into liposomes. Protein concentrations throughout refer to the monomeric forms of the proteins.

2 of 4 | PNAS Ghanbarpour et al.https://doi.org/10.1073/pnas.2101562118 A model for a partnership of lipid transfer proteins and scramblases in membrane expansion

and organelle biogenesis

Dow

nloa

ded

by g

uest

on

Nov

embe

r 25

, 202

1

analysis that this scrambling activity is essential for autophago-some growth (10, 11).We next explored whether ATG9A and TMEM41B/VMP1

might interact directly with the lipid transport protein ATG2A.We coexpressed 3XFLAG-tagged ATG9A and untagged ATG2Ain Expi293F cells, finding that ATG2A robustly copurifies withATG9A in an affinity purification, associating nearly stoichiometrically(Fig. 2A). To assess whether ATG2A might interact with TMEM41Bor VMP1, we overexpressed ATG2A in Expi293F cells, and passedthe lysate over 3XFLAG-TMEM41B or -VMP1 immobilized onanti-FLAG resin. ATG2A is robustly retained by both TMEM41Band VMP1, indicating that it interacts with either protein.ATG2A is an elongated structure proposed to interact with

apposed ER and autophagosome membranes via its ends, withits N terminus at one and C-terminal portions at its other end(3). If so, ATG2A should interact with TMEM41B/VMP1 via oneend and ATG9A via the other. To date, we are only able to makea soluble N-terminal, but not C-terminal fragment of ATG2A. Wefind that this fragment, mini-ATG2A (residues 1 to 345), associates

with TMEM41B and VMP1 but not ATG9A in interaction ex-periments similar to the ones described above for the full-lengthprotein (Fig. 2A and B). To increase confidence in our results, wealso used mini-ATG2A in flotation assays with either protein-freeliposomes or liposomes reconstituted with ATG9A, TMEM41B,VMP1, or TMEM41B/VMP1 (Fig. 2C). In this experiment, mini-ATG2A was incubated with liposomes, and then liposomes and anyassociated mini-ATG2A were separated from unbound protein bydensity gradient centrifugation. We found that mini-ATG2A asso-ciates robustly only with liposomes reconstituted with TMEM41B,VMP1, or TMEM41B/VMP1 but not empty or ATG9A-containingliposomes. (Intact ATG2A associates nonspecifically with liposomesand so flotation assays with the full-length protein were not in-formative.) These data are consistent with the “bridge” model(Fig. 2D). Although a direct interaction between the lipid trans-port protein and scramblases should not be necessary for bulklipid transport in principle, such interactions between ATG2A andATG9A or TMEM41B/VMP1 suggest there may be inherent ad-vantages to coupling scramblase and transport activities spatially.

ATG2

ATG9

ER Lumen

VMP1/TMEM41B

cytosol

IM lumen

anti-FLAGheavy chain

ATG9A

ATG2A

<chaperone

ATG2AFLAG-ATG9A

mini-ATG2A-his6

+ + -- + +- - +

+ + -- + +- - +

anti-ATG2

anti-ATG9A

Coomassie Western blotsanti-HIS

(in mini-ATG2)

+ + -- + +- - +

Co-immunoprecipitationExpression

Western blotsmini-

ATG2

250

150

100

50

75

MW(kDa)

0.12% of input

*

anti-HIS(in mini-ATG2)

anti-HIS(in mini-ATG2)

0.0

0.5

1.0

norm

ailiz

ed re

lativ

e in

tens

ity

(”To

p”/”B

otto

m” m

ini-A

TG2)

+---+

++--+

+-+-+

+++-+

+--++

liposomesVMP1

TMEM41BATG9A

mini-ATG2A-his6

“Top” fraction: membrane-associated

mini-ATG2A

“Bottom” fraction: soluble

mini-ATG2Aliposomes

VMP1TMEM41B

ATG9Amini-ATG2A-his6

+---+

++--+

+-+-+

+++-+

+--++

+---+

++--+

+-+-+

+++-+

+--++

anti-hisWestern blot

ATG9A interactions with ATG2A TMEM41B, VMP1 interactions with ATG2A

0%

20%

10%

5%

liposomes/proteoliposomes+ mini-ATG2

150K g’s

liposomes/proteoliposomesand associated mini-ATG2

unassociated mini-ATG2

20%

10%

5%stepped density gradient:

% iodixanol indicated

before centrifugation after centrifugation

Flotation Assay:

“Top” fraction

“Bottom” fraction

mini-ATG2A interactions with TMEM41B, VMP1, not ATG9A

A B

C

D

ATG2AVMP1

TMEM41B

+--

+-+

++-

--+

-+-

MW(kDa)

250

50

37

ATG2A* *

VMP1**

anti-FLAGheavy chain

TMEM41B* *

<chaperone<

<

0.06% inputanti-ATG2 Western blot:

Coomassie:Coomassie:

mini-ATG2A-his6VMP1

TMEM41B

anti-his Western blot:0.06% input

VMP1 & mini-ATG2co-migrate

TMEM41B

+--

+-+

++-

--+

-+-

*

E TMEM41B/VMP1

ATG9

ATG2

ER/”omegasome”

Fig. 2. The lipid transporter ATG2A interacts with ATG9, TMEM41B, and VMP1 and could function as a lipid transport bridge. (A) 3XFLAG-ATG9A wascoexpressed with either untagged intact ATG2A or mini-ATG2A-his6 (residues 1 to 345). Then 3XFLAG-ATG9A was bound to anti-FLAG resin along withassociated proteins. ATG9A copurifies with full-length ATG2A but not with mini-ATG2A. (B) 3XFLAG-TMEM41B or -VMP1 were bound to anti-FLAG resinincubated with cell lysates containing either untagged ATG2A or mini-ATG2A-his6 and washed. Both TMEM41B and VMP1 robustly pull down ATG2A andinteract, though more weakly, with mini-ATG2A. All experiments in A and B were repeated at least three times; representative Coomassie-stained gels andWestern blots are shown. (C) In the flotation assay, mixtures of liposomes or proteoliposomes and mini-ATG2A-his6 were loaded onto a density gradient andcentrifuged. Liposomes/proteoliposomes and associated proteins float at the top, whereas soluble proteins remain at the bottom of the gradient. Top andBottom fractions were analyzed for mini-ATG2A by sodium dodecyl sulfate polyacrylamide gel electrophoresis and Western blotting. Mini-ATG2 associateswith proteoliposomes containing TMEM41B, VMP1, or their mixture, but not empty or ATG9A-containing liposomes, consistent with interactions in A andB. The experiment was repeated three times. Quantification, with SD indicated. (D) Model for ATG2A-mediated lipid transport, in which one end ofATG2A associates with a scramblase complex comprising TMEM41B/VMP1 at the ER, and the other end associates with ATG9A in the nascent auto-phagosome, allowing lipids to flow from the ER to the autophagosome. TMEM41B/VMP1 equilibrates the two leaflets of the ER membrane following lipidextraction from the cytosolic leaflet, and ATG9A equilibrates the leaflets of the isolation membrane following lipid delivery to the cytosolic leaflet,allowing for isolation membrane expansion. (E) ATG2 and scramblases could partner to expand an ATG9-containing vesicle into a cup-shaped doublemembrane like the autophagosome.

Ghanbarpour et al. PNAS | 3 of 4A model for a partnership of lipid transfer proteins and scramblases in membrane expansionand organelle biogenesis

https://doi.org/10.1073/pnas.2101562118

CELL

BIOLO

GY

BRIEFRE

PORT

Dow

nloa

ded

by g

uest

on

Nov

embe

r 25

, 202

1

DiscussionWe consider plausible that the autophagosome could grow evenfrom a single ATG9-containing vesicle, acting as a seed mem-brane. In this scenario, ATG2 allows lipid transport from theER to the seeding vesicle, with TMEM41B and VMP1 reequi-librating the leaflets of the ER as lipids are extracted and ATG9in the “seed” scrambling ER-derived lipids as they are delivered(see also refs. 10, 11). This would allow for the expansion of themembrane surface area of the seed even while the volume ofcontents enclosed within the membrane remains relatively con-stant. Expansion in this way would result in a double membranestructure like the autophagosome. High membrane curvature isenergetically costly, so the nascent autophagosome would notform a double membrane sheet, with an expansive high curvaturecircumference, but would instead spontaneously curl up into acup-shaped structure (12), with a much smaller high curvaturearea, as observed in the maturing autophagosome. Interestingly,a protein structurally related to ATG2, VPS13, plays a role inprospore formation in yeast and acrosome formation in humans,where both the prospore and acrosome are also cup-shapeddouble-membrane structures initiated from a small number ofvesicles (13, 14). Autophagosomes, prospores, and acrosomes mayarise via similar mechanisms involving a partnership of lipidtransfer proteins and scramblases. Of note, human VPS13A wasreported to form a complex with a predicted scramblase, XK (15).VMP1 and TMEM41B have both been implicated in multiple

processes other than autophagy, all potentially associated in someway with membrane dynamics. The discovery that VMP1 andTMEM41B are scramblases can explain much of the apparentlydisparate biology associated with these proteins. First, both VMP1and TMEM41B are implicated in lipid homeostasis as depletion ofeither leads to the accumulation of neutral lipids into oversizedcytoplasmic lipid droplets (7, 8, 16). Both proteins are enriched at

organelle-organelle contact sites along with key lipid-synthesismachinery (16–18), and thus are ideally situated to facilitate re-distribution of lipids from the ER via soluble lipid transfer pro-teins localized at contact sites. The defects in lipid homeostasismay reflect challenges in lipid-synthesis and neutral lipid con-sumption when efficient efflux of phospholipids is hindered.Further, VMP1 was also recently shown as essential for lipo-protein production (19). Lipoprotein assembly takes place in theER lumen, where apolipoproteins are assembled with lipidsasymmetrically derived from the luminal leaflet of the ER mem-brane prior to packaging into vesicles for transport out of the cell.Depletion of VMP1 leads to the production of amorphouslipoprotein-like particles that suggest partial budding into boththe cytoplasmic and luminal leaflets simultaneously, and a com-plete loss of effective secretion. This suggests that analogous tocoupling of scramblases and cytoplasmic lipid transport proteins,there is a comparable functional coupling of scramblase activityand the effective building of lipoprotein particles. And lastly, theproteins were recently reported as essential for corona- and fla-virus replication, although no mechanisms were identified (20, 21).These viruses form replication compartments derived from theER membrane. Coronavirus replication compartments in partic-ular, described as double membrane spherical structures, resembleautophagosomes. We speculate that the replication compartmentsmight form de novo, like the autophagosome, involving a part-nership between lipid transfer proteins and scramblases.

Data Availability. All study data are included in the article and/orSI Appendix.

ACKNOWLEDGMENTS. This work was supported by funding from the NIH(GM135290 to T.J.M., GM131715 to K.M.R., and F32 GM137568 to A.G.) andan NSF award to D.P.V.

1. S. Maeda, C. Otomo, T. Otomo, The autophagic membrane tether ATG2A transferslipids between membranes. eLife 8, e45777 (2019).

2. T. Osawa et al., Atg2 mediates direct lipid transfer between membranes for auto-phagosome formation. Nat. Struct. Mol. Biol. 26, 281–288 (2019).

3. D. P. Valverde et al., ATG2 transports lipids to promote autophagosome biogenesis.J. Cell Biol. 218, 1787–1798 (2019).

4. W. A. Prinz, J. H. Hurley, A firehose for phospholipids. J. Cell Biol. 219, e202003132 (2020).5. A. R. Young et al., Starvation and ULK1-dependent cycling of mammalian Atg9 be-

tween the TGN and endosomes. J. Cell Sci. 119, 3888–3900 (2006).6. C. J. Shoemaker et al., CRISPR screening using an expanded toolkit of autophagy

reporters identifies TMEM41B as a novel autophagy factor. PLoS Biol. 17, e2007044(2019).

7. F. Moretti et al., TMEM41B is a novel regulator of autophagy and lipid mobilization.EMBO Rep. 19, e45889 (2018).

8. K. Morita et al., Genome-wide CRISPR screen identifies TMEM41B as a gene requiredfor autophagosome formation. J. Cell Biol. 217, 3817–3828 (2018).

9. B. Ploier, A. K. Menon, A fluorescence-based assay of phospholipid scramblase ac-tivity. J. Vis. Exp. 54635 (2016).

10. K. Matoba et al., Atg9 is a lipid scramblase that mediates autophagosomal membraneexpansion. Nat. Struct. Mol. Biol. 27, 1185–1193 (2020).

11. S. Maeda et al., Structure, lipid scrambling activity and role in autophagosome for-mation of ATG9A. Nat. Struct. Mol. Biol. 27, 1194–1201 (2020).

12. U. Seifert, Configurations of fluid membranes and vesicles. Adv. Phys. 46, 13–137

(1996).13. J. S. Park, A. M. Neiman, VPS13 regulates membrane morphogenesis during sporu-

lation in Saccharomyces cerevisiae. J. Cell Sci. 125, 3004–3011 (2012).14. R. Da Costa et al., Vps13b is required for acrosome biogenesis through functions

in Golgi dynamic and membrane trafficking. Cell. Mol. Life Sci. 77, 511–529 (2020).15. Y. Aoki et al., Phosphorylation of serine 114 on Atg32 mediates mitophagy.Mol. Biol.

Cell 22, 3206–3217 (2011).16. L. C. Tábara, R. Escalante, VMP1 establishes ER-microdomains that regulate mem-

brane contact sites and autophagy. PLoS One 11, e0166499 (2016).17. L. C. Tábara et al., Vacuole membrane protein 1 marks endoplasmic reticulum sub-

domains enriched in phospholipid synthesizing enzymes and is required for phos-

phoinositide distribution. Traffic 19, 624–638 (2018).18. Y. G. Zhao et al., The ER-localized transmembrane protein EPG-3/VMP1 regulates

SERCA activity to control ER-isolation membrane contacts for autophagosome for-

mation. Mol. Cell 67, 974–989.e6 (2017).19. H. Morishita et al., A critical role of VMP1 in lipoprotein secretion. eLife 8, e48834 (2019).20. W. M. Schneider et al., Genome-scale identification of SARS-CoV-2 and pan-coronavirus

host factor networks. Cell 184, 120–132.e14 (2021).21. H. H. Hoffmann et al., TMEM41B is a pan-flavivirus host factor. Cell 184, 133–148.e20

(2021).

4 of 4 | PNAS Ghanbarpour et al.https://doi.org/10.1073/pnas.2101562118 A model for a partnership of lipid transfer proteins and scramblases in membrane expansion

and organelle biogenesis

Dow

nloa

ded

by g

uest

on

Nov

embe

r 25

, 202

1