Phospholipase D and phosphatidic acid in the biogenesis and cargo loading of extracellular vesicles

Extracellular vesicles released by viable cells (exosomes and microvesicles) have emerged as important organelles supporting cell-cell communication. Because of their potential therapeutic significance, important efforts are being made toward characterizing the contents of these vesicles and the mechanisms that govern their biogenesis. It has been recently demonstrated that the lipid modifying enzyme, phospholipase D (PLD)2, is involved in exosome production and acts downstream of the small GTPase, ARF6. This review aims to recapitulate our current knowledge of the role of PLD2 and its product, phosphatidic acid, in the biogenesis of exosomes and to propose hypotheses for further investigation of a possible central role of these molecules in the biology of these organelles.

ruffles (12,13). In spite of the attempts of many groups to identify the PLD binding site for ARF (14,15), to date, evidence for direct PLD-ARF interaction is missing. PLD1 and PLD2 are known to differentially localize in the cell, possibly through specific posttranslational modifications. PLD1 (16) and PLD2 (17) can be palmitoylated on two cysteine residues in their PH domain, facilitating the sorting of these isoforms into different compartments. At steadystate, PLD1 preferentially localizes to the perinuclear area, potentially associated with the endoplasmic reticulum, the Golgi apparatus, and/or late endosomes (18,19). PLD2 is most often reported to be associated with the plasma membrane (6,18), but also localizes in the cytosolic cell interior (12) and vesicular compartments (20). PLD2 has been shown to participate in the biogenesis of extracellular vesicles (EVs)/exosomes in two different cell types (21,22).

EVs, SIGNALING ORGANELLES UNDER THE CONTROL OF PLD2
EVs are limited by a lipid bilayer, and can be released by any type of cell. EVs have the same topology as the cell, and they contain cytosolic components and various types of nucleic acids. They also carry transmembrane proteins and lipids. In the last years, it became clear that EVs are true organelles, supporting cell-to-cell communication over short or long distances (23,24). Moreover, the concept of "disease EVs" was validated in various pathological contexts, like cancer, neurodegeneration, and cardiovascular disease (25). These "bad EVs" can transport pathogenic molecules (including proteins, lipids, and nucleic acids) and participate in disease progression. In cancer, for example, bad EVs promote carcinogenesis, tumor growth, and angiogenesis, suppress immune response, mold the premetastatic niche, and impair response to therapy. According to the current nomenclature, one distinguishes three types of EVs, namely exosomes, microvesicles, and apoptotic bodies. Microvesicles originate from direct budding at the cell surface and their diameter can vary between 50 and 1,000 nm. Apoptotic bodies result from cell fragmentation and can reach a 5,000 nm in diameter. Apoptosis is not covered in this review. Exosomes find their origin in endosomal membranes. They originate from intraluminal vesicles (ILVs) that accumulate inside late endosomes and multivesicular bodies (MVBs). MVB fusion with the plasma membrane allows ILV secretion in the extracellular environment, where these small vesicles are then designated as "exosomes". Their size can vary from 40 to 150-200 nm. The term exosome is abusively used in the literature. It  (18). They contain two conserved HKD domains necessary for their enzymatic activity; a PH domain, which binds phosphoinositides and anionic lipids (80) and also serves as docking site for recruitment of SH2-domain-containing proteins like the growth factor receptor bound protein 2 (Grb2) (58); a phosphoinositide binding phox domain (PX) involved in targeting proteins to cell membranes; and a putative polybasic PIP 2 -binding site, located between the catalytic HKD motifs (1,15). The other four isoenzymes lack PX and PH domains and form a group of "non-classical PLDs". PLD3 (81) and PLD4 (82) contain the canonical pair of HKD domains; although to our knowledge, no catalytic activity has been demonstrated for these isoenzymes. They are localized in the endoplasmic reticulum (ER) where they are anchored by an N-terminal transmembrane domain and the catalytic domains are oriented in the ER lumen (83). PLD6 is anchored by an N-terminal transmembrane domain into the outer surface of mitochondria and hydrolyzes cardiolipin to generate PA (84). There is no publication yet concerning PLD5, but it shares similarities with PLD3 and PLD4 (85). often refers to the fraction of the secretome pelleting at 100,000 g (further referred to as "exosome-enriched EVs" in this document). Yet, this fraction can be contaminated with small vesicles that are not of exosomal origin, and complementary methods need to be used to prove the exosomal nature of EVs. Our understanding of the molecular mechanisms controlling exosome production is far from complete and some of these mechanisms appear to vary between cell types (26). Several molecules, including various specific Rabs, cortactin, and SNAREs, have been described to control the fusion of "secretory" MVBs with the plasma membrane. The endosomal sorting complex required for transport (ESCRT) was the first described mechanism for the formation of ILVs and MVBs (27,28). It consists of four multimeric complexes (ESCRT-0, -I, -II, and -III) and associated proteins (e.g., VPS4 and ALIX). ESCRTs are assembled in an orderly manner at the endosome. ESCRT-0, -I, and -II recognize and sequester ubiquitinated membrane proteins, while the ESCRT-III complex is responsible for membrane budding and abscission to form ILVs (29,30). Besides ESCRTs, neutral sphingomyelinase 2 (nSMase2) (31) and also syndecans, syntenin (32), heparanase (33), SRC (34), and ARF6 (21) were shown to participate in the budding of ILVs that will significantly contribute to the pool of exosomal vesicles (Fig. 2).
In a model of human mammary carcinoma cells (MCF-7), the depletion of ARF6 or ARNO, an ARF6 GEF, leads to a decrease in several exosomal proteins, such as syntenin, ALIX, CD63, and syndecan, and affects ILV budding (21). Investigation of the ARF6 effectors implicated in this process pointed to a role of PLD2 and excluded other ARF6 effectors, such as PLD1 and also phosphatidylinositol 4-phosphate 5-kinase (PIP5K) or PIP5K, enzymes producing phosphatidylinositol 4,5-bisphosphate [PI(4,5)P 2 ] (21). It is noteworthy that it has been proposed that PI(4,5) P 2 , the product of PIP5Ks, can activate PLD2, which leads to enhanced PA formation able to activate further PIP5Ks (12,20). However, this does not seem to happen on the limiting membrane of MVBs, as the knockdown of PIP5Ks did not impact on syntenin intra-endosomal budding (21). These data are consistent with the observations that ARF is able to activate PLD in the absence of PI(4,5)P 2 (35). Basically, on endosomal membranes, ARF6-PIP5K rather supports the recycling of syntenin and associated transmembrane protein cargo, pending the coincident detection of peptide and PI(4,5)P 2 by the syntenin PDZ domains (see the pathway in Fig. 2A) (36,37). On the contrary, ARF6-PLD2 supports the budding of syntenin cargoladen ILVs inside the MVBs (see the pathway in Fig. 2B) (21). This model is in line with the observations of Laulagnier et al. (38) showing that stimulation of RBL-2H3 cells with ionomycin boosts the release of exosome-enriched EVs in a PLD2-dependent manner. Indeed, ionomycin is an ionophore that induces the hydrolysis of PIP 2 and the release of calcium to the cytoplasm. One could speculate that PIP 2 hydrolysis would impair the recycling pathway and may indirectly favor the formation of ILVs. Moreover, high levels of intra-cytosolic calcium could facilitate the MVB-plasma membrane fusion events necessary for the secretion of ILVs as exosomes (39). Interestingly, syntenin also directly interacts with PA. While ARNO appears to be the preferred ARF6 GEF for inward budding (away from the cytosol, with formation of ILVs inside the MVB) (21), the stimulator of ARF6 that supports recycling pathways (necessitating endosomal outward budding, toward the cytosol) still needs to be identified. Clearly, syntenin-ALIX interaction and ESCRT are necessary for the exosomal pathway. As ESCRT was recently shown to also stimulate inward budding (40), it would be interesting to test for the impact of ALIX and ESCRT on the syntenin recycling pathway.

KEY ROLE FOR PA PRODUCTION IN EXOSOME BIOGENESIS
PA (the product of PLD) represents the simplest phospholipid (Fig. 3). Despite its simple structure and relatively low abundance [1-4% of the total of phospholipids (41, 42)],

Fig. 2.
Mechanisms that control the recycling of syntenin cargo to the plasma membrane (A) or the sorting of cargo into ILVs (B). A: Syntenin-mediated endosomal recycling to the plasma membrane depends on the activation of ARF6 on perinuclear recycling endosomes, the recruitment/activation of the ARF6 effector, PIP5K, and the direct interaction of syntenin with PIP 2 (37,86). The molecular mechanism supporting syntenin-PIP 2 -dependent recycling is due to the direct interaction of syntenin with PIP 2 , and cargo, such as syndecans (37) or Frizzled 7 (36) receptors. B: ARF6 and PLD2 affect exosomes by controlling the budding of ILVs into MVBs (21). The loading of syntenin and of syntenin cargo, such as syndecans and CD63, in ILVs also depends on ARF6, but relies on its effector, PLD2 (21), and probably its reaction product, PA. PA is important for membrane dynamics, i.e., fission and fusion (43). The function of PA is likely in part due to its ability to induce a negative membrane curvature because of its small headgroup (forming a "cone" that might favor endosomal intraluminal budding). Moreover, PA can also directly interact with proteins (43,44). Unfortunately, no consensus amino-acid sequence defines a PA-binding site.
PA enables an electrostatic/hydrogen bond switch (45) when engaging with protein ligands and the negative charge of the PA headgroup is increased from 1 to 2 and stabilized upon formation of hydrogen bonds with lysine and arginine residues of the interacting protein (46). As commonly observed for protein-lipid interactions, the membrane lipid environment and the nature of the acyl chains can influence the recognition of PA by interacting proteins (47).
PA is implicated, for instance, in regulated exocytosis in several different cell lines, such as the secretion of von Willebrand factor from endothelial cells (48) and insulin from pancreatic -cells (49). In RBL-2H3 cells, PA controls the trafficking of glycoconjugates from the Golgi to the plasma membrane (50). Whether PA might be essential for budding in endosomes and for MVB formation has not been directly investigated. One study indicates that PA is 1.8-fold enriched in exosome-enriched EVs versus lysates from PC-3 cells. Yet PA accounts for only 0.16% of the total lipids in these organelles (51) and other studies failed to detect PA (21,38,52).
What might explain the discrete amount of PA in exosome-enriched EVs? The team of Michel Record observed that active phosphatidate phosphatase 1 (PAP1) is present in exosome-enriched EVs derived from RBL-2H3 cells (53). Upon the incubation of BODIPY-PA, 60% of the PA was hydrolyzed into diglycerides within 15 min. It might thus be that PA is rapidly degraded in these organelles. Alternatively, during ILV biogenesis, PA might flip-flop and be restricted to the abscission sites, the membrane domains with the highest curvature.
Yet, PA might be key at several stages of exosome formation because of its biophysical properties (Fig. 4A, B) and/ or due to its direct interactions with effector proteins involved in the process of ILV budding (Fig. 4C-E). Moreover, PA could be important for the transport (Fig. 4F) and plasma membrane docking (Fig. 4G) of ILV-filled MVBs.
Because of its negative curvature, PA production could stimulate the initiation of ILV budding (Fig. 4A). It is noteworthy that lipid-binding data suggest that syntenin can directly interact with PA (21). Irrespective of the curvature, syntenin interaction with PA could thus concentrate the syntenin/syndecan/CD63/ALIX complexes at the nascent bud, with ALIX-ESCRT further stimulating the budding process (Fig. 4C). PA could also support ILV-budding independently of ESCRT. Indeed, nSMase2 has also been proposed to interact with PA (54,55), and this binding activity could support nSMase2 targeting (and ceramide formation) to MVBs (54). Interestingly, PA can interact with SRC through the SH4, unique lipid binding region (ULRB), and SH3 domains, stimulating its activation by favoring the open conformation (56). In turn, in cells like MCF-7, SRC can activate PLD2 by phosphorylating its tyrosine 511 (57). Such a positive feedback loop could support sustained ILV production and would be totally consistent with the recently established role of SRC in exosome biogenesis (34). Whether the production of PA is required to initiate the feedback loop is unclear. Possibly, the scaffolding function of PLD2 might be implicated. SRC might recruit PLD2 by direct interactions. For example, the SH2 domain of SRC could interact with the two phosphotyrosine motifs (P-YxN) localized in its PX domain, as it was shown for Grb2 (58). Alternatively, the interaction could be indirect, as the PH domain of PLD2 seems necessary for coimmunoprecipitation with SRC (59) (Fig. 4D). Interestingly, there is evidence that RAS mediates the activation of PLD that is induced by SRC in vivo (60), but whether RAS might function in exosome biogenesis and cargo-loading by acting on the PLD2-PA axis is unexplored (Fig. 4). Possibly after PLD2 action, a flip-flop of PA from the endosomal membrane facing the cytosol to the luminal membrane could help the completion of the ILV bud (Fig. 4B). Indeed lipids such as PA can spontaneously move from one leaflet to the other within seconds or minutes (61).
In addition, annexins, a family of Ca 2+ -regulated phospholipid-binding proteins that can interact with PA (62), could help the fission of ILVs (Fig. 4E). Indeed, annexins have been proposed to be mediators or regulators of membrane fusion by inducing membrane curvature (63), and have been identified in EVs enriched in exosomes (64). Moreover, annexins localize at MVBs (65). The conversion of PC to PA mediated by PLD2 could even lower the Ca 2+ requirements for membrane fusion in the presence of annexins (66). Signaling proteins, such as ribosomal S6 kinases (S6Ks), are activated by PA and thereby support actin polymerization (67). Such actin polymerization, when taking place at the MVB outer membrane, could support the traffic of MVBs to the plasma membrane (Fig. 4F). It is noteworthy that PA also binds to the tyrosine kinase, Fer, and enhances its ability to phosphorylate cortactin, a protein that promotes actin polymerization (68). Recently, cortactin has been described to act as a positive regulator of exosome secretion Fig. 3. PA and PC structure. PA comprises a glycerol backbone with two fatty acids bonded to carbons 1 and 2 forming the nonpolar acyl chains of the lipid, and a phosphate group bonded to carbon 3, forming the negatively charged small polar headgroup of PA. The carbon 3 is bound to a choline group in PC giving a cylindrical shape to the molecule. (69,70). PA production on the limiting membrane of MVBs could thus help exosome release in several ways (Fig. 4G).

OTHER POSSIBLE ROLES FOR PLD2 IN EXOSOME BIOGENESIS
PLD2 could also act in the biogenesis of bis(monoacylglycero)phosphate (BMP) (also called lysobisphosphatidic acid by cell biologists) by stimulating the production of its precursor, phosphatidylglycerol (71). The BMP concentration rises from 1% to 15% of total phospholipids between early and late endosomes in BHK cells (72). BMP has not been detected elsewhere in the cell and exhibits unique pH-dependent fusogenic properties (72,73). BMP can induce the formation of multivesicular liposomes when their lumen is acidic and can interact with ALIX (73) and hsp70 (74), indicating that BMP can participate in exosome biogenesis (75).
PLD2 has been shown to function as a GEF for small GTPases, in particular RAC2 (76). It has been proposed that PDZ proteins might mediate the GEF activity of PLD2 (77). Whether PLD2 could work as a GEF for ARF6 and whether such activity might participate in exosome formation remains an open question.
Of course, direct PA-interacting proteins could also be targeted to exosomes upon PLD2 activation. Cytosolic proteins, such as protein kinase C, for example, accumulating in urinary exosomes (78) and interacting with PA (79), could be targeted to these vesicles due to PA interaction.

CONCLUSIONS AND FUTURE PERSPECTIVES
In summary, PLD2 emerges as a regulator of exosome production. The role of PLD2 in the biogenesis of exosomes can be related to its enzymatic activity and the production of PA in various ways.
First, PA could stimulate the production of ILVs because it is a cone-shaped phospholipid that induces negative membrane curvature and can favor membrane fusion and fission events. Furthermore, PA at the limiting membrane of MVBs could participate in the recognition and assembly of members of the ESCRT machinery because it directly interacts with syntenin that connects to ESCRT via ALIX. Additionally, PA might support a positive feedback loop including signaling proteins like SRC. It is noteworthy that PA, by its property to interact with a plethora of cytosolic proteins, could also work as an anchor for their incorporation into exosomes. Thus, PA might affect not only the number but also the composition of exosomes.
In conclusion, PLD2 might integrate an important number of signaling pathways that participate in ILV formation and/or the release of exosomes. Such mechanisms now need to be addressed. This new knowledge could pave the way for the development of new strategies to modulate exosome production and the biological effects of these vesicles.