Lipid droplet and early autophagosomal membrane targeting of Atg2A and Atg14L in human tumor cells.

Autophagy is a lysosomal bulk degradation pathway for cytoplasmic cargo, such as long-lived proteins, lipids, and organelles. Induced upon nutrient starvation, autophagic degradation is accomplished by the concerted actions of autophagy-related (ATG) proteins. Here we demonstrate that two ATGs, human Atg2A and Atg14L, colocalize at cytoplasmic lipid droplets (LDs) and are functionally involved in controlling the number and size of LDs in human tumor cell lines. We show that Atg2A is targeted to cytoplasmic ADRP-positive LDs that migrate bidirectionally along microtubules. The LD localization of Atg2A was found to be independent of the autophagic status. Further, Atg2A colocalized with Atg14L under nutrient-rich conditions when autophagy was not induced. Upon nutrient starvation and dependent on phosphatidylinositol 3-phosphate [PtdIns(3)P] generation, both Atg2A and Atg14L were also specifically targeted to endoplasmic reticulum-associated early autophagosomal membranes, marked by the PtdIns(3)P effectors double-FYVE containing protein 1 (DFCP1) and WD-repeat protein interacting with phosphoinositides 1 (WIPI-1), both of which function at the onset of autophagy. These data provide evidence for additional roles of Atg2A and Atg14L in the formation of early autophagosomal membranes and also in lipid metabolism.

ER ( 12 ). In this context, it was suggested that LD biogenesis is connected to phagophore formation, likely involving unknown adaptor proteins that facilitate such functional interactions ( 8 ).
Here we provide evidence that both human Atg2A and Atg14L colocalize at LDs under conditions when autophagy is not induced above constitutive basal level. Upon nutrient starvation and dependent on PtdIns(3)P production, both Atg14L and Atg2A also localize to early autophagosomal membranes, decorated with the PtdIns(3)P effectors DFCP1 and WIPI-1. Interestingly, our high-content imaging analysis further showed that the numbers of cellular LDs increase upon nutrient starvation. In this context, we discuss the function of ATGs in both LD and autophagosome biogenesis.
Yeast Atg2 was identifi ed through functional screening and shown to play an essential role in both autophagy and the cytoplasm-to-vacuole pathway ( 40,41 ). Subsequently, it was demonstrated that yeast Atg2 is targeted via Atg18, the ancestral human WIPI-1/-2 protein, to the phagophore ( 42,43 ). Recently, it was shown that yeast Atg2 can be targeted to the phagophore also in the absence of Atg18, but dependent on PtdIns(3)P ( 44 ). Human Atg2 proteins are essential for autophagosome formation ( 45 ), suggesting that WIPI proteins and Atg2 share overlapping functions at the phagophore.
Although the membrane origin of autophagosomes is still unclear, landmark experiments showed that phagophore formation initiates from discrete ER regions referred to as cradle ( 46,47 ) where several ATGs colocalize ( 48 ). Interestingly, analogous ER regions have also been postulated to facilitate lipid transfer between LDs and the subjected to automated image analysis using the In Cell Analyzer 1000 Workstation 3.4 software. The number of LDs per cell was determined by using different parameters for nuclei, cell, and inclusions. The cells were recognized by the nuclei (DAPI channel) and GFP channel. The characteristic cell area was set to 800 µm 2 (G361) or 1,500 µm 2 (HeLa), with a sensitivity of 25. For the detection of inclusions within the recognized cells, only inclusions with a size between 0.5 µm and 5 µm were counted [sensitivity 25 (G361) or 40 (HeLa)]. Additionally, HeLa cells with an inclusion intensity of <600 were excluded from the analysis.

Live-cell video microscopy
Live-cell imaging was conducted as described earlier ( 50 ), and media supplemented with 11.4 mM ascorbic acid to reduce phototoxicity. Movies with fi ve images per second were generated. ImageJ software with the MTrackJ plug-in was used for still image representation and to calculate migration distances of selected structures.

EM
For freeze-fracture immune-EM, unfi xed stably transfected GFP-Atg2A U2OS cells were scraped from the culture vessels, centrifuged to remove excess medium, and recentrifuged briefl y (<30 s) in 30% glycerol. Cells were mounted in 30% glycerol on gold-nickel alloy carriers and immediately rapidly frozen in Freon 22 cooled with liquid nitrogen. The samples were fractured in a BA310 freeze-fracture unit (Balzers AG) at -100°C. Replicas of the fractured cells were immediately made by electron beam evaporation of platinum-carbon and carbon at angles of 38° and 90° and to thicknesses of 2 and 20 nm, respectively. The replicas were incubated overnight in 5% SDS to remove cellular material except for those molecules adhering directly to the replicas ( 51 ). They were then washed in distilled water and incubated briefl y in 5% BSA before immunolabeling. Freeze-fracture replicas of the cells were immunogold labeled with primary rabbit polyclonal antibodies raised against the entire sequence of GFP (ab290, Abcam). The secondary antibodies used were goat anti-rabbit antibodies coupled to 18 nm colloidal gold (conjugates from Jackson ImmunoResearch) ( 52 ). Control specimens were prepared without the primary antibodies. Examination of the immunogoldlabeled freeze-fracture replicas was carried out using a Philips 410 transmission electron microscope. Observations on freezefracture immunogold replicas were based on examination of >200 cells from three separate experiments.
For standard EM, subconfl uent G361 cells were treated with 500 µM OA for 24 h and fi xed in 4% paraformaldehyde (sc-281692, Santa Cruz) and further in 2% glutaraldehyde and 0.5% osmium tetroxide in 0.1 M PBS. Subsequently, fi xed cells were dehydrated with ethanol, and embedded in Epon as previously described ( 53 ). Thin sections were (ultramicrotome) contrasted with uranyl acetate and lead citrate and examined using an EM410 electron microscope (Philipis) and documented digitally (Ditabis).
The transfection mixture was incubated for 20 min and combined with 100 µl DMEM/10% FCS containing 1× 10 4 HeLa cells or 3× 10 4 G361 cells in 96-well plates. Forty-eight hours after transfection, cells were subjected to starvation treatments and high-content LD analysis.

Antibodies and fl uorescent dyes
The following primary antibodies were used in this study: antitubulin (

Confocal laser scanning microscopy
Immunostaining and confocal laser scanning microscopy (LSM) were previously described ( 34 ). For the visualization of LDs cells were incubated with HCS LipidTOX Green or HCS LipidTOX Red neutral lipid stain (1:1,000) for 30 min at room temperature . For quantitative colocalization analysis, image projections from confocal LSM sections (in distances of 0.5 µm) were acquired with identical laser intensities and detector gains. Subsequently, images were background subtracted, and ADRP LD signals were thresholded and analyzed for individual cells using the ImageJ colocalization threshold plug-in. Alternatively, images were analyzed by using Image Pro Plus software (Media Cybernetics). Using Volocity 3.1 (Improvision), individual confocal LSM sections (in distances of 0.2 µm) were applied for 3D reconstruction and fl y-through movie presentations.
By confocal LSM of transiently expressed GFP-Atg2A the vesicular localization of Atg2A in the cytoplasm was apparent ( Fig. 1B ). This result was verifi ed by live-cell microscopy of GFP-Atg2/U2OS cells (supplementary Fig. IA; supplementary Videos I-III) and by confocal LSM of myc -Atg2A (supplementary Fig. IB) or Atg2A-myc (data not shown). Because yeast Atg2 was identifi ed as an essential protein of the autophagic machinery, we conducted treatments that modulate the autophagic activity throughout this study, along with appropriate controls. In general, we used NF medium or RM to induce autophagy above basal level, or WM to inhibit the formation of autophagosomes by inhibiting PtdIns (3)

Human Atg2A localizes to cytoplasmic LDs
In order to defi ne the identity of vesicular Atg2A structures, we conducted colocalization studies with a variety of cellular markers (e.g., lysosomes, mitochondria, or FYVEpositive endosomes; data not shown) and found that Atg2A specifi cally localizes to LDs. In detail, by differential interference contrast microscopy we found that LDs colocalized with myc -Atg2A vesicles (supplementary Fig. ID). This result was further confi rmed by visualizing transiently expressed myc -Atg2A in U2OS cells treated with LipidTOX Green, which labels neutral lipids (supplementary Clearly, myc -Atg2A was expressed at the surface of LDs as fl uorescence intensity profi ling showed maximal myc -Atg2A signal intensity ( Fig. 1D , in red) at the cytoplasmic face of LDs ( Fig. 1D , in green). Localization of myc -Atg2A at the cytoplasmic face of LDs labeled with LipidTOX Green is further highlighted by 3D reconstruction and fl y-through movie presentations using confocal LSM sections (supplementary Video V). By freeze-fracture immuno-EM of the GFP-Atg2A/U2OS cell line with anti-GFP antibodies, we were unable to detect

Sucrose density centrifugation
LD purifi cation was performed as described previously ( 56 ). Up to 1.

Bioinformatics analyses
Cluster analysis: Human Atg2A protein (gi:239047271) was used for a Basic Local Alignment Search Tool search against the NCBI nonredundant protein database 'nr' (version November 2012). All sequences producing High Scoring Segment-Pairs (HSPs) up to E-values of 10 were extracted as full-length sequences (813 sequences) and analyzed in Cluster Analysis of Sequences (CLANS) ( 57 ). The sequence similarity groups identifi ed in the CLANS map were used as a basis to identify all sequences of the Atg2 protein family and select the animal as well as suitable outgroup sequences from which to infer a phylogeny.
Phylogenetic inference: All sequences from the animal Atg2 sequence similarity groups, as well as the selected outgroup sequences, were combined into one fi le and aligned using Multiple Sequence Comparison by Log-Expectation (MUSCLE). Fragment or truncated sequences present in the alignment were manually identifi ed and removed prior to phylogenetic inference. The phylogeny was inferred based on the neighbor-joining approach (1,000 bootstrap replicates) using the ASATURA software, the JTT substitution matrix, and no mutational saturation cutoff .

Human Atg2A is expressed in vesicular structures that migrate bidirectionally along microtubules
We generated plasmids for transient and stable expression of N-terminal tagged GFP-or myc -Atg2A and C-terminal tagged Atg2A-myc in human tumor cell lines (U2OS, G361, and HeLa). In addition, human U2OS cells derived from G418-selected clones that stably express GFP-Atg2A protein at a low level (GFP-Atg2A/U2OS hereafter), were also generated for our studies. By Western blotting of Further, fl uorescence intensity profi ling using confocal LSM identifi ed a profound colocalization of endogenous adipophilin (ADRP), a bona fi de LD marker ( 58 ), with vesicular GFP-Atg2A (supplementary Fig. IIIA). This result is analyzed in more detail subsequently.

ADRP-positive LDs are decorated with GFP-Atg2A
Upon OA treatment, the amount of ADRP protein increases as well as the number and size of ADRP-positive LDs ( 59 ). We conducted quantitative confocal LSM visualizing GFP-Atg2A and endogenous ADRP. We used GFP-Atg2Atransfected G361 cells and treated the cells with or without OA for 24 h. Subsequently, the cells were treated for 3 h with nutrient-rich CM, NF medium, or WM. No differences in colocalization patterns of GFP-Atg2A and endogenous ADRP regarding autophagy-modulating conditions were observed (data not shown), and nutrient starvation results (NF, OA + NF) are presented ( Figs. 2 and 3 ; supplementary Fig. III). Clearly, GFP-Atg2A prominently colocalized with endogenous ADRP at LDs, observed as typical ring-like structures ( Fig. 2 ). By comparing confocal LSM images, the amount of ADRP-positive LDs decorated with GFP-Atg2A seemed to increase upon OA treatment (OA + NF) ( Fig. 2A, B ). As expected, OA-mediated stimulation of LD formation increased the number of ADRP-positive LDs (OA + NF) when compared with conditions without OA (NF), as the fl uorescence area of endogenous ADRP signifi cantly increased from 1.7% (NF) to 8.4% (OA + NF) in relation to the total cell area ( Fig. 3A , right panels). The fl uorescence GFP-Atg2A area also signifi cantly increased upon OA treatment from 2.3% (NF) to 8.8% (OA + NF) ( Fig. 3A , left panels). As GFP-Atg2A colocalized with ADRP in both the absence and presence of OA (OA + NF) ( Fig. 2 ; supplementary Fig. III), we quantifi ed the colocalization of GFP-Atg2A and ADRP. To measure the overall association of GFP-Atg2A and endogenous ADRP, we measured the Pearson's correlation coeffi cient (PCC) for individual cells (n = 34 cells per treatment; PCC values + 1 = 100% colocalization) using the ImageJ colocalization threshold tool. In the absence of OA and under conditions that induce autophagy (NF), the PCC value was 0.33 and signifi cantly increased to 0.49 when autophagy was induced upon the stimulation of LD formation by OA for 24 h (OA + NF) ( Fig. 3B ). The Mander's overlap coeffi cient (MOC) for GFP-Atg2A (0.036) overlapping with ADRP also signifi cantly increased when LD formation was stimulated (0.119) ( Fig. 3C ). Such increase in overlap was not observed for ADRP overlapping GFP-Atg2A ( Fig. 3D ) and reached an MOC of 0.92 before and 0.94 after OA treatment. This result argues that i ) ADRP-positive LDs are prominently decorated with GFP-Atg2A and that ii ) GFP-Atg2A localizes not exclusively to the LD compartment. Consistent with both this result and expectation, we also detected GFP-Atg2A in the autophagic compartment as detailed subsequently.

Independent of the autophagic status, Atg2A colocalized with Atg14L at LDs
Next we conducted colocalization studies for Atg2A with ATGs functioning during the initiation of autophagosome GFP-Atg2A at LDs (supplementary Fig. IIC), demonstrating that Atg2A does not become a membrane protein of the LD lipid monolayer but associates with LDs as a peripheral membrane protein. however, signifi cantly increased the number of GFP-Atg14L puncta-positive cells to an average of 63% ( Fig. 4B , left panels).
On this basis, we transiently coexpressed GFP-Atg14L and myc -Atg2A in U2OS cells and treated the cells with nutrient-rich CM, NF medium, or WM. We found that both ATGs colocalized at vesicular structures that were present irrespective of the autophagic status, indicating that these structures represent the colocalization of myc -Atg2A and GFP-Atg14L at LDs ( Fig. 4C ). In support, fl uorescence intensity profi ling of GFP-Atg14L-expressing cells incubated with LipidTOX Red demonstrated that in part GFP-Atg14L localizes at the cytoplasmic face of LDs ( Fig. 4D ), and this result was further highlighted by 3D reconstruction and fl ythrough movie presentation (supplementary Video VI). Of note, although vesicular GFP-Atg14L ( Fig. 4B ) and myc -Atg2A structures (supplementary Fig. IVA) increased upon autophagy induction, colocalization of GFP-Atg14L and myc -Atg2A was not signifi cantly changed (68% to 77% of GFP-Atg14L colocalized with myc -Atg2A, whereas 52% to 63% of myc -Atg2A colocalized with GFP-Atg14L) (supplementary Fig. IVB, C). This indicates that Atg2A and Atg14L localize in part to the autophagic compartment upon autophagy induction and in addition to the LD compartment irrespective of the autophagic activity.

Detection of GFP-Atg2A and GFP-Atg14L in the ADRPpositive LD fraction separated by density gradient centrifugation
We conducted cell fractionation upon density centrifugation prepared from GFP-Atg2A/U2OS cells transiently transfected with myc -Atg14L. As expected, endogenous ADRP was exclusively present in the LD fraction. GFP-Atg2A was also detected in this fraction of fl oating ADRPpositive LDs; however, GFP-Atg2A was more prominently detected in cytoplasmic fractions. To a minor extent, we also detected myc -Atg14L, while the GAPDH control was restricted to cytoplasmic fractions ( Fig. 5 ).

Nutrient starvation leads to an increase of LDs in the human tumor cell lines G361 and U2OS
The previously discussed results demonstrated that Atg2A and Atg14L colocalize at LDs irrespective of the autophagic status of the cell, but that a subpopulation of both ATGs is targeted to early autophagosomal membranes enriched in PtdIns(3)P, and positive for DFCP1 and WIPI-1, upon the induction of autophagy. As we detailed our analysis using conditions when autophagy was  for prolonged energy shortage at a later stage. This phenomenon might even be evolutionarily conserved as Caenorhabditis elegans accumulates LDs upon dauer formation ( 62,63 ). Starvation-induced nonspecifi c degradation of proteins and organelles can provide acetyl-CoA and also fatty acids and cholesterol required for neutral lipid synthesis. Also, starvation is known to trigger autophagosomal degradation of glycogen stores ( 64 ), which could liberate additional energy for lipid synthesis. Furthermore, our starvation treatments include lipoprotein depletion, which LDs per cell was found to increase signifi cantly after 3 h (U2OS and G361) and after 24 h (all cell lines) ( Fig. 8A ). To functionally address the role of Atg2A and Atg14L in the biogenesis of LDs, we performed siRNA-mediated depletion of Atg2A or Atg14L in HeLa (supplementary Fig. VIA) and in G361 (supplementary Fig. VIB) cells. We assessed the number of LDs upon treatment with CM or NF medium using high-content image analysis. Atg2A depletion in HeLa cells followed by nutrient starvation using NF medium resulted in a signifi cant increase of LD numbers in individual cells ( Fig. 8B , left panels). Atg14L depletion signifi cantly increased the number of LDs in i ) HeLa cells in CM as well as upon nutrient starvation using NF medium ( Fig. 8B , left panels) and ii ) G361 cells in CM ( Fig. 8B , right panels). In G361 cells, the size of LDs signifi cantly increased upon Atg14L depletion and treatments using CM as well as NF medium (supplementary Fig. VID). From this, we conclude that Atg2A and Atg14L have dual functions, acting in the regulation of autophagosome formation as well as contributing to LD metabolism ( Fig. 8C ).

DISCUSSION
In Saccharomyces cerevisiae , PtdIns(3)P-dependent recruitment of Atg2 to autophagosome formation sites has been demonstrated ( 43,44 ). Similarly, PtdIns(3)P-dependent targeting of human Atg2A to WIPI-1 positive phagophores upon starvation-induced autophagy was also found in this study, and WM treatment abrogated this specifi c recruitment of Atg2A. In addition we demonstrate that Atg2A is targeted to the surface of ADRP-positive LDs that move bidirectionally on microtubules, independent of autophagic status. Further, we identifi ed Atg14L as a novel LD-associated protein, colocalizing there with Atg2A irrespective of autophagic status. Evidence has been provided that Atg14L resides at the ER and regulates localized PtdIns(3)P generation, which is then bound by the WIPIs. Apart from being the site of phagophore formation, LDs are also proposed to form at the ER ( 60 ), and thus sites of autophagosomal membrane biogenesis and LD formation seem to be functionally connected. LDs have further been detected in very close association with the forming phagophore ( 47 ).
So far, three ATGs have been found to be targeted to LDs, LC3 ( 61 ), Atg2A ( 45 ) (this study), and Atg14L (this study), indicating that these ATGs either display dual functions in autophagy and LD biogenesis or that they functionally connect both pathways. Therefore, by high-content analysis, we quantifi ed the number of LDs in human HeLa, G361, and U2OS tumor cells upon short-term nutrient starvation and detected a prominent increase in the number and size of LDs per individual cell, resembling LD formation. This indicates that autophagy induction and LD formation can simultaneously occur in starved cells. Hence starvation-induced autophagy might provide energy to boost lipid storage, preparing the cell to compensate Fig. 8. Starvation treatment increases LD formation. HeLa, G361, and U2OS cells were treated with CM or NF medium for 3 h, fi xed, and stained with DAPI (blue) and LipidTOX Green (green). Automated image acquisition and analysis was conducted using a high-content imaging platform (In Cell Analyzer 1000). A: Up to 9,305 cells were quantifi ed (n = 3-5) to assess the number of LDs per individual cell (left panel). Representative images for G361 cells are presented (right panel). Scale bar, 20 µm. B: HeLa and G361 cells were transiently transfected with scrambled control siRNA (siControl), siRNA targeting Atg2A (siAtg2A), or Atg14L (siAtg14L), and downregulation of Atg2A and Atg14L mRNA was verifi ed by quantitative PCR (supplementary Fig. VIA,  B). Cells transfected with siRNA were incubated with CM or NF medium for 3 h and fi xed, and LD abundance was quantifi ed by high-content analysis (n = 3-4). Mean ± SD. P values: n.s. (not signifi cant) P у 0.05, * P < 0.05, ** P р 0.01, *** P р 0.001. Supporting material is available (supplementary Fig. VIC, D). C: A model for the roles of human Atg2A and Atg14L in LD biogenesis and autophagy. could decrease ER cholesterol levels resulting in sterol regulatory element binding protein (SREBP)-mediated stimulation of lipid synthesis ( 65 ). However, autophagy was also found to degrade LDs, thereby contributing to the regulation of lipid metabolism in liver cells ( 10 , 61 ) and adipocyte differentiation ( 66 ), suggesting that the function of autophagosomal proteins at LDs and in adipogenesis is complex.
Whereas Atg2A and Atg14L are targeted to LDs independent of the autophagic status, both proteins localize to DFCP1 and WIPI-1-positive omegasome and phagophore membranes, distinct to LD structures, upon autophagy stimulation. In addition, we could demonstrate that Atg2A and Atg14L depletion resulted in increased LD abundance upon treatment with CM (Atg14L depletion) or NF medium (Atg2A depletion and Atg14L depletion). Therefore, our study provides evidence that Atg2A as well as Atg14L fulfi ll at least two functions in human cells: one exerted as a peripheral membrane protein at the surface of ADRP-positive LDs and the other involving the regulation of autophagy, dependent on PtdIns(3)P generation ( Fig. 8C ). It is possible that LD-localized Atg2A and Atg14L impair excessive lipid storage inside LDs and ensure effi cient energy use (e.g., generated by autophagy induction). Another function of Atg2A and Atg14L at the LD surface might involve modulation of the fatty acid and cholesterol esterifi cation/hydrolysis cycle, contributing to the regulation of lipid storage. Our results are in line with recent work on mammalian Atg2, shown to play an essential role in autophagosome formation and to also regulate LD size and distribution ( 45 ). However, it is also possible that ATGs, such as Atg2A and Atg14L, fulfi ll distinct roles at LDs and autophagosomal membranes, but that they also mediate a functional interconnection during simultaneous formation of LDs and autophagsomes, maybe by mediating intermembrane lipid traffi c.
Because Atg2 shares protein sequence homologies with Vps13 (supplementary Fig. VII), a protein shown to promote endosome/trans-Golgi network (TGN) membrane protein cycling ( 67 ), it might be plausible that Vps13 and Atg2 proteins generally function in membrane formation and morphology, and that distinct members provide specifi city for particular vesicle subsets. In this context, it has been proposed that the yeast Atg18/Atg2 complex is required to generate negative curvature at the forming autophagosome or fulfi lls an essential function at the elongating tips of the phagophore ( 43 ). Likewise it was recently found that Vps13 regulates membrane bending and promotes the expansion of the prospore membrane in yeast ( 68 ). From this, one would assume that malfunctions of either of the human Atg2/Vps13 proteins should be correlated to distinct human pathologies with defects in intracellular membrane biogenesis. Indeed, hereditary mutations of human Vps13 orthologs cause the diseases chorea acanthocytosis and Cohen syndrome ( 69,70 ). As Atg2 functions in LD and autophagosome biogenesis, it can be assumed that functional or genetic alterations of human