ATG2A-mediated DAG transfer recruits DGAT2 for lipid droplet growth

Helin Elhan; Alicia Damm; Justin L Korfhage; Daniel Álvarez; Mehdi Zouiouich; Francesca Giordano; Stefano Vanni; Thomas J Melia; Abdou Rachid Thiam

doi:10.1038/s41594-025-01689-0

. 2025 Nov 17;32(12):2601–2613. doi: 10.1038/s41594-025-01689-0

ATG2A-mediated DAG transfer recruits DGAT2 for lipid droplet growth

Helin Elhan ¹, Alicia Damm ¹, Justin L Korfhage ², Daniel Álvarez ^3,⁴, Mehdi Zouiouich ¹, Francesca Giordano ^5,⁶, Stefano Vanni ^3,⁷, Thomas J Melia ², Abdou Rachid Thiam ^1,^✉

PMCID: PMC12700830 PMID: 41249819

Abstract

Lipid droplet (LD) growth mechanisms and the roles of LD-associated lipid transfer proteins remain poorly understood. Here we show that the autophagy lipid transfer protein ATG2A has an anabolic role and promotes LD expansion by transferring diacylglycerol (DAG), triacylglycerol (TAG) and phosphatidic acid, from the endoplasmic reticulum to LDs. In ATG2A deficiency, synthesized lipids are incorporated inefficiently into LDs and assemble new LDs. In addition, DAG O-acyltransferase 2 (DGAT2), which synthesizes TAG and expands LD, fails to relocate to LDs. In vitro, DAG recruits DGAT2 to LDs. These findings support the idea that ATG2A-mediated DAG transfer recruits DGAT2 to LDs, promoting LD expansion. ATG2A alone promotes LD growth by transferring TAG and DAG, but its effectiveness in LD expansion is reduced when DGAT2 is inhibited. This synergistic action with DGAT2 prevents the buildup of nonmembrane lipids within the endoplasmic reticulum and favors TAG synthesis on the LD surface.

Subject terms: Endoplasmic reticulum, Membrane structure and assembly, Membrane trafficking, Membrane lipids

Elhan et al. show that ATG2A acts with DGAT2, the enzyme producing triacylglycerol (TAG), in lipid droplet growth. By delivering diacylglycerol to lipid droplets, ATG2A not only fuels TAG production but also promotes the recruitment of DGAT2 to droplet surfaces.

Main

The endoplasmic reticulum (ER) orchestrates protein and lipid synthesis, among other vital functions¹. Its integrity relies on several quality control mechanisms^2,3. Stress, such as excess misfolded proteins or lipid imbalances, triggers degradation pathways to remove defective components, preserving ER function^4,5. A key aspect of ER quality control is avoiding the buildup of nonmembrane lipids—fatty acids, phosphatidic acid (PA), diacylglycerol (DAG) and triacylglycerol (TAG)—which alter membrane properties⁶. These nonmembrane lipids are either converted into membrane-compatible phospholipids or stored in lipid droplets (LDs).

LDs regulate lipid metabolism and cellular homeostasis by mitigating various stresses, such as ER stress resulting from lipid imbalances^5,7–12. During lipogenesis, TAGs are synthesized in the ER from PA and DAG⁸, which also have signaling functions. However, they can alter membrane properties, and this is mitigated by TAG phase separation into nascent LDs¹³, catalyzed by factors such as membrane curvature and lipid composition^14,15. Elevated DAG levels enhance TAG LD nucleation^14,16,17. Seipin, an ER protein, aids LD nucleation by interacting with TAGs^18–20. Newly formed LDs at seipin gather more TAGs, budding into the cytosol to mature^20–22. The detailed mechanisms of these steps remain elusive^13,23. Mature LDs have a neutral lipid core covered by a phospholipid monolayer and proteins that influence their function.

Key enzymes for TAG synthesis, such as glycerol-3-phosphate acyltransferase (GPAT)4, acyl-CoA synthetase long-chain family member (ACSL)3, Lipin-1γ and DAG O-acyltransferase (DGAT)2, localize to the LD surface in various cell types, indicating that lipid synthesis occurs directly there^24,25. Lipin-1γ targets ER and LDs from the cytosol. However, GPAT4, ACSL3 and DGAT2 are ER embedded and diffuse to reach the LD surface^24,26,27. The mechanisms for their relocalization, substrate access and the importance of lipid synthesis on LDs versus in the ER remain unclear^9,28.

Lipid substrates PA and DAG in the ER may diffuse laterally to the LD and fuel lipid synthesis on LDs. However, lipid transfer proteins (LTPs) are located at the ER–LD interface^29–35, especially during lipogenesis^30,34. LTPs such as ORP5/8 and MOSPD2 transport one lipid at a time at organelle contact sites^34,36. In contrast, bridging lipid transfer proteins (BLTPs), for example, VPS13A/C and ATG2A, create a tunnel that facilitates efficient lipid flux between organelles^31,37. Although the function of ATG2A on LDs may be independent of autophagy^38,39, its role has often been studied in catabolic processes. Its ability to transport various lipids raises the question of whether it could also function in anabolic processes.

The consistent association of ATG2A with LDs indicates an essential function, which remains unclear owing to its role in both autophagy and LD biology. Although these processes are interconnected, studies comparing ATG2A knockout (KO) phenotypes with those of other autophagy-related protein KOs during catabolism^32,33,38,39 suggest somewhat different implications for ATG2A in LDs and autophagy.

ATG2A transfers phospholipids from the ER to autophagosomes for growth^37,40, working with scramblases VMP1 and TMEM41B on the ER, and ATG9 on autophagosomes^41–45. This coordination maintains membrane phospholipid density, symmetry and mechanical properties. In contrast, LDs have different constraints: their phospholipid monolayer does not require the same homeostasis as bilayer membranes and can lose phospholipids without compromising integrity⁴⁶. The limited LD surface area and monolayer structure suggest that LDs probably do not supply phospholipids directly, and scramblases are irrelevant. Therefore, the idea of ATG2A continually transferring phospholipids to or from LDs seems unlikely. Thus, ATG2A may help facilitate the movement of lipids into and out of LDs, rather than just phospholipids. For example, during starvation, ATG2A appears to transfer fatty acids from LDs to mitochondria for energy production^39,47. This indicates a role for ATG2A in lipid mobilization. However, whether ATG2A transports lipids toward LDs, for storage or other purposes, remains an open question.

Here, we examined whether ATG2A has a role in lipid storage within LDs and emphasized its involvement in transferring nonmembrane lipids such as DAG, TAG and PA from the ER to LDs during lipogenesis.

Results

ATG2A promotes LD growth during lipid storage

To explore the role of ATG2A in lipid storage, we compared LDs in wild-type (WT) and ATG2A/B double KO (DKO) HeLa cells. Under standard culture conditions, DKO cells had more LDs, while WT cells had fewer but larger LDs (Fig. 1a,b, 0 µM oleic acid (OA)). The absence of ATG2 impairs autophagy and disrupts lipophagy^39,47, which could explain our observed phenotypes. Thus, to specifically assess the direct effect of ATG2A in lipid storage, we induced LD formation by loading the cells with OA to predispose them to storage, stimulating anabolic over catabolic processes. Cells were exposed to 50, 100 or 200 µM OA for 24 h. Both WT and DKO showed comparable LD volume and LD sizes increased with OA concentration (Extended Data Fig. 1a–d). LDs were larger in WT than in DKO cells, which formed more numerous, smaller LDs (Fig. 1b). In a time-course experiment, this pattern persisted after 1 and 4 h of 200 µM OA treatment: WT cells had fewer, larger LDs, while DKO cells consistently showed smaller, more abundant LDs (Extended Data Fig. 1e–g). Importantly, ATG2A–eGFP transfection in DKO cells partially restored the WT LD size phenotype, highlighting the role of ATG2A in regulating neutral lipid packaging (Extended Data Fig. 1h,i).

Fig. 1 — a,b, The induction of LD formation leads to smaller but more numerous droplets in the absence of ATG2: confocal images of HeLa WT and ATG2 DKO cells ±24 h incubation with 200 µM OA (a) and quantification of LD size (µm²) and number per cell (b). In the SuperPlots, the small dots correspond to individual cells, while the large dots correspond to the mean value of all cells from one independent biological replicate. n = 88 (WT 0 µM), 106 (DKO 0 µM), 122 (WT 200 µM) and 110 (DKO 200 µM) cells. Data were collected from three independent biological replicates and are presented as mean ± s.d. Statistical analysis was performed using a two-tailed Mann–Whitney test; **** indicates P < 0.0001. c,d, ATG9 deficiency does not affect LD morphology: confocal images of WT and ATG9 KO cells ±24 h OA (c) and quantification as in b (d). Small dots indicate single cells and large dots indicate replicate means. n = 133 (WT 0 µM), 129 (ATG9 KO 0 µM), 97 (WT 200 µM) and 163 (ATG9 KO 200 µM) cells. Data were obtained from three independent biological replicates and are shown as mean ± s.d. Statistical analysis was performed using a two-tailed Mann–Whitney test. e,f, ATG2A promotes fatty acid incorporation into pre-existing LDs: confocal images of ATG2 DKO cells after a 1 h BPY-C12 pulse following OA preload (e) and a time course of BPY-C12 incorporation into LDs in DKO and ATG2A-rescued cells over 60 min (f). ATG2A-rescued DKO cells show faster and more efficient incorporation compared with control cells. n = 4 cells per condition across three independent biological replicates. Data are presented as mean ± s.e.m. g, Quantification of BPY-C12 signal in DKO (−) and ATG2A-rescued (+) cells. Small dots represent single cells and large dots represent replicate means. n = 21 cells per condition. Data are from three independent biological replicates and are shown as mean ± s.d. Statistical analysis was performed using a two-tailed unpaired Student’s t-test; *P = 0.0137. h, Lipidomics analysis of WT and DKO cells preloaded with OA and pulsed with OA/NBD-palmitate. NBD-TAG levels were higher in WT cells overexpressing (oe) ATG2A compared with DKO. Data are from three independent biological replicates, with dots representing replicate means and are shown as mean ± s.d. Statistical analysis was performed using a two-tailed unpaired Student’s t-test; **P = 0.0061. i,j, The absence of ATG2 promotes de novo LD nucleation: confocal images of DKO cells after BPY-C12 pulse (i) and quantification of the number of particles lacking LipidTOX signal in DKO (−) and ATG2A-rescued (+) cells (j). Small dots represent individual cells and large dots represent replicate means. n = 7 cells for DKO and n = 9 cells for ATG2A rescued. Data are from three independent biological replicates and are shown as mean ± s.d. Statistical analysis was performed using a two-tailed unpaired Student’s t-test; ****P < 0.0001. k,l, TD-ATG2A mutant is defective in lipid delivery to LDs: confocal images of ATG2 DKO cells expressing TD-ATG2A–eGFP mutant after a 1 h pulse with BPY-C12, showing defective lipid delivery to LDs (k) and quantification of BPY-C12 incorporation into LDs in TD-ATG2A-expressing cells (+) compared with DKO controls (−) (l). Small dots represent individual cells and large dots represent replicate means. n = 13 cells for DKO and n = 10 cells for TD-ATG2A-expressing cells. Data are from three independent biological replicates and are presented as mean ± s.d. Statistical analysis was performed using a two-tailed unpaired Student’s t-test. m, Quantification of the number of particles lacking LipidTOX signal in DKO (−) and TD-ATG2A-expressing (+) cells. Small dots represent individual cells and large dots represent replicate means. n = 10 for DKO and n = 11 for TD-ATG2A-expressing DKO cells. Data are from three independent biological replicates and are presented as mean ± s.d. Statistical analysis was performed using a two-tailed unpaired Student’s t-test. Scale bars, 10 µm; insets, 1 µm.

Source data

Extended Data Fig. 1 — **(a-c)** Induction of LD formation leads to smaller but more numerous droplets in the absence of ATG2. **(a)** Confocal images of HeLa WT and ATG2 DKO cells ± 24 h incubation increasing concentration of OA (0, 50, 100, 200 µM). LDs stained with BODIPY FL and nuclei with Hoechst. **(b-c)** Quantification of LD size (µm²) **(b)** and number **(c)** from (a). Data are shown as SuperPlots, where small dots represent individual cells (n = 88 (WT 0 µM); 106 (DKO 0 µM); 108 (WT 50 µM); 116 (DKO 50 µM); 92 (WT 100 µM); 75 (DKO 100 µM); 122 (WT 200 µM) and 110 (DKO 200 µM) cells) and large color-coded dots represent replicate means from 3 independent experiments. Data are presented as mean ± SD. Two-tailed Mann-Whitney test was used for statistical analysis; **p = 0.003, ****p < 0.0001. **(d)** Total LD volume per cell based on values shown in (b-c). Data are shown as SuperPlots, where small dots represent individual cells and large dots represent replicate means from 3 independent experiments. Data are presented as mean ± SD. Two-tailed Mann-Whitney test was used for statistical analysis. **(e)** Confocal images of WT and ATG2 DKO cells with 200 µM OA feeding during 1 h or 4 h. LDs were stained with BODIPY FL and the nuclei with Hoechst. **(f-g)** LDs size **(f)** and number **(g)** in cells shown in (e). Data are shown as SuperPlots, where small dots represent individual cells (n = 110 (WT 1 h); 88 (DKO 1 h); 126 (WT 4 h) and 84 (DKO 4 h) cells) and large color-coded dots represent replicate means from 3 independent experiments. Data are presented as mean ± SD. Two-tailed Mann-Whitney test was used for statistical analysis; ****p < 0.0001. **(h)** Confocal microscopy images of ATG2A-rescued DKO cells. Two examples are shown: on the left, transfected cells highlight an increase in LDs size in the presence of the protein. On the right, two cells in the same focal plane are displayed: one transfected (green square) and the other non-transfected, with endogenous levels of ATG2 (red square). LDs were stained with LipidTOX. **(i)** LDs size and number per cell under DKO (-) and ATG2A-rescued (+) conditions shown in (h). Data are shown as SuperPlots, where small dots represent individual cells and large dots represent replicate means from 3 independent experiments (n = 45 cells per condition). Data are presented as mean ± SD. Two-tailed Mann-Whitney test was used for statistical analysis; **p = 0.0027. **(j)** Confocal microscopy image of WT cell transfected with ATG2A–eGFP for 24 h, followed by treatment with OA for 24 h. Both a transfected cell (green square) and a non-transfected cell (red square) are visible within the same focal plane. LDs were stained with LipidTOX. **(k)** LDs size in non-transfected WT cells with endogenous ATG2A expression levels (-) compared to cells overexpressing ATG2A–eGFP (+). Data are shown as SuperPlots, where small dots represent individual cells (n = 41 cells for WT and 33 cells for WT + ATG2A condition) and large dots represent replicate means from 3 independent experiments. Data are presented as mean ± SD. Two-tailed unpaired Student’s t-test was used for statistical analysis; ****p < 0.0001. **(l)** Total LD volume per cell based on values shown in Fig. 1d. Data are shown as SuperPlots, where small dots represent individual cells and large dots represent replicate means from 3 independent experiments. Data are presented as mean ± SD. Two-tailed Mann-Whitney test was used for statistical analysis. **(m)** Mass spectrometry-based lipidomics of C18:1 PA levels in WT and DKO cells. PA levels (mol%) were quantified by mass spectrometry and normalized to the mean of WT (set to 1). Each dot represents one biological replicate. Data are presented as mean ± SD. Two-tailed unpaired Student’s t-test was used for statistical analysis; *p = 0.0253. Scale bars for all images: 10 µm; insets: 1 µm.

Source data

Next, we overexpressed ATG2A in WT cells, which have normal autophagy. ATG2A localized to LDs and yielded larger LDs than nontransfected cells (Extended Data Fig. 1j,k), indicating that it increases LD size. Then, we used ATG9 KO cells, which are autophagy deficient but retain functional ATG2, to assess whether they recapitulate the DKO phenotype. Upon OA supplementation, ATG9 KO produced a comparable total LD volume to WT cells. Although the number of LDs was slightly higher in ATG9 KO cells, the difference was not statistically significant, and LD size remained similar between ATG9 KO and WT cells (Fig. 1c,d and Extended Data Fig. 1l). This contrasts sharply with the phenotype observed in ATG2 DKO cells, supporting our model that ATG2A promotes LD growth independently of autophagy. Thus, ATG2A may transport lipids from the ER to LDs, thereby accelerating lipid storage in LDs and reducing their accumulation in the ER.

ATG2A promotes lipid incorporation into LDs and limits LD nucleation

To investigate the role of ATG2A in lipid incorporation, we performed a pulse-chase experiment. We transfected WT or DKO cells with ATG2A–eGFP and loaded them with OA for 24 h. This preloading created enough pre-existing LDs, bound or not by ATG2A, to examine the protein impact on lipid incorporation. We then performed a 1-h pulse with OA containing 1% BODIPY-C12 fatty acid tracer (BPY-C12) and track incorporation into LDs (Fig. 1e).

We compared BPY-C12 incorporation into LDs between ATG2A-transfected and nontransfected DKO cells. ATG2A-rescued cells showed significantly higher incorporation into LDs (Fig. 1f,g). In WT cells, overexpressing ATG2A also increased BPY-C12 incorporation (Extended Data Fig. 2a–c). In contrast, WT and ATG9 KO cells exhibited similar incorporation levels, indicating that lipid uptake into LDs in this context is independent of autophagy (Extended Data Fig. 2d,e). Moreover, in a pulse-labeling experiment, when OA was combined with 7-nitrobenz-2-oxa-1, 3-diazol-4-yl (NBD)-palmitic acid, we observed higher levels of NBD-TAG in WT cells overexpressing ATG2A compared with DKO cells, which accumulated more PA (Fig. 1h and Extended Data Fig. 1m). Overall, these findings support the idea that ATG2A helps transfer lipids to LDs for storage.

Extended Data Fig. 2 — **(a)** Confocal images of HeLa WT cells at 0 and 60 minutes after a BPY-C12 pulse, following a 24 h OA preloading to generate mature LDs. The same focal plane shows a non-transfected cell (red square) and a neighboring ATG2A–eGFP-expressing cell (green square). Pre-existing LDs are labeled with LipidTOX. Split channels and zoomed insets on the right highlight stronger BPY-C12 accumulation in LDs of ATG2A-expressing cells. Representative of 3 independent experiments. **(b)** Quantification of BPY-C12 signal in LDs after 1 h pulse in ATG2 DKO cells, with (+) or without ATG2A expression (-). ATG2A expression significantly increases fatty acid incorporation into LDs. Data are shown as SuperPlots, where small dots represent individual cells (n = 41 cells for WT and 33 cells for WT + ATG2A condition) and large dots represent replicate means from 3 independent experiments. Data are presented as mean ± SD. Two-tailed Mann-Whitney test was applied for statistical analysis; ****p < 0.0001. **(c)** Time course of BPY-C12 incorporation into LDs in WT and cells over 60 minutes. WT cells overexpressing ATG2A show faster and more efficient incorporation compared to control cells. n = 9 cells per condition across 3 independent biological replicates. Data are presented as mean ± SEM. **(d)** Confocal images of ATG9 KO cells after 1 h pulse with BPY-C12 following 24 h OA preloading. LDs were stained with LipidTOX. Representative of 3 independent experiments. **(e)** Time-course of BPY-C12 incorporation into LDs over 60 minutes in WT and ATG9 KO cells. Similar incorporation kinetics suggest that fatty acid transfer into LDs occurs independently of autophagy. **(f)** Live-cell imaging after a 1 h OA pulse shows early recruitment of ATG2A to nascent LDs, where it colocalizes with the early LD marker LiveDrop (hpGPAT4, blue arrow). At this stage, unlike mature LDs, nascent LDs are poorly labeled by LipidTOX. Representative of 3 independent experiments. **(g)** After a 4 h OA pulse, BPY-C12 and LipidTOX signals become increasingly overlapped, reflecting the maturation of nascent LDs. Representative of 3 independent experiments. **(h)** TLC of purified LD fractions from DGAT2-inhibited WT cells ± ATG2A overexpression, revealed under UV. TopFluor-DAG (Ref) indicates DAG migration. DAG enrichment is observed upon ATG2A overexpression. Representative of two independent experiments. Scale bars for all images: 10 µm; insets: 1 µm.

Source data

During the pulse, we observed bright BODIPY-positive particles in DKO cells with weak or no LipidTOX staining, probably nascent LDs (Fig. 1i and Supplementary Video 1). Indeed, LipidTOX labels hydrophobic regions and is less effective for smaller LDs with fewer lipids, which were labeled by LiveDrop and ATG2A (Extended Data Fig. 2f). In addition, extending the pulse to 4 h showed that the BPY-C12 and LipidTOX signals nearly fully overlapped as the nascent LDs matured (Extended Data Fig. 2g). After the 1-h pulse, fewer nucleated LDs appeared in ATG2A-rescued DKO cells than in WT (Fig. 1i,j). Thus, WT cells efficiently package TAGs into a limited number of ATG2A-bound LDs, while in DKO cells, slower lipid delivery to existing LDs results in the formation of more LDs to store the same amount of TAG.

To further confirm the lipid transport of ATG2A to LD for growth, we transfected DKO cells with a transport-deficient ATG2A mutant (TD-ATG2A) containing ten mutations of vital hydrophobic to charged residues within the groove for lipid transport³⁹. TD-ATG2A localized to LDs (Fig. 1k). However, lipid incorporation following the pulse was similar in DKO and transfected cells, indicating that TD-ATG2A is defective in delivering lipids to LDs (Fig. 1l). Similarly, cells with TD-ATG2A, like DKO cells, formed many new LDs after the pulse (Fig. 1m).

While the groove of ATG2A can accommodate and transport various glycerophospholipids, these molecules localize to the LD surface. In our experiments, we observed the BPY-C12 signal within the LD core (Fig. 1e). Therefore, we hypothesize that ATG2A transports lipids that can be accommodated in the LD oil core. Following the supply of fatty acids, TAG and DAG are the lipids that can reside in the LD core⁶.

ATG2A mediates DAG delivery to the LD core

Fatty acids are converted to PA, then to DAG at the ER and LD surfaces (Fig. 2a). DAG is transformed to TAG by DGATs. If ATG2A only transports TAGs to LDs, inhibiting DGATs should substantially diminish the delivery of fluorescent lipids inside LDs. First, to confirm the efficacy of the inhibitors, cells were delipidated for 24 h, incubated with or without DGAT inhibitors for 5 h and then loaded with OA for several hours. Inhibitor-treated cells exhibited a notable reduction in LD formation compared with controls, confirming the effectiveness of the protocol (Extended Data Fig. 3a). Next, we treated cells with DGAT1/2 inhibitors before the 1-h pulse and examined BPY–lipid incorporation in both DKO and ATG2A-rescued cells (Fig. 2b). BPY–lipid incorporation into the LD core was readily detected (Fig. 2c). Interestingly, DGAT inhibition did not significantly affect BPY incorporation even in rescued cells (Fig. 2d and Extended Data Fig. 3b). These results indicate ATG2A can transfer lipids into the LD core independently of DGAT activity. This leaves DAG as the primary lipid transferred by ATG2 to LDs, although it may also contribute to the transport of TAG.

Extended Data Fig. 3 — **(a)** Validation of DGAT1 and DGAT2 inhibitor (DGATi) efficacy. HeLa WT cells were lipid-depleted and subsequently incubated with DGATi for 5 h. OA was then added to induce LD formation. LDs were stained with LipidTOX. **(b)** Confocal images of ATG2A-rescued DKO cells after a 1 h BPY-C12 pulse, with or without DGATi. These images support the quantification of BPY-C12 incorporation into LDs shown in Fig. 2d. **(c-d)** Additional data complementing Fig. 3a-b. WT cells expressing endogenous levels of ATG2 show partial incorporation of TMR-PA into LDs after a 1 h pulse, with signal intensity higher than in DKO but lower than in ATG2A-overexpressing cells. **(d)** Quantification of TMR-PA enrichment in LDs relative to the cytosol in WT, DKO, and ATG2A-rescued cells. SuperPlots are shown, with small dots representing individual cells (n = 30 per condition) and large color-coded dots indicating replicate means from 3 independent biological experiments. Data are presented as mean ± SD. Statistical analysis was performed using two-tailed Mann-Whitney test; **p = 0.0021, ****p < 0.0001. **(e)** Additional confocal images of ATG2A-rescued cells after a 1 h TMR-PA pulse under Lipin-1 inhibition, supporting Fig. 3c. Cells were transfected with ATG2A–eGFP for 24 h, treated with 200 µM OA for 24 h to induce LD maturation, then incubated with propranolol for 5 h prior to the 1 h TMR-PA pulse. Both non-transfected (1) and transfected (2) cells show absence of TMR-PA incorporation into the LD core. **(f)** Confocal images of ATG2A-rescued DKO cells showing the rare phenotype of TMR-PA accumulation at the LD surface, but not in the core, after Lipin-1 inhibition. Images are representative of 3 independent experiments. **(g)** Number of LDs per cell in ATG2A-rescued DKO cells, categorized by TMR-PA localization phenotype: droplets with TMR-PA detected at the surface (red dots) and droplets with no detectable TMR signal (green dots). SuperPlots are shown, with small dots representing individual cells (n = 26 per condition) and large color-coded dots indicating replicate means from 3 independent biological experiments. Data are presented as mean ± SD. Two-tailed Mann-Whitney test was performed; ****p < 0.0001. Scale bars: 10 µm; insets: 1 µm.

Source data

Our results suggest that ATG2A transports and accumulates DAG in LDs, creating a sink for TAG biosynthesis in LDs and decreasing ER DAG availability for other uses such as phospholipid synthesis via the Kennedy pathway. To test the impact of ATG2A on DAG levels, we induced LD formation in WT and DKO cells for 24 h to establish an LD pool for ATG2A binding. We then performed a 1-h OA pulse and used mass spectrometry to measure cellular DAG versus TAG levels, knowing that most DAGs, nonmembrane lipids, reside in LDs. We observed a higher DAG/TAG ratio in WT cells overexpressing ATG2A compared with DKO cells (Fig. 2e). This difference was more pronounced when DGATs were inhibited before the OA pulse, resulting in greater DAG accumulation. DGAT inhibition had no significant effect on DAG levels in DKO cells. We interpret these findings as follows: in WT cells, ATG2A retrieves DAG for LDs, protecting it from other pathways in the ER. This increases the DAG/TAG ratio, especially with DGAT inhibition. Without ATG2A, DAG stays longer in the ER and converts into other lipids, lowering the DAG/TAG ratio. Overall, these data support the role of ATG2A in the transport of DAG to LDs, influencing ER DAG metabolism.

To directly visualize the incorporation of DAGs into LDs, we preincubated DKO cells, either nonrescued or rescued with ATG2A, with a DGAT inhibitor for 5 h, followed by incubation with TopFluor–DAG. We observed robust incorporation of TopFluor into the LD core, as confirmed by thin-layer chromatography (TLC) (Fig. 2f and Extended Data Fig. 2h). Notably, incorporation was significantly reduced in DKO cells compared with ATG2A-rescued cells (Fig. 2g). These findings provide compelling evidence that ATG2A binding to LDs facilitates the incorporation of DAG into their core.

PA is metabolized and stored in LD via an ATG2A-dependent pathway

PA is metabolized to DAG, which is then converted into phospholipids or TAG (Fig. 2a). To further test our hypothesis, we investigated the fate of PA. We repeated the pulse experiment by supplementing cells with an acyl chain Tamra-labeled PA (TMR-PA) instead of BPY-OA (Fig. 3a). In both DKO and rescued cells, a clear TMR signal was observed within the LD core, indicating that PA was metabolized into neutral lipids and stored within LDs. Consistent with our model of the role of ATG2A in bulk lipid transfer to LDs, TMR incorporation was higher in ATG2A-rescued DKO cells compared with DKO cells (Fig. 3b).

Fig. 3 — a,b, ATG2A enables PA incorporation into the LD core: confocal images of HeLa ATG2 DKO cells transfected or not with ATG2A–eGFP after 24 h OA-induced LD formation and a 1-h pulse with TMR-PA (a) and quantification of TMR-PA enrichment in LDs relative to cytosol in DKO and ATG2A-rescued cells (b). SuperPlots are shown, where small dots represent individual cells (n = 30 per condition) and large color-coded dots represent replicate means from three independent biological experiments. Data are presented as mean ± s.d. Statistical analysis was performed using a two-tailed Mann–Whitney test; ****P < 0.0001. c,d, Lipin inhibition prevents PA accumulation in LDs: confocal images of ATG2A-rescued cells treated with OA for 24 h, followed by Lipin-1 inhibition using propranolol for 5 h, then 1 h pulse with TMR-PA (c). The same focal plane includes a nontransfected control (1) and an ATG2A-rescued cell (2). Quantification of TMR-PA signal in LDs in conditions ± ATG2A under Lipin-1 inhibition (d). SuperPlots are shown, where small dots represent individual cells (n = 23 for DKO and n = 16 for ATG2A rescued) and large color-coded dots represent replicate means from three independent biological experiments. Data are presented as mean ± s.d. Statistical analysis was performed using a two-tailed unpaired Student’s t-test. e,f, In the absence of ATG2, PA accumulates at non-LD membranes: after a 1 h TMR-PA pulse, cells were exposed to hypotonic swelling by incubation in diluted DMEM (1:20 in water) for 15 min at 37 °C (e). This treatment causes spatial separation of organelles, which adopt a vesicular structure, including the ER marked with EROX–BFP, allowing clear visualization of TMR-PA localization. Line scans were performed to measure TMR-PA signal at the surface of ER and non-ER vesicles. The partition coefficient (P) was defined as the ratio of intensity on vesicle membranes (I_VES) to that on lipid droplets (I_LD). Representative images show a TMR-positive non-ER vesicle (red arrow) and an ER membrane (blue arrow). Quantification of the PA partition coefficient in conditions ± ATG2A (f). SuperPlots are shown, where small dots represent vesicle/LD values (n = 80 vesicle/LD per condition from ~5 cells) and large color-coded dots represent replicate means from three independent biological experiments. Data are presented as mean ± s.d. Statistical analysis was performed using a two-tailed unpaired Student’s t-test; ****P < 0.0001. Scale bars for all images, 10 µm; insets, 1 µm.

Source data

We repeated the experiment with the inhibition of Lipin-1 by propranolol⁴⁸, which should prevent the conversion of PA to DAG. Cells were pretreated with propranolol before the TMR-PA pulse (Fig. 3c). The primary phenotype was the near absence of signal in the LD core (Fig. 3d and Extended Data Fig. 3e), which aligns with our model of DAG and/or TAG transfer to LDs by ATG2A. In minor instances, TMR was shown as a surface signal on specific ATG2A-positive LDs (Extended Data Fig. 3f,g), suggesting that ATG2A may directly transport PA or derived phospholipids to LDs. These data suggest that PA is converted to DAG in the ER and then delivered to the LD for TAG conversion by DGAT2. Alternatively, PA could be directly transported from the ER to LDs via ATG2, and then converted to DAG and TAG by Lipin-1 and DGAT2. In either case, ATG2A appears to coordinate with lipogenic enzymes in LD growth.

Without ATG2A, PA may follow different fates because it cannot integrate LDs. Using cell swelling^49–51, we generated large organelle vesicles to better visualize and measure PA localization by fluorescence. DKO and ATG2A-rescued DKO cells were transfected with EROX–BFP, labeling the ER lumen. After the PA pulse, cell swelling was induced with hypotonic medium (Fig. 3e). TMR signals indicated significantly higher fluorescent lipid levels on non-LD structures, such as the ER, in DKO versus rescued cells (Fig. 3f and Extended Data Fig. 4a). This suggests that impaired ER-to-LD transport of TMR-PA-derived lipids in DKO cells causes the lipid to be metabolized into phospholipids, which then redistribute to the ER and other organelles. Combined with the DGAT inhibition data (Fig. 2b–d), these findings support that ATG2A transports DAG (and possibly TAG) from the ER to LDs during lipogenesis.

Extended Data Fig. 4 — **(a)** Same experiment and images as in Fig. 3e (1 h TMR-PA pulse followed by hypotonic swelling), with enhanced brightness/contrast on the TMR-PA channel to better visualize PA enrichment in lipid droplets. This adjustment highlights the strong PA accumulation in LDs upon ATG2A rescue, contrasting with the more diffuse PA signal on membrane vesicles observed in ATG2 DKO cells. **(b)** Immuno-electron microscopy of HeLa WT cells overexpressing ATG2A–eGFP, labeled with anti-GFP (15 nm gold) and anti-PDI (10 nm gold) to mark ATG2A and ER, respectively. ATG2A localizes at ER-LD contact sites, shown by colocalization of gold particles at membrane interfaces (red arrows). Images are representative of 3 independent experiments.

ATG2A is at ER–LD contact sites and can transport DAG, TAG and PA in vitro

To test whether ATG2A localizes to regions where the ER and LDs are in close proximity, a prerequisite for our model, we performed immuno-electron microscopy on ultrathin cryosections of WT cells overexpressing ATG2A–eGFP (Fig. 4a). Dual immunogold labeling was used to detect ATG2A–eGFP (15 nm gold) and the ER marker protein disulfide isomerase (PDI; 10 nm gold), allowing high-resolution visualization of ER–LD interfaces. We found that ATG2A was indeed localized at sites where the ER closely apposed the surface of LDs, as indicated by the colocalization of 15 nm and 10 nm gold particles (Fig. 4a and Extended Data Fig. 4b). This spatial organization suggests that ATG2A may help bridge or stabilize these membrane interfaces. Altogether, these ultrastructural observations support a role for ATG2A in lipid exchange between the ER and LDs, which may contribute to the biogenesis and maintenance of LDs.

Fig. 4 — a, ATG2A localizes at ER–LD contact sites. Immuno-electron microscopy of ultrathin cryosections from HeLa WT cells overexpressing ATG2A–eGFP, labeled with anti-GFP (15 nm gold) and anti-PDI (10 nm gold) to mark ATG2A and ER, respectively. ATG2A is detected at ER–LD interfaces, indicating its enrichment at membrane contact sites. The representative cropped region shows colocalization of 15 nm and 10 nm gold particles at ER–LD junctions (red arrows). Images are representative of three independent experiments. b, A schematic of the FRET dequenching assay performed to detect lipid transport in a reconstituted system. (Donor) liposomes harboring Rh-PE head-labeled and NBD-PS tail-labeled lipids or harboring Rh-PE head-labeled and NBD-DAG tail-labeled lipids are mixed with aLDs (acceptor) in the presence of recombinant ATG2A protein. In the donor liposomes, rhodamine quenches the NBD. Increasing fluorescence in the assay indicates dilution of the fluorophores into the acceptor compartment and thus is evidence of lipid transport. c, Quantification of the NBD signal observed under different experimental conditions as a percentage of maximum dequenching scaled from basal fluorescence of membranes before addition of ATG2A to maximum fluorescence measured after dissolution of membranes by Triton X-100 at the end of the assay. Data are presented as mean ± s.e.m. from three independent membrane-protein preparations, each performed in technical duplicates. d, DAG transfer assay. NBD-labeled DAG was incorporated into donor liposomes and its transfer to aLDs was monitored by fluorescence. Owing to the higher quantum yield of NBD in the hydrophobic aLD core compared with membranes, DAG transfer results in a fluorescence increase. e, ATG2A promotes DAG transfer. Upon addition of ATG2A, NBD-DAG fluorescence increased over time, indicating efficient DAG transfer from donor membranes to aLDs. Triton X-100 (TX-100) was added at the end to normalize NBD fluorescence. Data are presented as mean ± s.e.m. from three independent membrane-protein preparations, each performed in technical duplicates. f, Unbiased coarse-grain molecular dynamics (CG-MD) simulations illustrate the spontaneous binding of different lipids such as DOPC, DOPA, DAG or TAG initially placed in bulk solvent, to the lipid-binding pocket of ATG2A. The protein is shown in purple, the lipid tails are in yellow and the groups are in black. The number of lipids incorporated into the tunnel is indicated in the figure. g, Lipid tail (left) and headgroup solvation (right) for each lipid in the last 500 ns of the CG-MD trajectories. Each lipid-bound system was simulated for 500 ns, and the analysis was performed for five independent CG-MD replicas per lipid. Data are presented as mean ± s.d. across replicas. The box plots indicate the distribution across all lipids and time frames: center line, median; box limits, 25th and 75th percentiles; whiskers, 1.5× interquartile range; points, outliers.

Next, we tested the ability of ATG2A to transport DAG from the ER to LDs in vitro using a Förster resonance energy transfer (FRET) dequenching assay with liposomes and artificial LDs (aLDs) (Fig. 4b). Donor liposomes had 66% dioleoylphosphatidylcholine (DOPC), 30% dioleoylphosphatidylethanolamine (DOPE), 2% Rh-phosphatidylethanolamine (PE) and 2% NBD-DAG or NBD-phosphatidylserine (PS), allowing comparison of DAG with a phospholipid. aLDs from triolein and DOPC served as acceptors. As lipids transfer from donor to acceptor, the quenching of NBD by Rh-PE decreases, resulting in increased fluorescence. Experiments were conducted with and without ATG2A, and the data obtained with ATG2A were subtracted from the data obtained without it. Both NBD-PS and NBD-DAG were transported by ATG2A (Fig. 4c). However, NBD-DAG shows more dequenching than NBD-PS, probably because DAG can move across the bilayer (Fig. 4c), allowing inner-leaflet DAG to participate in lipid transport as effectively as outer-leaflet DAG. Conversely, NBD-PS on the inner leaflet may become trapped, limiting its transport. Furthermore, DAG can penetrate the aLD core, which is an infinite receptacle (Fig. 2b). In contrast, phospholipids stay at the aLD interface, which is finite and thus limits the level of transport. Last, the relationship between fluorescence dequenching and fluorophore dilution is nonlinear⁵². Our data suggest that the fluorescence maximum rises from ~8% (NBD-PS) to ~30% (NBD-DAG), representing a nearly tenfold increase in dilution. Supporting this idea, adding more acceptor aLDs led to greater dequenching of NBD-DAG compared with NBD-PS, suggesting that the additional diffusion area (or volume in this case) provided a greater capacity for labeled DAG than for labeled PS (Fig. 4c). To further test the capacity of ATG2A to transport lipids to LDs, we conducted a similar lipid transport assay, which indicated that ATG2A may transport TAG as efficiently as DAG (Extended Data Fig. 5e) but not cholesterol esters (Extended Data Fig. 5f). We found that PA was also transferred (Extended Data Fig. 5g–i).

Extended Data Fig. 5 — **(a-d)** In vitro assay of ATG2A-mediated DAG transfer to artificial lipid droplets. **(a)** Schematic of the TopFluor-DAG transfer assay between liposomes and artificial lipid droplets (aLDs), with recombinant full-length ATG2A bridging the two via its hydrophobic tunnel. **(b)** Confocal images of aLDs and TopFluor-DAG liposomes after 3 h incubation, without or with ATG2A. LipidTOX stains aLDs. Images are representative of 3 independent experiments, each performed with a distinct batch of mixed lipids. Scale bar: 50 µm; insets: 20 µm. **(c)** Quantification of TopFluor-DAG incorporation into aLDs (n = 75 per condition), shown as individual droplets (small dots) and experiment means (large dots, 3 independent experiments). Two-tailed Mann-Whitney test was used for statistical analysis; ****p < 0.0001. **(d)** Kinetics of TopFluor-DAG incorporation over time, averaged from 3 independent experiments. Data are shown as mean ± SD. Two-tailed unpaired Students’ t-test was used; **p = 0.0035. **(e)** Depiction of DAG and TAG transfer from donor liposomes (66% DOPC, 30% DOPE, 2% fluorescent lipids) to artificial lipid droplets (aLDs) composed of 70% DOPC, 30% DOPE, and triolein. Control reactions without ATG2A show baseline spontaneous lipid mixing. **(f)** FRET dequenching lipid transport assay comparing transfer of head-labeled lipids (Rh-PE, NBD-PS) versus tail-labeled neutral lipids (Rh-TAG, NBD-12-CE) between donor (61% DOPC, 30% DOPE, 5% DGS-NTA(Ni), 2% fluorescent lipids) and acceptor liposomes (65% DOPC, 30% DOPE, 5% PI(4,5)P2). Controls without ATG2A reveal spontaneous membrane fusion levels. Both assays (e-f) show fluorescent lipid transport, though they do not clarify which lipid species are preferentially transferred. **(g-i)** Phosphatidic acid transfer. **(g)** Tail-labeled NBD-PA and TMR-PA were incorporated into liposomes at 2% each such that NBD fluorescence was quenched by TMR. The remaining liposome composition was 0.5% pyrene-PC, 30% DOPE, and 65.5% DOPC. Acceptor artificial lipid droplets were constructed from 99.5% DOPC and 0.5% pyrene-PC. **(h)** PA transfer reactions were performed using 25 µM donor and acceptor phospholipids, with the acceptor membrane surface area approximately double that of the donor. ATG2A (100 nM) or control buffer was added at time zero, and transfer was monitored over 1-5 minutes at 30 °C. **(i)** Initial lipid transfer rate was determined by fitting an exponential plateau curve to the data and calculating the maximum slope. Data represent 3 combinations of independent membrane and protein preparations. Individual biological replicates were performed in technical duplicate; data are shown as mean ± SEM. Two-tailed ratio paired t-test was applied; *p < 0.05.

Source data

To confirm DAG transport, we used two complementary fluorescence assays. First, we incubated TopFluor–DAG liposomes with micrometric aLDs to assess DAG incorporation (Extended Data Fig. 5a). The liposomes often docked to the aLDs owing to their large size. Still, consistent with the dequenching assay, ATG2A significantly enhanced TopFluor–DAG delivery to the core of aLDs (Extended Data Fig. 5b–d). Second, we used liposomes containing 2% NBD-stearoyl-arachidonoyl-DG. We monitored the increase in NBD fluorescence as DAG transferred to the hydrophobic core of nanometric aLDs, wherein the quantum yield is higher⁵³ (Fig. 4d). Donor and acceptor membrane areas were adjusted to present roughly twice the membrane surface area on the acceptors, and reactions were triggered by adding or not adding ATG2A. Triton X-100 was then added to reveal total NBD fluorescence. As before, ATG2A strikingly increased the NBD signal, confirming the transfer of DAG to the aLD core (Fig. 4e).

To investigate whether the cavity of ATG2A can accommodate lipids beyond glycerophospholipids, we resorted to unbiased coarse-grained simulations of ATG2A in water⁵⁴, in the presence of DAG or TAG, and compared the results to the data on ATG2A with DOPC lipids and dioleoylphosphatidic acid (DOPA) (Fig. 4f,g). Our simulations revealed that ATG2A can bind DAG, TAG and PA, and accommodate a large number of lipids within its cavity (Fig. 4f). The smaller size and tighter packing of the DAG molecules are consistent with the larger number of bound lipids observed in our simulations. The orientation of the DAG and TAG within the cavity mirrors that of phospholipids, with the lipid tails interacting with the exposed hydrophobic residues of the protein’s beta sheets and the polar head groups oriented toward the cavity opening. Notably, both DAG and TAG engage in fewer interactions with water molecules in comparison with DOPC (Fig. 4g), and DAG exhibits a higher propensity for cavity incorporation than TAG (Fig. 4f,g). These data support the ability of ATG2A to bind and potentially transport DAG, TAG and PA.

DGAT2 localization to LDs requires ATG2 and DAG

GPAT4, ACSL3, Lipin-1 and DGAT2, involved in TAG biosynthesis, bind to LDs²⁴. Since ATG2A may regulate substrate flow from the ER to LDs and promote local TAG synthesis and LD growth, we investigated whether its absence affects these enzymes.

We expressed GFP-tagged versions of the enzymes in WT and DKO cells, loaded with OA for 24 h. No differences in LD recruitment were seen for GPAT4, ACSL3 and Lipin-1 (Fig. 5a). However, DGAT2 was often absent from LDs in DKO cells (Fig. 5a,b). Overexpression of DGAT2 did not increase LD size (Extended Data Fig. 6a,b). These findings suggest ATG2A transports DAG to LDs in synchrony with DGAT2 moving from ER to LDs. Although ATG2A may also transport PA, the localization of Lipin-1 LD was unaltered by the absence of ATG2A, indicating a critical role of ATG2A in DAG transport.

Fig. 5 — a,b, DGAT2 is not recruited to LDs in the absence of ATG2: confocal images of HeLa WT and ATG2 DKO cells expressing the indicated enzymes of the TAG synthesis pathway, after 24 h transfection and 24 h OA treatment (a). Proteins are shown in green; lipid droplets are stained with LipidTOX. Scale bars, 10 µm; insets, 1 µm. Quantification of enzyme recruitment to LDs (b). Data are shown as SuperPlots displaying individual LDs (small dots) and the mean per independent experiment (large color-coded dots). Approximately n = 1,600 LDs per condition were analyzed from three independent biological replicates. Data are presented as mean ± s.d. Statistical analysis was performed using a two-tailed Mann–Whitney test; ****P < 0.0001. c–f, DAG is required for DGAT2 targeting to LDs: a schematic of GERV generation: WT cells were transfected with DGAT2–GFP and the ER lumen marker EROX–BFP, then exposed to hypotonic shock to generate GERVs (c). A schematic of DEGERV generation: GERVs were incubated with droplets composed of TAG or TAG:DAG to generate DEGERVs. DGAT2 redistribution to aLDs depends on their DAG content (d). Confocal images of DEGERVs from cells expressing DGAT2–GFP. DGAT2 signal is low on pure TAG aLDs but enriched on DAG-containing aLDs (e). Yellow arrows indicate aLDs. Quantification of DGAT2 intensity on aLDs (f). Each symbol (circle, square or triangle) represents the mean of one independent experiment; small uncolored dots show individual values. Approximately n = 20 cells were analyzed per condition. Data are presented as mean ± s.d. Statistical analysis was performed using a two-tailed Mann–Whitney test; ****P < 0.0001.

Source data

Extended Data Fig. 6 — a) Confocal images showing the absence of DGAT2 recruitment in HeLa WT and ATG2 DKO cells. Cells were transfected with DGAT2–GFP for 24 h and treated with OA for 24 h. All channels are displayed to complement Fig. 5a. LDs were stained with LipidTOX. **(b)** Quantification of LDs sizes in WT and DKO cells overexpressing DGAT2 after 24 h OA induction. Data are shown as SuperPlots, where small dots represent individual cells (n = 17 cells per condition) and large color-coded dots represent replicate means from 3 independent experiments. Data are presented as mean ± SD. Two-tailed unpaired Student’s t-test was used; ****p < 0.0001. **(c)** Additional examples complementing Fig. 5e. WT cells were transfected with DGAT2–GFP and the ER lumen marker EROX–BFP, then subjected to hypotonic shock to generate GERVs. These GERVs were incubated with droplets composed of either TAG alone or TAG mixed with DAG to produce DEGERVs. DGAT2 redistribution to the surface of aLDs depends on their DAG content, as indicated with yellow arrows. Scale bars: 10 µm; insets: 1 µm, unless otherwise indicated in the figure.

Source data

The neutral lipid and phospholipid composition of LDs influences protein binding^55,56. We investigated whether DAG affects DGAT2 movement to LDs using droplet-embedded giant ER vesicles (DEGERVs). DEGERVs embed aLDs within giant ER vesicles (GERVs)⁴⁹, offering a controlled setting to study ER protein transfer to LDs (Fig. 5c,d). To prepare DEGERVs, cells were transfected with a fluorescent ER marker and DGAT2–GFP. After 24 h, they underwent hypotonic swelling and lysis to produce GERVs^49,57 (Fig. 5c). These GERVs were mixed with aLDs, which fused with the GERV membrane^57–59, forming DEGERVs (Fig. 5d).

When aLDs composed solely of TAG were used, DGAT2 barely moved from the GERV to the aLD (Fig. 5e,f). In contrast, other ER proteins, including ACSL3 and Far1, relocated efficiently to aLDs⁵⁷. This suggests that the relocation of DGAT2 from the ER to LDs is inefficient with TAG. We repeated the experiment with a 9/1 TAG/DAG mixture and found a strong relocation and enrichment of DGAT2 to the aLDs (Fig. 5e,f and Extended Data Fig. 6c). This result suggests that DAG-containing LDs help move DGAT2 from the ER to LDs. ATG2A may transport DAG, causing DGAT2 to relocate and convert DAG into TAG on LDs, promoting LD growth.

Moderate ATG2A recruitment to LDs is key for DGAT2 relocalization and LD growth

If ATG2A-mediated DAG transport relocates DGAT2 to LD, we thought that ATG2A overexpression would enhance DGAT2 LD targeting. DGAT2–GFP efficiently relocalized to LDs in WT cells and not in DKO as previously found (Fig. 5a). Surprisingly, however, when both ATG2A and DGAT2 were co-overexpressed, we observed multiple phenotypes wherein DGAT2–GFP either remained in the ER or relocated to LDs (Fig. 6a). After careful analysis, we found that DGAT2–GFP relocation to LDs depended on ATG2A–mCherry expression level, which we varied (Fig. 6a,b and Extended Data Fig. 7a): low ATG2A expression allowed DGAT2 to efficiently localize to LDs in DKO, similar to WT cells; high ATG2A prevented DGAT2 relocation, possibly due to protein competition for LD binding^57,60. This suggests that ATG2A LD levels must be tightly controlled to avoid displacing proteins such as DGAT2, which has a low affinity for pure TAG LDs⁵⁷. For comparison, Lipin-1γ was not displaced by ATG2A (Extended Data Fig. 7b). Consistent with our interpretation, ATG2A has a slow off-rate from LDs (Extended Data Fig. 7c,d) and displaces even the high-affinity LD protein Plin1 (Extended Data Fig. 7e,f)^57,61.

Fig. 6 — a, Confocal images of HeLa ATG2 DKO cells cotransfected with DGAT2–GFP and ATG2A–mCherry and treated with OA for 24 h. LDs were stained with LipidTOX. Low ATG2A levels enable DGAT2 recruitment to LDs. b, Quantification of DGAT2–GFP recruitment to LDs in WT, DKO and DKO cells co-expressing low or high ATG2A–mCherry. Low ATG2A promotes DGAT2 accumulation, whereas high ATG2A reduces LD targeting. SuperPlots show individual LDs (small dots; n = 160 for DKO and WT, n = 270 for low ATG2A and n = 184 for high ATG2A; ~15 cells per condition) and mean per experiment (large color-coded dots, from three independent experiments). Data are mean ± s.d.; two-tailed Mann–Whitney test was used; ****P < 0.0001. c, Correlation of DGAT2 and ATG2A signal intensities on LDs in DKO cells cotransfected with either WT or lipid transfer-deficient ATG2A (TD-ATG2A). DGAT2 recruitment correlates with WT ATG2A levels but not with TD-ATG2A, indicating dependence on ATG2A lipid transfer activity. d, Confocal images of WT cells, nontransfected or overexpressing ATG2A–eGFP, treated with DGAT1 or DGAT2 inhibitors (inh). LDs were stained with BODIPY FL or LipidTOX and nuclei with Hoechst. e, Left: SuperPlots of mean LD size per cell (n = 40 cells per condition) with the mean per experiment indicated by large color-coded dots. Right: LD size distribution across all cells. ATG2A overexpression enhances LD growth in DGAT1-inhibited cells, while DGAT2 inhibition shifts the LD population toward smaller droplets. Data are mean ± s.d.; two-tailed unpaired Student’s t-test was used; **P = 0.0050 and ****P < 0.0001. f, Normalized density plots of LD size in ATG2A-overexpressing cells under control, DGAT1i or DGAT2i conditions, confirming the shift toward smaller LDs upon DGAT2 inhibition. Scale bars, 10 µm; insets, 1 µm.

Source data

Extended Data Fig. 7 — **(a)** Confocal images of HeLa WT cells showing DGAT2–GFP recruitment to LDs with or without co-expression of ATG2A–mCherry. Cells were transfected for 24 h, treated with OA for 24 h, and stained with LipidTOX. Images are representative of 3 independent experiments. **(b)** Confocal images of WT cells co-expressing Lipin-1γ–GFP and ATG2A–mCherry, showing that Lipin-1 localizes to LDs independently of ATG2A expression. Images are representative of 3 independent experiments. **(c)** FRAP analysis of ATG2A mobility. Cells expressing ATG2A–eGFP were treated with OA and stained with LipidTOX. GFP fluorescence was photobleached, and images were acquired every 30 seconds. Scale bars: 5 µm; insets: 1 µm. Images are representative of 3 independent experiments. **(d)** Line plot showing normalized fluorescence intensity in the photobleached region of ATG2A–eGFP expressing cells. The grey curve represents the normalized fluorescence intensity of eGFP-positive LDs that were not bleached. Data were collected from 3 independent experiments and presented as mean ± SEM. **(e)** Confocal microscopy images of WT cells demonstrating competition between ATG2 and Plin1 for LD recruitment. Cells were cotransfected with ATG2–eGFP and PLIN1–mCherry for 24 h, then treated with OA for 24 h to induce LD formation. LDs were stained with LipidTOX. Images are representative of 3 independent experiments. **(f)** Line profile shows the intensity levels of ATG2A–eGFP and PLIN1–mCherry on droplets depicted in the inset. **(g)** Representative confocal images supporting the correlation analysis shown in Fig. 6c. Examples of cells with high and low levels of TD-ATG2A, both lacking DGAT2 enrichment on LDs are shown. Images are representative of 3 independent experiments. **(h)** The red curve shows a non-linear Gaussian fit indicating no DGAT2 recruitment in cells expressing the lipid transfer-deficient TD-ATG2A mutant, regardless of expression level. **(i)** Confocal images of DKO cells illustrating LD size under control conditions and following inhibition of DGAT1 or DGAT2. LDs are stained with BODIPY FL and nuclei with Hoechst. **(j)** Quantification of LDs size in cells shown in (i), in comparison to the non-transfected WT condition shown in Fig. 6d. Data are displayed as Superplots showing the mean size of LDs per cell (n = 132 (WT no inh); n = 193 (WT +DGAT1i); n = 137 (WT +DGAT2i); n = 146 (DKO no inh); n = 157 (DKO +DGAT1i); n = 183 (DKO +DGAT2i)) and the means per independent experiments (large dots, 3 independent experiments). Data are presented as mean ± SD. Two-tailed Mann-Whitney test was used; *p = 0.0153, **p = 0.0025, ****p < 0.0001. **(k)** Non-normalized density plots of LD size distribution in WT ATG2A-overexpressing cells under control (no inh), DGAT1i and DGAT2i conditions. DGAT2 inhibition resulted in a higher density of small LDs compared to other conditions, confirming the shift observed in (Fig. 6e, right panel). Scale bars: 10 µm; insets: 1 µm, unless otherwise indicated in the figure.

Source data

To validate that the DGAT2 relocation required the ATG2A lipid transport activity, we repeated the experiments with the TD-ATG2A. In this case, we observed no relocation of DGAT2 (Fig. 6c and Extended Data Fig. 7g,h). These data suggest that a low ATG2A recruitment level to LDs is both necessary and sufficient for DAG to be delivered to LDs, facilitating subsequent DGAT2 relocation. These data also indicate that ATG2A does not directly recruit DGAT2, but rather via its DAG transport activity.

Finally, we examined the synergy between ATG2A and DGAT2 in lipid storage by looking at how DGAT1/2 activity affects LD size. DKO cells had smaller LDs than WT cells (Extended Data Fig. 7i,j) due to increased LD nucleation, which causes lipids to spread into more numerous, smaller LDs, and the absence of DGAT2 on LDs, which limits their growth. Inhibiting either DGAT1 or DGAT2 had only a slight impact on LD size. The way lipids are directed for both nucleation and growth makes it challenging to determine the exact role of DGATs in LD expansion.

Therefore, WT cells, which do not exhibit aberrant nucleation, provide an optimal context for investigating how ATG2A levels influence LD size with DGAT1 and DGAT2. We repeated the ATG2A–eGFP expression experiments in WT cells. As previously observed, ATG2A overexpression led to larger LDs (Fig. 6d–f), probably due to its efficiency in channeling DAG and TAG into LDs.

Inhibition of DGAT1 in ATG2A-overexpressing cells significantly shifted the LD size distribution toward larger LDs (Fig. 6e,f). In this condition, only DGAT2 remains active. Coupled with DAG transport mediated by ATG2A, this results in enhanced LD growth and an increased number of larger LDs. This is probably because the DAG normally used by DGAT1 becomes available for transfer to LDs, where it is utilized by DGAT2 to promote TAG synthesis. These findings highlight a functional synergy between ATG2A and DGAT2 in driving LD expansion.

Conversely, DGAT2 inhibition did not significantly alter the average LD size compared with the control condition, despite impaired TAG synthesis at the LD surface. This could be due to TAG being transported to LDs by ATG2A, although the LDs appeared more enriched in DAG (Extended Data Fig. 2h). Notably, the LD size distribution shifted toward smaller LDs (Fig. 6e,f and Extended Data Fig. 7k). In this context, DAG is more efficiently converted into TAG by DGAT1 in the ER, leading to the formation of numerous smaller LDs. ATG2A may also contribute by transferring residual ER-derived TAG to LDs, partially compensating for the loss of DGAT2 activity at the LD surface. As a result, the average LD size remains similar to that in the absence of inhibition. Together, these results suggest that ATG2A and DGAT2 cooperate to promote efficient lipid storage and LD expansion.

Discussion

Thermodynamic and biophysical principles govern LD formation. Elevated TAG levels in the ER can spontaneously nucleate nascent LDs, which are controlled by seipin and accelerated by DAG^16,18,20,62. Subsequently, LDs bud into the cytosol, maintaining a single point of physical contiguity with the ER, which is gated by seipin^63,64. This contact point enables proteins to traffic between the two organelles^63,65,66. For instance, DGAT2 and GPAT4 reside in both compartments, but only the movement of GPAT4 to LDs is impacted by seipin²⁴.

TAG lipogenesis enzymes are found in both the ER and LDs, indicating that lipid precursors such as DAG migrate to LDs for TAG synthesis within them. A key question is whether the seipin complex is the only entry point for ER-derived lipids to enter LDs. During TAG biosynthesis, lipid intermediates such as DAG and PA are produced, and the ER probably quickly removes these potentially toxic intermediates, with LDs serving as detoxifying compartments. If lipids, such as TAG, DAG and PA, access LDs only through seipin, it could cause a bottleneck, leading to ER accumulation and possible membrane damage or stress^67–70. Although LTPs may mediate ER-to-LD lipid transport, their specific roles and cargo are not well characterized. These proteins may help form or localize to ER–LD contact sites that facilitate the transfer of lipids such as DAG, TAG and PA.

Here, we demonstrate that ATG2A facilitates lipid incorporation into LDs and growth by transferring DAG from the ER to LDs. Only a few, and potentially dynamic, ATG2A bridges are formed between the ER and LD, facilitating DAG channeling to LDs. These bridges are distinct from seipin at the ER–LD connection. The ATG2A-mediated DAG transport causes DGAT2 to relocate from the ER to LDs, thereby accelerating lipid storage and expanding LDs. The adaptive LD phospholipid monolayer and neutral lipid core furnish a protective environment, functioning as a sink for DAG, PA and eventually free fatty acids, which can be gradually converted into storage lipids without causing stress in the ER bilayer. This mechanism of lipid storage on LDs would detoxify the ER of harmful lipids, thereby acting as a branch of ER lipid and membrane quality control.

Our in vitro and in silico data show that ATG2A can bind and transport DAG, PA and possibly TAG. While TAG retrieval from the ER to LDs is possible, our pulse-chase experiments with fluorescent fatty acids reveal a minimal effect of DGAT inhibition on lipid incorporation into LDs (Fig. 2), indicating that DAG is mainly transported. The retrieval of DAG compared with TAG may prevent potential DAG signaling cascades or its rapid conversion to phospholipids via the Kennedy pathway.

The synergy between ATG2A and DGAT2 mirrors that of ATG2 and ATG9 during autophagosome biogenesis. These ATG2A partner proteins use the transported lipids to shape and grow the organelles. On LDs, other lipogenic enzymes may work with different LTPs to acquire and metabolize lipid substrates, such as fatty acids or PA, transferred from the ER to LDs. Such collaborative actions during lipogenesis probably protect the ER and enhance the efficiency of lipid storage in LDs.

Our study shows that ATG2A helps transfer various lipids between organelles, adjusting its activity based on the cellular metabolic status. During energy shortages, ATG2A transports PE, which is essential for ATG8 lipidation and the formation of the autophagosomal membrane³⁹. It also facilitates the movement of fatty acids from LDs to mitochondria⁴⁷. When energy is plentiful, it transports DAG from the ER to LDs for storage. The mechanism by which ATG2A performs these diverse tasks remains unclear, and a key question is whether ATG2A operates unidirectionally in lipid transport. If so, the orientation of the protein or its mode of binding to organelles could determine the directionality of lipid transfer. One possibility is that the orientation of ATG2A between the organelles’ interface changes depending on the metabolic status.

In conclusion, our data provide a unique description of LD growth mechanisms and highlight a unique ER protection mechanism: retrieving membrane-destabilizing lipid substrates to LDs, effectively outsourcing lipid storage to these organelles.

Methods

Cell culture

HeLa WT, ATG2A/B DKO and ATG9A KO cell lines (a gift from Dr. Juan Bonifacino, NIH) were maintained in high-glucose Dulbecco’s modified Eagle medium (DMEM, Dutscher) with 4.5 g l⁻¹ glucose, stabilized glutamine and sodium pyruvate, supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin (Gibco BRL). Cells were cultured at 37 °C in a humidified 5% CO₂ atmosphere. For imaging, cells were seeded on 35 mm glass-bottom dishes (P35G-0-20-C, MatTek Life Sciences) and grown for 24 h. Mycoplasma contamination was routinely checked by PCR.

Plasmids and cell transfection

Transient transfections were performed using the indicated plasmids and Polyethyleneimine HCl MAX (PEI MAX, 24765, Polysciences) following the manufacturer’s instructions. Cells were transfected for 24 h before imaging or analysis. The plasmids used in this study include pEGFP-C1-hATG2A (36456, Addgene). All other plasmids were gifted by their respective authors: ATG2A–mCherry by Dr. Olivier Vincent (IIBM - Universidad Autónoma de Madrid); TD-ATG2A–eGFP, ATG2A–3xFlag by Prof. Thomas Melia (Yale School of Medicine); DGAT2–GFP by Prof. Robert Yang (University of New South Wales); ACSL3–mCherry by Prof. Joachim Füllekrug (Heidelberg University); Lipin-1γ–GFP by Prof. Toyoshi Fujimoto (Juntendo University); GPAT4–EGFP by Prof. Sander Kersten (Cornell University); hpGPAT4–mCherry by Dr. Sarah Cohen (University of North Carolina at Chapel Hill); PLIN1–mCherry by Prof. David B. Savage (University of Cambridge); and EROX–BFP (Addgene plasmid 68126) by Dr. Erik Snapp. All newly received plasmids were sequence verified before use.

OA and fluorescent fatty acid treatment

A 2 mM OA stock was prepared by mixing 10% BSA in DPBS with DMEM (1:10, v/v) and adding OA (O1383, Sigma-Aldrich). The solution was incubated at 37 °C for 20 min with intermittent vortexing, filtered (0.2 µm) and stored at 4 °C. For fluorescent labeling, BODIPY 558/568 C12 (D3835, Invitrogen) was added at a 1:100 molar ratio and processed the same way. For LD induction experiments, cells were treated with 200 µM OA.

Fluorescent probes

Nuclei were stained with Hoechst 33342 (0.025% v/v; 62249, Thermo Fisher). LDs and membranes were labeled using HCS LipidTOX Deep Red (0.05% v/v; H34477) or BODIPY FL (0.025% v/v; D2183).

Incorporation of labeled lipids

For BPY-C12 incorporation, cells were incubated with 200 µM BPY-C12 + OA for 1 h at 37 °C. For PA labeling, TopFluor TMR-PA (810240 P, Avanti) was dried under argon, resuspended in ethanol (0.5 mM) and applied at 2 µM in DMEM for 1 h. For DAG labeling, TopFluor–DAG (810300, Avanti) was prepared similarly, with 2 µM in PBS + 2% FBS for 1 h.

Chemical inhibition of DGAT and Lipin enzymes

DGAT1/2 were inhibited using 5 µM of each inhibitor (PZ0207, PZ0233; Sigma) for 5 h. Lipin-1γ was inhibited with 100 µM propranolol hydrochloride (P0884, Sigma) for 5 h at 37 °C, 5% CO₂.

In vitro DAG FRET transfer assay

Purified 3xFlag-ATG2A was produced in Expi293 cells following the protocol in ref. ⁴¹. Liposomes and aLDs were prepared by combining solubilized lipids in desired molar ratios in borosilicate glass tubes, which were then dried under nitrogen. The dried lipids were further dried for 1 h in a vacuum chamber. Membranes were resuspended in 500 mM NaCl and 20 mM HEPES, pH 8.0. Liposomes were resuspended by vortexing, with 50 μl transferred to a 96-well assay plate. Donor membranes were included at 25 μM total phospholipid, and acceptor liposomes were added at 50 μM; aLDs could not be determined when fluorophores were absent. Reactions were performed in a BioTek Synergy H1 microplate reader at 30 °C. Rhodamine-PE (Avanti Research; 810150), NBD-PS (Avanti Research; 810195), NBD-PS (Avanti Research; 810198), Rhodamine-TAG (Avanti Research; 810103), NBD-12 cholesterol (Avanti Research; 810252), 1-NBD-stearoyl-2arachidonoyl DAG (Cayman Chemical; 10011300) and triolein (Sigma; T7140) were used in these experiments. Experiments were carried out with and without ATG2A, and the first condition served as a reference for the plotted data in the presence of ATG2A.

NBD-DAG transfer assay

Donor liposomes were prepared by incorporating 2 mol% NBD-labeled stearoyl-arachidonoyl DAG (NBD-DG) into a lipid mixture containing 0.5 mol% pyrene-phosphatidylcholine (pyrene-PC), 30 mol% DOPE and 65.5 mol% DOPC. Acceptor aLDs were assembled from 99.5 mol% DOPC and 0.5 mol% pyrene-PC. The assay leverages the fact that the NBD fluorophore exhibits a lower quantum yield in bilayer membranes compared with the hydrophobic core of lipid droplets; thus, the transfer of NBD-DG from liposomes to aLDs results in an increase in NBD fluorescence intensity. Lipid transfer reactions were performed using 25 µM total phospholipid from both donor and acceptor vesicles. The acceptor surface area was adjusted to be roughly twice that of the donor liposomes to promote unidirectional transfer. Reactions were initiated by adding ATG2A (final concentration 100 nM) or control buffer at time zero (T = 0). Protein was added over 1–5 min, although this addition period was not shown on the kinetic plot’s time axis. Reactions were conducted at 30 °C. At the end of each assay, Triton X-100 was added to solubilize all membrane components, equalizing the environment of NBD fluorophores and allowing normalization of fluorescence to the total NBD-lipid content. Data represent three independent membrane and protein preparation combinations, with each biological replicate measured in technical duplicate.

PA transfer assay

Donor liposomes were prepared by incorporating 2 mol% NBD-labeled PA (NBD-PA) and 2 mol% TMR-labeled PA (TMR-PA), enabling FRET-based quenching of NBD fluorescence by TMR. The remaining lipid composition included 0.5 mol% pyrene-PC, 30 mol% DOPE and 65.5 mol% DOPC. Acceptor aLDs consisted of 99.5 mol% DOPC and 0.5 mol% pyrene-PC. Lipid transfer reactions involved 25 µM total phospholipid from both donor and acceptor vesicles, with the acceptor membrane surface area approximately twice that of the donor to promote transfer directionality. Reactions were initiated by adding 100 nM ATG2A or an equivalent volume of control buffer at 30 °C, with protein addition completed within 1–5 min (this addition period was not shown on the kinetic data’s time axis). Lipid transfer kinetics were measured by tracking the dequenching of NBD fluorescence. Initial transfer rates were determined by fitting the fluorescence data to an exponential rise-to-plateau function and deriving the maximum slope. Each data point represents the average of three independent biological replicates, each prepared with separate protein and membrane samples, measured in technical duplicates.

Fluorescence transfer assay with liposomes and emulsion droplets

Liposomes were prepared with 70% PC, 30% PE and 3% TopFluor–DAG, dried, resuspended in HKM buffer and sonicated. Emulsions (70% PC and 30% PE) were dried, resuspended in trioctanoate and labeled with LipidTOX. Liposomes were mixed with ATG2A-3xFlag and emulsions. Samples were deposited on BSA-treated coverslips. Experiments were repeated three times with different lipid batches.

ATG2A transport-deficient mutations

The ten mutations in the ATG2A groove are the following: F28N, V82E, L101D, V169Q, F171R, V178D, L229D, L234N, M259K and L285E. In silico analysis using the OmegaFold prediction program indicated that the mutations cause minimal structural perturbation³⁹. The selected combination is predicted to fold similarly to WT ATG2A, with the secondary and tertiary structures largely preserved, except near the mutations

Molecular dynamics simulations and visualization tools

Molecular dynamics simulations were performed with the GROMACS (v 2021.x)⁷¹ package and the MARTINI 3 force field⁷². Visual Molecular Dynamics (VMD)⁷³ was used for the renders of Atg2 filled with lipids.

Protein preparation and coarse-graining: the atomistic structure of the protein was obtained from the AlphaFold Protein Structure Database (AF-Q2TAZ0-F1) and converted to CG using the Martinize⁷⁴ script. Residues 37–1,720 from ATG2A were considered, as helices at the N- and C-terminal parts of the protein were removed to facilitate lipid binding. An additional elastic network with a force constant of 1,000 kJ mol⁻¹ nm⁻² was used to restrain the secondary structure of the protein, with a 0 nm elastic bond lower cutoff and an upper cutoff of 0.8 nm.
System setup and lipid-binding protocol: the CG ATG2A model was then placed in the center of a cubic box, with a 2.0 nm distance between the protein and the box edge. One lipid molecule was randomly placed in the bulk solvent in each of the five replicas, without any distance restrictions. The system was then solvated, considering a Van der Waals distance of 0.21 nm, neutralized and ionized with 0.12 M NaCl. Each replica was simulated for 250 ns, and the procedure was repeated iteratively, such that the structure of the protein with n lipids bound in it served as the starting structure for the addition of the (n + 1)th lipid, solvating and ionizing again every time a new lipid was added. This protocol was concluded once no more binding to ATG2A was observed.
Equilibration and production run parameters: initial equilibration was carried out by performing energy minimization using the steepest descent algorithm, followed by a short MD run of 125 ps. Production runs were performed at 310 K using a velocity-rescale thermostat⁷⁵, with separate temperature coupling for protein and nonprotein particles. The md integrator was used for the production runs, with a time step of 25 fs. The Parrinello–Rahman barostat⁷⁶ was used to maintain a pressure of 1 bar, along with an isotropic pressure coupling scheme and an nstpcouple parameter set to 10. The Coulombic terms were calculated using the reaction field, with an epsilon (dielectric constant) of 15 and a cutoff distance of 1.1 nm. A cutoff scheme was applied for the Van der Waals terms, using a cutoff distance of 1.1 nm, and the Verlet cutoff scheme was used for the potential shift⁷⁷. The nonbonded interactions were calculated by generating a pair-list using the Verlet scheme with a buffer tolerance of 0.005.
Solvation analysis and lipid solvation computation: a tcl script was used to compute the solvation number of the lipids, counting the number of water molecules within 5.0 Å of the lipid tail beads (headgroup and backbone beads were excluded) for each trajectory frame, for each replica. For that purpose, the final lipid-bound systems were simulated for 500 ns and the total solvation was computed at each frame. The solvation of the headgroups was computed in a similar manner, but considering the headgroups and backbone beads instead.

Immunogold labeling

HeLa cells were fixed with a mixture of 2% PFA and 0.125% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4, for 2 h, and processed for ultracryomicrotomy as described previously⁷⁸. Ultrathin cryosections were double-immunogold labeled using anti-GFP (rabbit polyclonal, Life Technologies; 1:100) and anti-PDI (mouse monoclonal, Genetex; 1:300) primary antibodies. Labeling was performed sequentially to allow selective detection of each antigen with distinct gold particles. Sections were first incubated with the anti-GFP antibody, followed by protein A conjugated to 15-nm gold particles (CMC, UMC Utrecht). After thorough washing and blocking, sections were incubated with the anti-PDI antibody, followed by a rabbit anti-mouse bridging antibody (Dako), and subsequently with protein A conjugated to 10-nm gold particles. Immunogold-labeled cryosections were observed under an 80 kV JEOL 1400 microscope equipped with an Orius High speed (Gatan) camera and, for higher resolution, under an FEI Tecnai 12 microscope equipped with a OneView 4k Gatan camera.

Mass spectrometry analysis

For each condition, 100,000 WT or ATG2A/B DKO HeLa cells were seeded in 6-well plates. Two lipidomics experiments were performed: (1) WT + ATG2A–eGFP and DKO cells were incubated with OA for 24 h, then pulsed 1 h with OA + BPY-C12 to assess DAG/TAG ratios under DGAT inhibition or control; (2) cells were treated with OA for 24 h and pulsed 1 h with OA + NBD-palmitate (1:1) to quantify DAG, TAG and phospholipids. After treatment, cells were washed with ice-cold 150 mM ammonium bicarbonate, scraped, pelleted (19,000g, 5 min, 4 °C), resuspended in lysis buffer, snap frozen and stored at −80 °C. Lipids were extracted and analyzed by mass spectrometry in triplicate. Extracts were resuspended in 100 μl chloroform/methanol (2:1) and data were normalized to total lipid content and the mean WT value per experiment.

LD purification

LDs were purified following the protocol described by Peterson et al.⁷⁹. HeLa WT cells, with or without ATG2A–eGFP, were treated with 200 µM OA for 24 h. For selected samples, DGAT2 was inhibited for 5 h, followed by a 1-h pulse with TopFluor–DAG/OA (1:1) in BSA-supplemented DMEM. Cells from two 150 mm dishes per condition were collected, pelleted (500g, 10 min, 4 °C), and resuspended in 3 ml ice-cold HLM buffer with protease inhibitors. Cells were homogenized (40 passages through a 21 G needle) and centrifuged (1,000g, 10 min, 4 °C) to remove nuclei. The post-nuclear supernatant was adjusted to 20% sucrose, layered with 5% sucrose and HLM buffer, and centrifuged at 24,000g (TST41 rotor, 30 min, 4 °C) to float LDs. About 500 µl of the LD fraction was collected from the top for analysis.

TLC

For TLC analysis, LD samples were normalized to the protein content of the corresponding post-nuclear supernatant. When LD protein was too low, equal volumes were used, referencing the lowest post-nuclear supernatant concentration (200 µl). Lipids were extracted with chloroform/methanol (2:1, 1:20 sample-to-solvent), vortexed and agitated for 20 min at room temperature. Phase separation was induced by adding water, the organic phase was collected, dried under nitrogen and resuspended in 50 µl chloroform/methanol (2:1). About 20 µl of each extract was spotted on TLC plates (100933, Merck Millipore) and developed in hexane/diethyl ether/acetic acid (40:60:2). Plates were air dried and fluorescent lipids were visualized under UV.

GERV production and extraction

After transfection for 24 h the cultured cells were transferred into a hypotonic culture media DMEM:H₂O (5:95% v/v) at pH 7.4, at 37 °C, 5% CO₂, for 15 min, to induce cell volume increase and giant organelle vesicles. For DEGERVs experiments, cells were mechanically lysed by extensive pipetting and the solution was filtered at 10 μm to extract giant organelle vesicles, including GERVs.

DEGERV production

TAG droplets were prepared using an oil-in-hypotonic media emulsion as described in ref. ⁵⁷. Briefly, 8 μl of triolein was mixed with 300 µl of hypotonic media. For the TAG:DAG droplets, a 9:1 (v/v) ratio was used to create the emulsion. The mixture was then sonicated and vortexed to produce small droplets. The resulting emulsion was gently mixed with the GERV solution via pipetting and incubated on the bench for 10 min to form DEGERVs. The DEGERVs sample was subsequently transferred onto a glass coverslip pretreated with 10% (w/w) BSA and washed five times with hypotonic media.

FRAP experiments

Fluorescence recovery after photobleaching (FRAP) experiments were performed by photobleaching the total surface of a droplet with the 488-nm laser line at 100% laser power. Recovery of fluorescence was monitored every 30 s for 15 min immediately after photobleaching. The FRAP curves were normalized by the fluorescence before and after the bleach in the region of interest.

Confocal microscopy imaging

All images were acquired using a Carl Zeiss LSM800 microscope with an oil-immersed ×63 objective. eGFP, NBD, TopFluor and BODIPY FL fluorescence was excited at 488 nm, and emission was detected between 510 and 550 nm, while mCherry and TMR tagged protein fluorescence was excited at 561 nm, and emission was detected between 580 and 650 nm. Deep-red fluorescence was excited at 640 nm and was detected above 650 nm. BFP fluorescence was excited at 405 nm and detected below 500 nm. For both cells and DEGERVs experiments, z-stack were acquired with a stack gap between 0.4 and 0.8 μm to cover the whole object of interest. All fluorescence signals were analyzed with Fiji (see below).

Quantification of LDs size and number

LDs were segmented using the StarDist algorithm (ZeroCostDL4Mic), and cells were segmented in Fiji (v2.9.0) by generating an RGB overlay from the LD fluorescent signal. Parameters such as object area and count were quantified using CellProfiler (v4.2.6) based on masks from segmented cells and LDs. Data were processed in Python (Spyder, v3.10) and visualized using GraphPad Prism (v8.4.3).

Quantification of fluorescent lipid and protein signals on LD

For the quantification of labeled lipid signal, fluorescence intensity was measured within LDs, and the background signal was subtracted. For protein recruitment to the surface of LDs, signal intensity was measured by drawing dots on individual LDs using the LD marker channel as a reference. The fluorescence intensity of the protein signal on LDs was then normalized to the average background signal measured in the cytosol. For kinetic experiments (Fig. 1, ATG2 DKO and Extended Data Fig. 2, WT), signal intensity was measured over time in four ATG2 DKO and nine WT cells. The values at the 60 min time point correspond to static measurements performed on a larger number of cells (1 h time point).

Statistics and reproducibility

No statistical method was used to predetermine sample size and no data were excluded from the analyses. All experiments were performed in at least three independent biological replicates. Data were analyzed using GraphPad Prism 8.4.3 (GraphPad Software). In most cases, SuperPlots were used to display biological variability and replicate consistency, with individual cell or LD measurements represented as small dots and the corresponding mean value per biological replicate as larger dots. Data are presented as the mean of independent biological replicates ± s.d., except in cases of small sample sizes, where ± s.e.m. is shown.

Normality was assessed using the D’Agostino and Pearson test. For normally distributed data, a two-tailed unpaired Student’s t-test was used; for non-normally distributed data, a two-tailed Mann–Whitney test was applied. For the lipidomic analysis (Fig. 2e), statistical significance was determined using a one-way analysis of variance. Comparisons with P < 0.05 were considered statistically significant and are indicated with asterisks (*P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001); nonsignificant values (P ≥ 0.05) are indicated on the figures.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41594-025-01689-0.

Supplementary information

Reporting Summary^{(59.2KB, pdf)}

Peer Review File^{(1.6MB, pdf)}

Supplementary Video 1^{(2.2MB, m4v)}

Live-cell imaging showing 1 h OA pulse-chase incorporation into lipid droplets, highlighting ATG2A-dependent fatty acid uptake (related to Extended Data Fig. 2). On left, LD (magenta) ATG2A–eGFP (green) are shown. On right, only the LD signal is displayed.

Source data

Source Data Fig. 1^{(141.9KB, xlsx)}

Statistical source data.

Source Data Fig. 2^{(109.2KB, xlsx)}

Statistical source data.

Source Data Fig. 3^{(20.5KB, xlsx)}

Statistical source data.

Source Data Fig. 5^{(108.3KB, xlsx)}

Statistical source data.

Source Data Fig. 6^{(384.3KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 1^{(56.5KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 2^{(16.9KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 2h^{(2.7MB, pdf)}

Unprocessed TLC.

Source Data Extended Data Fig. 3^{(13.7KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 5^{(26.2KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 6^{(11KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 7^{(31.8KB, xlsx)}

Statistical source data.

Acknowledgements

We thank the group members for their helpful discussions. We thank J. Bonifacino for sharing the knockout cells used in this study and for helpful discussions. This work was supported by the Fondation pour la Recherche Médicale (FRM) no. EQU202103012564, the Programme Emergence de la Ville de Paris (DDEEES 165) and the ANR-24-CE11-3433-01 to A.R.T.; FRM no. FDT202504020489 to H.E., the National Institutes of Health (GM100930) to T.J.M. and 1F31DK136246 to J.L.K. S.V. acknowledges support from the Swiss National Science Foundation (no. 310030_219264). This work was supported by a grant from the Swiss National Supercomputing Centre (CSCS) under project ID 637 on Alps. For the immuno-electron microscopy, the present work has benefited from Imagerie-Gif core facility supported by l’Agence Nationale de la Recherche (FBI ANR-24-INBS-0005 (BIOGEN); SPS ANR-17-EUR-0007, EUR SPS-GSR). We also acknowledge the ImagoSeine core facility of Institut Jacques Monod, member of France-BioImaging (ANR-10-INBS-04) and IBiSA, with the support of Labex “Who Am I”, Inserm Plan Cancer, Region Ile-de-France and Fondation Bettencourt Schueller.

Extended data

Author contributions

H.E. and A.R.T. designed the research. H.E., A.D., F.G., M.Z., T.J.M., J.L.K., S.V., D.A. and A.R.T. designed experiments. H.E. conducted all experiments with assistance from A.D., who carried out the in vitro experiments using DEGERVs, and M.Z. with the analysis. J.L.K. from the laboratory of T.J.M. conducted the in vitro transfer assays and provided the purified full-length ATG2A protein. F.G. performed the IEM analyses. S.V. and D.A. performed the in silico analyses. A.R.T. wrote the manuscript, which was reviewed and edited by all co-authors.

Peer review

Peer review information

Nature Structural & Molecular Biology thanks Michael Ragusa and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Melina Casadio, in collaboration with the Nature Structural & Molecular Biology team.

Data availability

Mass spectrometry data generated in this study have been deposited in the MetaboLights repository under accession number MTBLS12951 (https://www.ebi.ac.uk/metabolights/MTBLS12951), including raw mass spectrometry files and processed lipidomics identification tables. All other data supporting the findings of this study are available from the corresponding author upon reasonable request. Source data are provided with this paper.

Competing interests

A.R.T. is an advisor and cofounder of Oria Bioscience. The other authors declare no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

is available for this paper at 10.1038/s41594-025-01689-0.

Supplementary information

The online version contains supplementary material available at 10.1038/s41594-025-01689-0.

References

1.Schwarz, D. S. & Blower, M. D. The endoplasmic reticulum: structure, function and response to cellular signaling. Cell. Mol. Life Sci.73, 79–94 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Oakes, S. A. & Papa, F. R. The role of endoplasmic reticulum stress in human pathology. Annu. Rev. Pathol. Mech. Dis.10, 173–194 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Halbleib, K. et al. Activation of the unfolded protein response by lipid bilayer stress. Mol. Cell67, 673–684 (2017). [DOI] [PubMed] [Google Scholar]
4.Roberts, M. A. & Olzmann, J. A. Protein quality control and lipid droplet metabolism. Annu. Rev. Cell Dev. Biol.36, 115–139 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Vevea, J. D. et al. Role for lipid droplet biogenesis and microlipophagy in adaptation to lipid imbalance in yeast. Dev. Cell35, 584–599 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Thiam, A. R., Farese, R. V. Jr & Walther, T. C. The biophysics and cell biology of lipid droplets. Nat. Rev. Mol. Cell Biol.14, 775 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Garcia, E. J., Vevea, J. D. & Pon, L. A. Lipid droplet autophagy during energy mobilization, lipid homeostasis and protein quality control. Front. Biosci.23, 1552 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Olzmann, J. A. & Carvalho, P. Dynamics and functions of lipid droplets. Nat. Rev. Mol. Cell Biol.1, 137–155 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Chitraju, C. et al. Triglyceride synthesis by DGAT1 protects adipocytes from lipid-induced ER stress during lipolysis. Cell Metab.26, 407–418 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Hariri, H. et al. Mdm1 maintains endoplasmic reticulum homeostasis by spatially regulating lipid droplet biogenesis. J. Cell Biol.218, 1319–1334 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Renne, M. F. & Hariri, H. Lipid droplet–organelle contact sites as hubs for fatty acid metabolism, trafficking, and metabolic channeling. Front. Cell Dev. Biol.9, 726261 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Zadoorian, A., Du, X. & Yang, H. Lipid droplet biogenesis and functions in health and disease. Nat. Rev. Endocrinol.19, 443–459 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Thiam, A. R. & Ikonen, E. Lipid droplet nucleation. Trends Cell Biol.31, 108–118 (2021). [DOI] [PubMed] [Google Scholar]
14.Ben M’barek, K. et al. ER membrane phospholipids and surface tension control cellular lipid droplet formation. Dev. Cell41, 591–604 (2017). [DOI] [PubMed] [Google Scholar]
15.Thiam, A. R. & Forêt, L. The physics of lipid droplet nucleation, growth and budding. Biochim. Biophys. Acta1861, 715–722 (2016). [DOI] [PubMed] [Google Scholar]
16.Zoni, V. et al. Pre-existing bilayer stresses modulate triglyceride accumulation in the ER versus lipid droplets. eLife10, e62886 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Adeyo, O. et al. The yeast lipin orthologue Pah1p is important for biogenesis of lipid droplets. J. Cell Biol.192, 1043–1055 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Zoni, V. et al. Seipin accumulates and traps diacylglycerols and triglycerides in its ring-like structure. Proc. Natl Acad. Sci. USA118, 2017205118 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Prasanna, X. et al. Seipin traps triacylglycerols to facilitate their nanoscale clustering in the endoplasmic reticulum membrane. PLoS Biol.19, e3000998 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Wang, H. et al. Seipin is required for converting nascent to mature lipid droplets. eLife5, e16582 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Salo, V. T. et al. Seipin facilitates triglyceride flow to lipid droplet and counteracts droplet ripening via endoplasmic reticulum contact. Dev. Cell50, 478–493 (2019). [DOI] [PubMed] [Google Scholar]
22.Chorlay, A. et al. Membrane asymmetry imposes directionality on lipid droplet emergence from the ER. Dev. Cell50, 25–42 (2019). [DOI] [PubMed] [Google Scholar]
23.Mathiowetz, A. J. & Olzmann, J. A. Lipid droplets and cellular lipid flux. Nat. Cell Biol.26, 331–345 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Wilfling, F. et al. Triacylglycerol synthesis enzymes mediate lipid droplet growth by relocalizing from the ER to lipid droplets. Dev. Cell24, 384–399 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Wang, H. et al. Lipin-1γ isoform is a novel lipid droplet-associated protein highly expressed in the brain. FEBS Lett.585, 1979–1984 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Dhiman, R., Caesar, S., Thiam, A. R. & Schrul, B. Lipid droplet biogenesis and function in health and disease. Semin. Cell Dev. Biol.108, 4–13 (2020). [DOI] [PubMed] [Google Scholar]
27.Kory, N., Farese, R. V. Jr & Walther, T. C. Targeting fat: mechanisms of protein localization to lipid droplets. Trends Cell Biol.26, 535–546 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Nguyen, T. B. et al. DGAT1-dependent lipid droplet biogenesis protects mitochondrial function during starvation-induced autophagy. Dev. Cell42, 9–21 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Du, X. et al. ORP5 localizes to ER–lipid droplet contacts and regulates the level of PI(4)P on lipid droplets. J. Cell Biol.219, e201905162 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Guyard, V. et al. ORP5 and ORP8 orchestrate lipid droplet biogenesis and maintenance at ER–mitochondria contact sites. J. Cell Biol.221, e202112107 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Kumar, N. et al. VPS13A and VPS13C are lipid transport proteins differentially localized at ER contact sites. J. Cell Biol.217, 3625–3639 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Pfisterer, S. G. et al. Lipid droplet and early autophagosomal membrane targeting of Atg2A and Atg14L in human tumor cells. J. Lipid Res.55, 1267–1278 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Velikkakath, A. K. G., Nishimura, T., Oita, E., Ishihara, N. & Mizushima, N. Mammalian Atg2 proteins are essential for autophagosome formation and important for regulation of size and distribution of lipid droplets. Mol. Biol. Cell23, 896–909 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Zouiouich, M. et al. MOSPD2 is an endoplasmic reticulum–lipid droplet tether functioning in LD homeostasis. J. Cell Biol.221, e202110044 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Hynynen, R. et al. OSBP-related protein 2 is a sterol receptor on lipid droplets that regulates the metabolism of neutral lipids. J. Lipid Res.50, 1305–1315 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Chung, J. et al. PI4P/phosphatidylserine countertransport at ORP5-and ORP8-mediated ER–plasma membrane contacts. Science349, 428–432 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Valverde, D. P. et al. ATG2 transports lipids to promote autophagosome biogenesis. J. Cell Biol.218, 1787–1798 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Tamura, N. et al. Differential requirement for ATG2A domains for localization to autophagic membranes and lipid droplets. FEBS Lett.591, 3819–3830 (2017). [DOI] [PubMed] [Google Scholar]
39.Korfhage, J. L. et al. ATG2A-mediated bridge-like lipid transport regulates lipid droplet accumulation. Preprint at bioRxiv10.1101/2023.08.14.553257 (2023).
40.Osawa, T. et al. Atg2 mediates direct lipid transfer between membranes for autophagosome formation. Nat. Struct. Mol. Biol.26, 281–288 (2019). [DOI] [PubMed] [Google Scholar]
41.Hama, Y., Morishita, H. & Mizushima, N. Regulation of ER-derived membrane dynamics by the DedA domain-containing proteins VMP1 and TMEM41B. EMBO Rep.23, e53894 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Tábara, L.-C. & Escalante, R. VMP1 establishes ER-microdomains that regulate membrane contact sites and autophagy. PLoS ONE11, e0166499 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.McEwan, D. G. & Ryan, K. M. ATG2 and VPS13 proteins: molecular highways transporting lipids to drive membrane expansion and organelle communication. FEBS J.289, 7113–7127 (2022). [DOI] [PubMed] [Google Scholar]
44.Maeda, S. et al. Structure, lipid scrambling activity and role in autophagosome formation of ATG9A. Nat. Struct. Mol. Biol.27, 1194–1201 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Matoba, K. et al. Atg9 is a lipid scramblase that mediates autophagosomal membrane expansion. Nat. Struct. Mol. Biol.27, 1185–1193 (2020). [DOI] [PubMed] [Google Scholar]
46.Chorlay, A., Forêt, L. & Thiam, A. R. Origin of gradients in lipid density and surface tension between connected lipid droplet and bilayer. Biophys. J.120, 5491–5503 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Mailler, E. et al. The autophagy protein ATG9A enables lipid mobilization from lipid droplets. Nat. Commun.12, 6750 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Han, G.-S. & Carman, G. M. Characterization of the human LPIN1-encoded phosphatidate phosphatase isoforms. J. Biol. Chem.285, 14628–14638 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Santinho, A., Carpentier, M., Lopes Sampaio, J., Omrane, M. & Thiam, A. R. Giant organelle vesicles to uncover intracellular membrane mechanics and plasticity. Nat. Commun.15, 3767 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
50.King, C., Sengupta, P., Seo, A. Y. & Lippincott-Schwartz, J. ER membranes exhibit phase behavior at sites of organelle contact. Proc. Natl Acad. Sci. USA117, 7225–7235 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Jaiswal, A., Hoerth, C. H., Pereira, A. M. Z. & Lorenz, H. Improved spatial resolution by induced live cell and organelle swelling in hypotonic solutions. Sci. Rep.9, 12911 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.François-Martin, C. & Pincet, F. Actual fusion efficiency in the lipid mixing assay—comparison between nanodiscs and liposomes. Sci. Rep.7, 43860 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Sarkar, P. et al. Environment-sensitive fluorescence of 7-nitrobenz-2-oxa-1, 3-diazol-4-yl (NBD)-labeled ligands for serotonin receptors. Molecules26, 3848 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Srinivasan, S., Álvarez, D., John Peter, A. T. & Vanni, S. Unbiased MD simulations identify lipid binding sites in lipid transfer proteins. J. Cell Biol.223, e202312055 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Chorlay, A. & Thiam, A. R. Neutral lipids regulate amphipathic helix affinity for model lipid droplets. J. Cell Biol.219, e201907099 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Caillon, L. et al. Triacylglycerols sequester monotopic membrane proteins to lipid droplets. Nat. Commun.11, 3944 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Damm, A. et al. Steric repulsion counteracts ER–to–lipid droplet protein movement. Sci. Adv.11, eadu6998 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Chorlay, A., Santinho, A. & Thiam, A. R. Making droplet-embedded vesicles to model cellular lipid droplets. STAR Protoc.1, 100116 (2020). [DOI] [PMC free article] [PubMed]
59.Chorlay, A. & Thiam, A. R. An asymmetry in monolayer tension regulates lipid droplet budding direction. Biophys. J.114, 631–640 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Kory, N., Thiam, A.-R., Farese, R. V. Jr & Walther, T. C. Protein crowding is a determinant of lipid droplet protein composition. Dev. Cell34, 351–363 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
61.Ajjaji, D. et al. Dual binding motifs underpin the hierarchical association of perilipins1–3 with lipid droplets. Mol. Biol. Cell30, 703–716 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Chung, J. et al. LDAF1 and seipin form a lipid droplet assembly complex. Dev. Cell51, 551–563 (2019). [DOI] [PMC free article] [PubMed]
63.Mizrak, A. et al. Single-molecule analysis of protein targeting from the endoplasmic reticulum to lipid droplets. Preprint at bioRxiv10.1101/2024.08.27.610018 (2024).
64.Szymanski, K. M. et al. The lipodystrophy protein seipin is found at endoplasmic reticulum lipid droplet junctions and is important for droplet morphology. Proc. Natl Acad. Sci. USA104, 20890–20895 (2007). [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Wang, S. et al. Seipin and the membrane-shaping protein Pex30 cooperate in organelle budding from the endoplasmic reticulum. Nat. Commun.9, 2939 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Salo, V. T. et al. Seipin regulates ER–lipid droplet contacts and cargo delivery. EMBO J.35, 2699–2716 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Santinho, A., Chorlay, A., Foret, L. & Thiam, A. R. Fat inclusions strongly alter membrane mechanics. Biophys. J.120, 607–617 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Santinho, A. et al. Membrane curvature catalyzes lipid droplet assembly. Curr. Biol.30, 2481–2494 (2020). [DOI] [PubMed] [Google Scholar]
69.Ganesan, S., Shabits, B. N. & Zaremberg, V. Tracking diacylglycerol and phosphatidic acid pools in budding yeast. Lipid Insights8s1, LPI.S31781 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Zhukovsky, M. A., Filograna, A., Luini, A., Corda, D. & Valente, C. Phosphatidic acid in membrane rearrangements. FEBS Lett.593, 2428–2451 (2019). [DOI] [PubMed] [Google Scholar]
71.Van Der Spoel, D. et al. GROMACS: fast, flexible, and free. J. Comput. Chem.26, 1701–1718 (2005). [DOI] [PubMed] [Google Scholar]
72.Marrink, S. J., Risselada, H. J., Yefimov, S., Tieleman, D. P. & De Vries, A. H. The MARTINI force field: coarse grained model for biomolecular simulations. J. Phys. Chem. B111, 7812–7824 (2007). [DOI] [PubMed] [Google Scholar]
73.Humphrey, W., Dalke, A. & Schulten, K. VMD: visual molecular dynamics. J. Mol. Graph.14, 33–38 (1996). [DOI] [PubMed] [Google Scholar]
74.de Jong, D. H. et al. Improved parameters for the martini coarse-grained protein force field. J. Chem. Theory Comput.9, 687–697 (2012). [DOI] [PubMed] [Google Scholar]
75.Bussi, G., Donadio, D. & Parrinello, M. Canonical sampling through velocity rescaling. J. Chem. Phys.126, 014101 (2007). [DOI] [PubMed] [Google Scholar]
76.Parrinello, M. & Rahman, A. Polymorphic transitions in single crystals: a new molecular dynamics method. J. Appl. Phys.52, 7182–7190 (1981). [Google Scholar]
77.de Jong, D. H., Baoukina, S., Ingólfsson, H. I. & Marrink, S. J. Martini straight: boosting performance using a shorter cutoff and GPUs. Comput. Phys. Commun.199, 1–7 (2016). [Google Scholar]
78.Slot, J. W. & Geuze, H. J. Cryosectioning and immunolabeling. Nat. Protoc.2, 2480–2491 (2007). [DOI] [PubMed] [Google Scholar]
79.Peterson, C. W., Deol, K. K., To, M. & Olzmann, J. A. Optimized protocol for the identification of lipid droplet proteomes using proximity labeling proteomics in cultured human cells. STAR Protoc.2, 100579 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Reporting Summary^{(59.2KB, pdf)}

Peer Review File^{(1.6MB, pdf)}

Supplementary Video 1^{(2.2MB, m4v)}

Source Data Fig. 1^{(141.9KB, xlsx)}

Statistical source data.

Source Data Fig. 2^{(109.2KB, xlsx)}

Statistical source data.

Source Data Fig. 3^{(20.5KB, xlsx)}

Statistical source data.

Source Data Fig. 5^{(108.3KB, xlsx)}

Statistical source data.

Source Data Fig. 6^{(384.3KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 1^{(56.5KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 2^{(16.9KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 2h^{(2.7MB, pdf)}

Unprocessed TLC.

Source Data Extended Data Fig. 3^{(13.7KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 5^{(26.2KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 6^{(11KB, xlsx)}

Statistical source data.

Source Data Extended Data Fig. 7^{(31.8KB, xlsx)}

Statistical source data.

Data Availability Statement

[CR1] 1.Schwarz, D. S. & Blower, M. D. The endoplasmic reticulum: structure, function and response to cellular signaling. Cell. Mol. Life Sci.73, 79–94 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR2] 2.Oakes, S. A. & Papa, F. R. The role of endoplasmic reticulum stress in human pathology. Annu. Rev. Pathol. Mech. Dis.10, 173–194 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR3] 3.Halbleib, K. et al. Activation of the unfolded protein response by lipid bilayer stress. Mol. Cell67, 673–684 (2017). [DOI] [PubMed] [Google Scholar]

[CR4] 4.Roberts, M. A. & Olzmann, J. A. Protein quality control and lipid droplet metabolism. Annu. Rev. Cell Dev. Biol.36, 115–139 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR5] 5.Vevea, J. D. et al. Role for lipid droplet biogenesis and microlipophagy in adaptation to lipid imbalance in yeast. Dev. Cell35, 584–599 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR6] 6.Thiam, A. R., Farese, R. V. Jr & Walther, T. C. The biophysics and cell biology of lipid droplets. Nat. Rev. Mol. Cell Biol.14, 775 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR7] 7.Garcia, E. J., Vevea, J. D. & Pon, L. A. Lipid droplet autophagy during energy mobilization, lipid homeostasis and protein quality control. Front. Biosci.23, 1552 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR8] 8.Olzmann, J. A. & Carvalho, P. Dynamics and functions of lipid droplets. Nat. Rev. Mol. Cell Biol.1, 137–155 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR9] 9.Chitraju, C. et al. Triglyceride synthesis by DGAT1 protects adipocytes from lipid-induced ER stress during lipolysis. Cell Metab.26, 407–418 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Hariri, H. et al. Mdm1 maintains endoplasmic reticulum homeostasis by spatially regulating lipid droplet biogenesis. J. Cell Biol.218, 1319–1334 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR11] 11.Renne, M. F. & Hariri, H. Lipid droplet–organelle contact sites as hubs for fatty acid metabolism, trafficking, and metabolic channeling. Front. Cell Dev. Biol.9, 726261 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Zadoorian, A., Du, X. & Yang, H. Lipid droplet biogenesis and functions in health and disease. Nat. Rev. Endocrinol.19, 443–459 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR13] 13.Thiam, A. R. & Ikonen, E. Lipid droplet nucleation. Trends Cell Biol.31, 108–118 (2021). [DOI] [PubMed] [Google Scholar]

[CR14] 14.Ben M’barek, K. et al. ER membrane phospholipids and surface tension control cellular lipid droplet formation. Dev. Cell41, 591–604 (2017). [DOI] [PubMed] [Google Scholar]

[CR15] 15.Thiam, A. R. & Forêt, L. The physics of lipid droplet nucleation, growth and budding. Biochim. Biophys. Acta1861, 715–722 (2016). [DOI] [PubMed] [Google Scholar]

[CR16] 16.Zoni, V. et al. Pre-existing bilayer stresses modulate triglyceride accumulation in the ER versus lipid droplets. eLife10, e62886 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Adeyo, O. et al. The yeast lipin orthologue Pah1p is important for biogenesis of lipid droplets. J. Cell Biol.192, 1043–1055 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Zoni, V. et al. Seipin accumulates and traps diacylglycerols and triglycerides in its ring-like structure. Proc. Natl Acad. Sci. USA118, 2017205118 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR19] 19.Prasanna, X. et al. Seipin traps triacylglycerols to facilitate their nanoscale clustering in the endoplasmic reticulum membrane. PLoS Biol.19, e3000998 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR20] 20.Wang, H. et al. Seipin is required for converting nascent to mature lipid droplets. eLife5, e16582 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR21] 21.Salo, V. T. et al. Seipin facilitates triglyceride flow to lipid droplet and counteracts droplet ripening via endoplasmic reticulum contact. Dev. Cell50, 478–493 (2019). [DOI] [PubMed] [Google Scholar]

[CR22] 22.Chorlay, A. et al. Membrane asymmetry imposes directionality on lipid droplet emergence from the ER. Dev. Cell50, 25–42 (2019). [DOI] [PubMed] [Google Scholar]

[CR23] 23.Mathiowetz, A. J. & Olzmann, J. A. Lipid droplets and cellular lipid flux. Nat. Cell Biol.26, 331–345 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR24] 24.Wilfling, F. et al. Triacylglycerol synthesis enzymes mediate lipid droplet growth by relocalizing from the ER to lipid droplets. Dev. Cell24, 384–399 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR25] 25.Wang, H. et al. Lipin-1γ isoform is a novel lipid droplet-associated protein highly expressed in the brain. FEBS Lett.585, 1979–1984 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR26] 26.Dhiman, R., Caesar, S., Thiam, A. R. & Schrul, B. Lipid droplet biogenesis and function in health and disease. Semin. Cell Dev. Biol.108, 4–13 (2020). [DOI] [PubMed] [Google Scholar]

[CR27] 27.Kory, N., Farese, R. V. Jr & Walther, T. C. Targeting fat: mechanisms of protein localization to lipid droplets. Trends Cell Biol.26, 535–546 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.Nguyen, T. B. et al. DGAT1-dependent lipid droplet biogenesis protects mitochondrial function during starvation-induced autophagy. Dev. Cell42, 9–21 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR29] 29.Du, X. et al. ORP5 localizes to ER–lipid droplet contacts and regulates the level of PI(4)P on lipid droplets. J. Cell Biol.219, e201905162 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR30] 30.Guyard, V. et al. ORP5 and ORP8 orchestrate lipid droplet biogenesis and maintenance at ER–mitochondria contact sites. J. Cell Biol.221, e202112107 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR31] 31.Kumar, N. et al. VPS13A and VPS13C are lipid transport proteins differentially localized at ER contact sites. J. Cell Biol.217, 3625–3639 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR32] 32.Pfisterer, S. G. et al. Lipid droplet and early autophagosomal membrane targeting of Atg2A and Atg14L in human tumor cells. J. Lipid Res.55, 1267–1278 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR33] 33.Velikkakath, A. K. G., Nishimura, T., Oita, E., Ishihara, N. & Mizushima, N. Mammalian Atg2 proteins are essential for autophagosome formation and important for regulation of size and distribution of lipid droplets. Mol. Biol. Cell23, 896–909 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR34] 34.Zouiouich, M. et al. MOSPD2 is an endoplasmic reticulum–lipid droplet tether functioning in LD homeostasis. J. Cell Biol.221, e202110044 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR35] 35.Hynynen, R. et al. OSBP-related protein 2 is a sterol receptor on lipid droplets that regulates the metabolism of neutral lipids. J. Lipid Res.50, 1305–1315 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR36] 36.Chung, J. et al. PI4P/phosphatidylserine countertransport at ORP5-and ORP8-mediated ER–plasma membrane contacts. Science349, 428–432 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR37] 37.Valverde, D. P. et al. ATG2 transports lipids to promote autophagosome biogenesis. J. Cell Biol.218, 1787–1798 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR38] 38.Tamura, N. et al. Differential requirement for ATG2A domains for localization to autophagic membranes and lipid droplets. FEBS Lett.591, 3819–3830 (2017). [DOI] [PubMed] [Google Scholar]

[CR39] 39.Korfhage, J. L. et al. ATG2A-mediated bridge-like lipid transport regulates lipid droplet accumulation. Preprint at bioRxiv10.1101/2023.08.14.553257 (2023).

[CR40] 40.Osawa, T. et al. Atg2 mediates direct lipid transfer between membranes for autophagosome formation. Nat. Struct. Mol. Biol.26, 281–288 (2019). [DOI] [PubMed] [Google Scholar]

[CR41] 41.Hama, Y., Morishita, H. & Mizushima, N. Regulation of ER-derived membrane dynamics by the DedA domain-containing proteins VMP1 and TMEM41B. EMBO Rep.23, e53894 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR42] 42.Tábara, L.-C. & Escalante, R. VMP1 establishes ER-microdomains that regulate membrane contact sites and autophagy. PLoS ONE11, e0166499 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR43] 43.McEwan, D. G. & Ryan, K. M. ATG2 and VPS13 proteins: molecular highways transporting lipids to drive membrane expansion and organelle communication. FEBS J.289, 7113–7127 (2022). [DOI] [PubMed] [Google Scholar]

[CR44] 44.Maeda, S. et al. Structure, lipid scrambling activity and role in autophagosome formation of ATG9A. Nat. Struct. Mol. Biol.27, 1194–1201 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR45] 45.Matoba, K. et al. Atg9 is a lipid scramblase that mediates autophagosomal membrane expansion. Nat. Struct. Mol. Biol.27, 1185–1193 (2020). [DOI] [PubMed] [Google Scholar]

[CR46] 46.Chorlay, A., Forêt, L. & Thiam, A. R. Origin of gradients in lipid density and surface tension between connected lipid droplet and bilayer. Biophys. J.120, 5491–5503 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR47] 47.Mailler, E. et al. The autophagy protein ATG9A enables lipid mobilization from lipid droplets. Nat. Commun.12, 6750 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR48] 48.Han, G.-S. & Carman, G. M. Characterization of the human LPIN1-encoded phosphatidate phosphatase isoforms. J. Biol. Chem.285, 14628–14638 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR49] 49.Santinho, A., Carpentier, M., Lopes Sampaio, J., Omrane, M. & Thiam, A. R. Giant organelle vesicles to uncover intracellular membrane mechanics and plasticity. Nat. Commun.15, 3767 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR50] 50.King, C., Sengupta, P., Seo, A. Y. & Lippincott-Schwartz, J. ER membranes exhibit phase behavior at sites of organelle contact. Proc. Natl Acad. Sci. USA117, 7225–7235 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR51] 51.Jaiswal, A., Hoerth, C. H., Pereira, A. M. Z. & Lorenz, H. Improved spatial resolution by induced live cell and organelle swelling in hypotonic solutions. Sci. Rep.9, 12911 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR52] 52.François-Martin, C. & Pincet, F. Actual fusion efficiency in the lipid mixing assay—comparison between nanodiscs and liposomes. Sci. Rep.7, 43860 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR53] 53.Sarkar, P. et al. Environment-sensitive fluorescence of 7-nitrobenz-2-oxa-1, 3-diazol-4-yl (NBD)-labeled ligands for serotonin receptors. Molecules26, 3848 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR54] 54.Srinivasan, S., Álvarez, D., John Peter, A. T. & Vanni, S. Unbiased MD simulations identify lipid binding sites in lipid transfer proteins. J. Cell Biol.223, e202312055 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR55] 55.Chorlay, A. & Thiam, A. R. Neutral lipids regulate amphipathic helix affinity for model lipid droplets. J. Cell Biol.219, e201907099 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR56] 56.Caillon, L. et al. Triacylglycerols sequester monotopic membrane proteins to lipid droplets. Nat. Commun.11, 3944 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR57] 57.Damm, A. et al. Steric repulsion counteracts ER–to–lipid droplet protein movement. Sci. Adv.11, eadu6998 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR58] 58.Chorlay, A., Santinho, A. & Thiam, A. R. Making droplet-embedded vesicles to model cellular lipid droplets. STAR Protoc.1, 100116 (2020). [DOI] [PMC free article] [PubMed]

[CR59] 59.Chorlay, A. & Thiam, A. R. An asymmetry in monolayer tension regulates lipid droplet budding direction. Biophys. J.114, 631–640 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR60] 60.Kory, N., Thiam, A.-R., Farese, R. V. Jr & Walther, T. C. Protein crowding is a determinant of lipid droplet protein composition. Dev. Cell34, 351–363 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR61] 61.Ajjaji, D. et al. Dual binding motifs underpin the hierarchical association of perilipins1–3 with lipid droplets. Mol. Biol. Cell30, 703–716 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR62] 62.Chung, J. et al. LDAF1 and seipin form a lipid droplet assembly complex. Dev. Cell51, 551–563 (2019). [DOI] [PMC free article] [PubMed]

[CR63] 63.Mizrak, A. et al. Single-molecule analysis of protein targeting from the endoplasmic reticulum to lipid droplets. Preprint at bioRxiv10.1101/2024.08.27.610018 (2024).

[CR64] 64.Szymanski, K. M. et al. The lipodystrophy protein seipin is found at endoplasmic reticulum lipid droplet junctions and is important for droplet morphology. Proc. Natl Acad. Sci. USA104, 20890–20895 (2007). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR65] 65.Wang, S. et al. Seipin and the membrane-shaping protein Pex30 cooperate in organelle budding from the endoplasmic reticulum. Nat. Commun.9, 2939 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR66] 66.Salo, V. T. et al. Seipin regulates ER–lipid droplet contacts and cargo delivery. EMBO J.35, 2699–2716 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR67] 67.Santinho, A., Chorlay, A., Foret, L. & Thiam, A. R. Fat inclusions strongly alter membrane mechanics. Biophys. J.120, 607–617 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR68] 68.Santinho, A. et al. Membrane curvature catalyzes lipid droplet assembly. Curr. Biol.30, 2481–2494 (2020). [DOI] [PubMed] [Google Scholar]

[CR69] 69.Ganesan, S., Shabits, B. N. & Zaremberg, V. Tracking diacylglycerol and phosphatidic acid pools in budding yeast. Lipid Insights8s1, LPI.S31781 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR70] 70.Zhukovsky, M. A., Filograna, A., Luini, A., Corda, D. & Valente, C. Phosphatidic acid in membrane rearrangements. FEBS Lett.593, 2428–2451 (2019). [DOI] [PubMed] [Google Scholar]

[CR71] 71.Van Der Spoel, D. et al. GROMACS: fast, flexible, and free. J. Comput. Chem.26, 1701–1718 (2005). [DOI] [PubMed] [Google Scholar]

[CR72] 72.Marrink, S. J., Risselada, H. J., Yefimov, S., Tieleman, D. P. & De Vries, A. H. The MARTINI force field: coarse grained model for biomolecular simulations. J. Phys. Chem. B111, 7812–7824 (2007). [DOI] [PubMed] [Google Scholar]

[CR73] 73.Humphrey, W., Dalke, A. & Schulten, K. VMD: visual molecular dynamics. J. Mol. Graph.14, 33–38 (1996). [DOI] [PubMed] [Google Scholar]

[CR74] 74.de Jong, D. H. et al. Improved parameters for the martini coarse-grained protein force field. J. Chem. Theory Comput.9, 687–697 (2012). [DOI] [PubMed] [Google Scholar]

[CR75] 75.Bussi, G., Donadio, D. & Parrinello, M. Canonical sampling through velocity rescaling. J. Chem. Phys.126, 014101 (2007). [DOI] [PubMed] [Google Scholar]

[CR76] 76.Parrinello, M. & Rahman, A. Polymorphic transitions in single crystals: a new molecular dynamics method. J. Appl. Phys.52, 7182–7190 (1981). [Google Scholar]

[CR77] 77.de Jong, D. H., Baoukina, S., Ingólfsson, H. I. & Marrink, S. J. Martini straight: boosting performance using a shorter cutoff and GPUs. Comput. Phys. Commun.199, 1–7 (2016). [Google Scholar]

[CR78] 78.Slot, J. W. & Geuze, H. J. Cryosectioning and immunolabeling. Nat. Protoc.2, 2480–2491 (2007). [DOI] [PubMed] [Google Scholar]

[CR79] 79.Peterson, C. W., Deol, K. K., To, M. & Olzmann, J. A. Optimized protocol for the identification of lipid droplet proteomes using proximity labeling proteomics in cultured human cells. STAR Protoc.2, 100579 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

ATG2A-mediated DAG transfer recruits DGAT2 for lipid droplet growth

Helin Elhan

Alicia Damm

Justin L Korfhage

Daniel Álvarez

Mehdi Zouiouich

Francesca Giordano

Stefano Vanni

Thomas J Melia

Abdou Rachid Thiam

Abstract

Main

Results

ATG2A promotes LD growth during lipid storage

Fig. 1. ATG2A facilitates lipid incorporation into pre-existing LDs and limits de novo LD nucleation.

Extended Data Fig. 1. ATG2A regulates LD size and number.

ATG2A promotes lipid incorporation into LDs and limits LD nucleation

Extended Data Fig. 2. ATG2A enhances fatty acid incorporation into lipid droplets.

ATG2A mediates DAG delivery to the LD core

Fig. 2. ATG2A promotes lipid incorporation inside LDs under DGAT inhibition.

Extended Data Fig. 3. DGAT and Lipin-1 inhibition reveals ATG2A-dependent lipid incorporation into LDs.

PA is metabolized and stored in LD via an ATG2A-dependent pathway

Fig. 3. PA is metabolized and stored in LDs via an ATG2A-dependent pathway.

Extended Data Fig. 4. ATG2A promotes PA enrichment in lipid droplets and localizes preferentially at ER-LD contact sites.

ATG2A is at ER–LD contact sites and can transport DAG, TAG and PA in vitro

Fig. 4. ATG2A transports DAG.

Extended Data Fig. 5. ATG2A mediates DAG and PA transfer to artificial lipid droplets in vitro.

DGAT2 localization to LDs requires ATG2 and DAG

Fig. 5. DGAT2 localization on LDs requires ATG2 and DAG.

Extended Data Fig. 6. DGAT2 recruitment to artificial lipid droplets depends on DAG content.

Moderate ATG2A recruitment to LDs is key for DGAT2 relocalization and LD growth

Fig. 6. DGAT2 relocalization to lipid droplets depends on ATG2A levels and activity.

Extended Data Fig. 7. ATG2A regulates DGAT2 recruitment and LD size.

Discussion

Methods

Cell culture

Plasmids and cell transfection

OA and fluorescent fatty acid treatment

Fluorescent probes

Incorporation of labeled lipids

Chemical inhibition of DGAT and Lipin enzymes

In vitro DAG FRET transfer assay

NBD-DAG transfer assay

PA transfer assay

Fluorescence transfer assay with liposomes and emulsion droplets

ATG2A transport-deficient mutations

Molecular dynamics simulations and visualization tools

Immunogold labeling

Mass spectrometry analysis

LD purification

TLC

GERV production and extraction

DEGERV production

FRAP experiments

Confocal microscopy imaging

Quantification of LDs size and number

Quantification of fluorescent lipid and protein signals on LD

Statistics and reproducibility

Reporting summary

Online content

Supplementary information

Source data

Acknowledgements

Extended data

Author contributions

Peer review

Peer review information

Data availability

Competing interests

Footnotes

Extended data

Supplementary information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES