Many lipid transfer proteins are found at MCSs where they facilitate the exchange of lipids between organelles. They can function as tethers or be co-opted by tethers or adaptor proteins that recruit them at specific sites thus imparting regiospecificity to the lipid transfer activity they carry out. However, not all tethers are LTPs and not all LTPs are tethers. LTPs constitute a large and structurally diverse functional class of proteins (Wong et al., 2017; Wong et al., 2019), encompassing members from several protein (super)families with their associated lipid-binding protein fold such as the Steroidogenic Acute Regulatory-lipid related Transfer (StART) proteins and their StART-like relatives (StARkin) (Alpy and Tomasetto, 2014; Wong and Levine, 2016; Horenkamp et al., 2018; Jentsch et al., 2018) or the Tubular Lipid-binding domain superfamily (TULIP) (Alva and Lupas, 2016; Wong and Levine, 2017) that includes the subgroup of Synaptotagmin-like Mitochondrial lipid-binding Protein domain (SMP) (Lee and Hong, 2006) exclusively associated with MCSs (Toulmay and Prinz, 2012) (Figure 3). The TULIP superfamily also includes the Takeout and BPI/CETP subgroups. Takeout and similar proteins such as the Juvenile Hormone Binding Protein (JHBP) serve as hydrophobic ligand carriers in the hemolymph of insects (Hamiaux et al., 2009; Suzuki et al., 2011) while the Cholesteryl Ester Transfer Protein (CETP) transports cholesteryl esters, triglycerides and PLs in blood plasma (Qiu et al., 2007); these two subgroups thus encompass hydrophobic ligand carrier extracellular proteins but make use of the same protein fold.

Yeast Oxysterol homology (Osh) proteins, mammalian OxySterol Binding Protein (OSBP) and Oxysterol-binding Related Proteins (ORP) proteins form another large LTP protein family (Kentala et al., 2016). At least some 22 crystal structures of LTP domains from Osh or ORP proteins have been tallied so far including Osh1 (Manik et al., 2017), Osh3 (Tong et al., 2013), Osh4 (Im et al., 2005; de Saint-Jean et al., 2011), Osh6 (Maeda et al., 2013; Moser von Filseck et al., 2015), ORP1, ORP2 (Wang et al., 2019) and ORP3 (Tong et al., 2021) bound to different phosphoinositides and/or cholesterol derivatives; these structures reveal the versatility of this particular LTP fold capable of accommodating different types of ligands (e.g., Osh6 binding PS or PI4P and OSBP binding cholesterol or PI4P). A similar versatility can be described for another large class of LTPs including the PRELI and StART/StARkin domains (Alpy and Tomasetto, 2014) (Figure 3) capable of binding a variety of lipids such as PLs (PA or PS) and sterols or ceramides, respectively.

Although not all LTPs are present at MCSs they all function by extracting/releasing lipids from and into membranes and shielding them from the aqueous phase by burying them in their protein core during the transfer step between the donor and acceptor organelles. The structural diversity of LTPs also translate into a certain diversity of their mechanisms of action. Mechanisms of non-vesicular transfer of lipids by LTPs roughly fall into three classes simplistically described in Figure 4. Each mechanism is exemplified and analyzed in more depth in the following Sections 3 through 6 of this review; the boundaries between categories are however not set in stone as some lipid exchange systems combine several of these mechanisms. LTPs can consist of a “simple” lipid transfer domain or have a more complex multi-domain architecture including several lipid exchange domains and/or other protein modules involved in specific membrane-binding and/or interaction with other membrane associated or embedded proteins thus determining their cellular location and tethering capabilities. Despite the structural diversity of LTPs, a few common features emerge: Their lipid transfer domains are usually ∼150–300 amino acids in size and use highly twisted β-sheets to fold into basket, barrel or cup -shaped structures with binding pockets lined with hydrophobic residues to accommodate the aliphatic hydrocarbon chains of lipids. Some of these structures resemble baskets, sealed at one end and with flexible loops or small helices acting as lids or gates at their entrance to reversibly entrap and release lipidic ligands and function as shuttles (Figure 4A) (e.g., Ups1/Mdm35 or Osh/ORP, StART/StARkin).

Others LTPs are tubular-shaped (e.g., SMP domains of the TULIP family) with a hydrophobic tunnel seemingly suited for the channeling of lipids from one end to the other by lipid sliding (Figure 4B). PLs bear charges as they all harbor negatively charged phosphate moieties but also other charged head groups such as serine, ethanolamine (decarboxylated serine) and choline (trimethylated ethanolamine). This energetic consideration is also relevant to the extraction/desorption steps where a lipid passes from a membrane into a proteinaceous binding pocket and vice versa). Because of the energetic penalty associated with burying polar headgroups into a hydrophobic cavity, some of these tubular LTPs display a lateral opening (also referred to as “seam”) that runs along the tube/barrel and whose edges are sometimes delineated by twisted α-helices. This solvent-accessible seam, accommodates the hydrophilic headgroups while aliphatic chains remained buried within the LTP core and enables the sliding of lipids alongside the tubular core of the LTP (Figure 3).

Lipid-conducting protein bridges or tunnels such as Vps13 and Atg2 that will be discussed in Section 5 constitute one spectacular exception as these LTPs-tethers are much larger (∼1,500–3,000 residues for yeast Atg2 and Vps13, respectively) and elongated (∼200–400Å) proteins: Their lipid transfer domains extend beyond their small lipid-binding signature Chorein-N motif (Figures 3, 4C).

ERMES was initially discovered by the Walter group using a synthetic genetic screen approach (Kornmann et al., 2009) and shown to be composed of the mitochondrial porin Mdm10 and of three SMP domain containing subunits: cytosolic Mdm12, ER-anchored Mmm1 and mitochondrial-attached Mdm34. Later on, a fifth subunit the calcium-binding GTPase Gem1, an ortholog of Miro GTPases was identified (Kornmann et al., 2011; Stroud et al., 2011) (Figure 6A). ERMES involvement in PL exchange (transfer of PS and PC) at the ER-to-Mitochondria interface has now been established combining in vitro (Kawano et al., 2017; Jeong et al., 2017) and in vivo approaches (Kojima et al., 2016). ERMES is not essential though as dominant mutations in Vps13 (Lang et al., 2015; John Peter et al., 2017), another LTP found at nucleus-to-vacuole junctions (NVJ) (Malia and Ungermann, 2016) and vacuole-and-mitochondria patch (vCLAMP) (Elbaz-Alon et al., 2014; Hönscher et al., 2014) MCSs can compensate for the lack of ERMES function. Furthermore, two different complexes have been identified at vCLAMPs one with Vps13 and one with Vps39 (González Montoro et al., 2018). Vps39 is also part of the dynamic HOPS (homotypic fusion and protein sorting) and CORVET tethering complexes (Balderhaar and Ungermann, 2013) involved in endosomal tethering (thus involving a vesicular trafficking pathway) while a dynamic exchange of/competition for Vps39 between HOPS and vCLAMP affects contact site formation at the vacuole (Montoro et al., 2021). These studies reveal the existence of redundant pathways (i.e., ERMES, Vps13-vCLAMP and Vps39-HOPS but not Vps39-vCLAMP) supporting lipid fluxes the cell. In yeasts deficient for the ER PS synthase, heterologous ectopic expression of bacterial PS synthase targeted to peroxisomes or LDs shows that PS reaches mitochondria to be transformed into PE (Shiino et al., 2020). Another recent study combining the simplification and systematic rewiring of PE/PC biosynthesis in yeast using enzyme retargeting (Figure 2) and saturated transposon mutagenesis analysis has elegantly revealed the robustness of lipid cellular distribution together with the existence of lipostatic adaption mechanisms (Peter et al., 2021) that remain to be studied. Other ERMES functions include involvement in mitochondrial protein import and assembly (Ellenrieder et al., 2017), calcium signaling and regulation of its cellular pools, distribution of mitochondria and their DNA (Friedman et al., 2011; Murley et al., 2013), iron homeostasis (Xue et al., 2017) and coenzyme Q biosynthesis (Eisenberg-Bord et al., 2019).

The exact roles and mechanisms of action of Mdm10 within the ERMES remain unknown. However, genetic studies point at a connection between mitochondria-associated membranes and mitochondrial protein biogenesis (as in import and insertion) (Ellenrieder et al., 2017; Meisinger et al., 2007). In yeast, the porin Mdm10 is also a subunit of the Sorting Assembly Machinery (SAM) complex involved in the biogenesis and insertion of many outer mitochondrial porins (such as VDAC). The cryo-EM structure of yeast Mdm10 was recently revealed as a component of the SAM complex, in association with another mitochondrial porin, Sam50, and its two associated peripheral membrane proteins Sam35 and Sam37 (Takeda et al., 2021) (Figure 8A). Porin Mdm10 is composed of 19 β-strands arranged into a canonical transmembrane β-barrel (Flinner et al., 2013); with this last structure, the entire protein superfamily of mitochondrial porins has now been surveyed (Zeth, 2010). The three largest soluble loops of Mdm10, one cytoplasmic and two in the inter-mitochondrial compartment, were too flexible to be resolved in the cryo-EM structure (Figure 8A). Ellenrieder et al. characterized the molecular interactions existing between ERMES and SAM and showed that the two machineries bind to opposite faces of the porin barrel and concluded that Mdm10 acts as the mitochondrial integral membrane anchor for ERMES and its SMP-containing components (Ellenrieder et al., 2016). When mapped to the available cryo-EM structure of Mdm10, its ERMES-binding surface corresponds to β-strand 14 and the following short cytoplasmic loop L14 while the SAM-interacting surface corresponds to β-strand 3. Anecdotally though, Ellenrieder et al. used a poor homology model to erroneously map the two interaction areas on the same cytoplasmic side of the outer mitochondrial membrane (Ellenrieder et al., 2016); the now-available Mdm10 structure indicates that the ERMES-binding interface is on the cytoplasmic leaflet while its SAM-binding interface rather lays within the internal leaflet of the outer mitochondrial membrane (Figure 8A). Although deletions of various Mdm10 loops do not seem to affect cell viability or interactions with SAM and ERMES, it still remains unclear whether the largest cytoplasmic loop (L18) of Mdm10 might also engage in additional direct molecular interactions with the soluble SMP core of ERMES (i.e., Mmm1-Mdm12-Mdm34) and/or the C-terminal uncharacterized region of Mdm34, respectively. In the recent SAM structure, the Sam37 peripheral subunit interacts with one of the short α-helices sitting at the center of the transmembrane β-barrel of Mdm10. Although the studies from the Pfanner and Becker groups hint at close interactions between Mdm10 and Mdm34, the molecular details are still unknown (Ellenrieder et al., 2016). Lastly, it is not excluded that Mdm10 might form dimers or tetramers with membrane-bending or at least membrane-reshaping/destabilizing properties with possible effects on the kinetics of lipid desorption and exchange also discussed in Section 7 of this review; this has been observed for Tom40, another 19 β-strand porin and main subunit of the translocase of the outer membrane (TOM) complex (Tucker and Park, 2019). These observations further complicate attempts to define subunit stoichiometry within an ERMES functional and structural unit.

The fifth ERMES subunit Gem1 is the orthologue of mitochondrial Rho1 GTPase (Miro) in Metazoans (Figure 8B). This regulatory calcium-binding GTPase is a tail-anchored membrane protein, i.e., harboring a single C-terminal transmembrane helix. It was identified in the genetic screen performed by Kornmann et al. (2009) and later also detected in proteomic analysis of native ERMES complexes affinity-purified using tagged Mmm1 subunits (Stroud et al., 2011). High resolution crystal structures of Drosophila and human Miro (Klosowiak et al., 2013; Klosowiak et al., 2016; Smith et al., 2020) provide excellent homology modeling templates for yeast Gem1, a modular protein containing two functional ras-like GTPase domains flanking two calcium-binding EF hands (Koshiba et al., 2011) (Figure 8B). Its function seems regulatory; in vivo studies show that the GTPase1 and the two EF domains of Gem1 are required for its association with ERMES while the GTPase2 is required for proper PL exchange (Frederick et al., 2004; Kornmann et al., 2011; Koshiba et al., 2011). More recently Gem1 and ERMES have been involved in the formation of a cellular structure, the mitochondrial-derived compartment, in response to amino acid over-abundance (English et al., 2020). In Metazoans, Miro is a substrate for the kinase Pink1 and the E3 ubiquitin ligase Parkin and is essential to Ca²⁺-dependent axonal transport of mitochondria along microtubules.

Some cellular processes such as mitochondrial biogenesis and autophagy can require production of large amounts of biological membranes on relatively large physical scales and short time scales. Autophagy is a complex and highly coordinated process (Hollenstein et al., 2019; Hollenstein and Kraft, 2020) involving the formation of a phagophore assembly site (PAS), a transient structure that expands into the autophagosome, a double membrane compartment that grows from lipids supplied by the ER to eventually engulf cytoplasmic macromolecules and organelles for their delivery to lysosomes for destruction/recycling. In yeast, protein Atg2 is involved in autophagy, transporting lipids from the ER to the expanding autophagosome (Valverde et al., 2019; Ktistakis, 2019; Osawa and Noda, 2019; Otomo and Maeda, 2019). In yeast bulk autophagy, the early PAS is formed by high-order assemblies of the Atg1 kinase complex with other Atg proteins to form a so-called Atg-complex condensate (Hawkins and Klionsky, 2020; Fujioka and Noda, 2021) (Figure 9C) where LLPS organizes the site of autophagosome formation (Fujioka et al., 2020). In the past years crystallography and cryo-EM have provided considerable insights into the molecular interactions between Atg proteins and membranes at the heart of PAS formation and autophagosome biogenesis (Lai et al., 2019). Still in yeast, protein Vps13 is found at vacuole-mitochondria (vCLAMP) and vacuole-ER (NVJ, the nuclear envelope derives from the ER) contact sites (Lang et al., 2015) and also promotes the formation of another double membrane structure, the prospore (Park and Neiman, 2012). Both proteins are evolutionarily conserved. In humans, they are two Atg2 homologues (Atg2A-B) while four Vps13 homologues (Vps13A-D) are thought to transfer lipids from ER to other organelles (Kumar et al., 2018) such as mitochondria, endo-lysosomes (Muñoz-Braceras et al., 2019) and LDs (Wang et al., 2021); membrane association and cellular localization are determined by a combination of structural motifs present within each protein (e.g., the WD-40 like domain in yeast Vps13) and the interaction with membrane-specific adaptor proteins binding at the C-terminal and N-terminal ends of the protein (e.g., WIPI in Atg2A and Spo71/73 in yeast Vps13) (Kumar et al., 2018; Dziurdzik and Conibear, 2021; Nakamura et al., 2021) (Figure 9B).

Both proteins are large ranging from ∼1500 residues for Atg2 to ∼3000 residues for Vps13 in Fungi. A number of structural and biochemical studies have shed light on the mechanisms of lipid transfer and membrane/organelle tethering by Atg2 (Valverde et al., 2019; Maeda et al., 2019; Osawa et al., 2019) and Vps13 (Kumar et al., 2018). Several low resolution (∼30 Å) NS-EM reconstructions of yeast Vps13 (De et al., 2017), human Atg2A and yeast Atg2 (Chowdhury et al., 2018), a 15 Å resolution cryo-EM analysis of human Atg2A (Valverde et al., 2019) and a 3.5 Å resolution cryo-EM structure of the N-terminal half (1390 residue-long) of Chaetomium thermophilum Vps13 (Li et al., 2020) revealed that both proteins form elongated structures capable of tethering and bridging two membranes (Figure 9B). Sub-3 Å resolution crystal structures of N-terminal fragments of Schizosaccharomyces pombe Atg2 (Osawa et al., 2019) and C. thermophilum Vps13 (Kumar et al., 2018) reveal that the signature chorein-N motifs (∼100 residues) adopt a U-shaped structure with a hydrophobic groove that can accommodate PL acyl chains to mediate lipid transfer (Figure 9A). These motifs are located at the N-terminal tip of the elongated Atg2 and Vps13 EM reconstructions (Figure 9B). Remarkably, the near-atomic resolution cryo-EM reconstruction of the N-terminal half of fungal Vps13 published by the Reinisch group reveals the existence of a hydrophobic groove (Figure 9B), extending well beyond the mere Chorein-N domain and suited for the “sliding-channeling” of phospholipids in bulk.

Atg2 and Vps13 both function as tethers bridging heterotypic membranes and as LTPs using a “tunnel” mechanism to channel lipids between membranes (Leonzino et al., 2021; Lees and Reinisch, 2020; Reinisch and Prinz, 2021). Mechanistically, they bypass the protein diffusion step described in Sections 2 and 3. They appear quite promiscuous as they bind most glycerophospholipid types and enable the efficient bulk transfer of PLs to support the expansion of the acceptor membranes (in particular for the autophagosome). Mutations introduced within the hydrophobic tunnel block lipid movements but not tethering, thus supporting this bridge model where Atg2 and Vps13 transfer lipids in bulk from the ER towards the expanding autophagosome or between organelles, respectively (Osawa et al., 2019; Valverde et al., 2019; Li et al., 2020) (Figure 9D). Since bulk lipid transfer is essential to support processes involving membrane expansion (such as phagosome growth) but also to connect organelles such as mitochondria and plastids that are essentially disconnected from vesicular trafficking pathways, it is possible that ERMES might serve a similar function to meet the substantial lipid requirements of the mitochondrial network in yeast. The mechanisms determining directionality of these lipid transfer processes will be discussed in Section 7.

Vps13 and Atg2 are the only two eukaryotic membrane-bridging systems extensively characterized so far. Bacterial organisms also seem to have evolved similar strategies to transport lipids “en masse” between two membrane compartments although the proteins assemblies and structures involved are not evolutionarily related. In Gram-negative bacteria, the lipopolysaccharide (LPS) transport system (Lpt) (Suits et al., 2008; Qiao et al., 2014; Botos et al., 2016; Owens et al., 2019) and the lipophilic envelope-spanning tunnel (Let) (Isom et al., 2020) span the entire periplasmic space to transfer LPS or PLs from the cell membrane towards the outer membrane and across the thin peptidoglycan layer, respectively. In Lpt the β-jellyroll folds present in LptA, the periplasmic C-terminal domains of LptC and LptF, and the periplasmic N-terminal domain of the outer transmembrane β-barrel translocon LptD interact and align to form a periplasmic bridge with a solvent exposed V-shaped hydrophobic groove suited for the transfer of LPS (Figure 10). Transport is powered by ATP hydrolysis (Mi et al., 2017; Owens et al., 2019) that ensures unidirectional vectorial transfer by flipping and pushing in the LPS substrates from the cell membrane into the lipid bridge. In Let where each LetB chain consists of seven repeats of a mammalian cell entry (MCE) domain, six chains of LetB associate to stack seven hexameric MCE rings on top of each other and create a protein tubular structure spanning the periplasmic space. Pore-lining loops in each MCE domains delineate a central channel capable of binding PLs (Figure 10).

The term transport should be used with caution. In integral membrane protein biology, transport implies an active process, requiring an energy source and therefore its expenditure, usually in the form of ATP (e.g., ABC transporters such as endosomal ABCG1 that transports sterols), or ion-gradients (e.g., proton gradient) generating membrane potentials. Such transporters are opposed to channels that “only” facilitate the movement of molecules along an existing concentration gradient (e.g., aquaporins water channels). Thus, transporters can function against concentration gradients while channels cannot (Li et al., 2021). In molecular terms, most LTPs are passive and do not directly expend energy (i.e., nucleotide hydrolysis or membrane potential gradient). Enzymes responsible for specific steps along the synthesis/transformation or catabolism of lipids are specifically distributed in specific organelles; thus, metabolic reactions taking place between compartments can drive LTP-mediated exchange of lipids along concentration gradients (Shiino et al., 2020) (Figure 12A). Ceramide transport by CERT from ER to Golgi is driven by the transformation of ceramide into sphingomyelin in the Golgi that cannot be transported back. LD growth is probably the most spectacular example showing the result of the activities of TAG synthesis in the ER and of LD-membrane attached or embedded enzymes on the generation and subsequent growth of these droplets, respectively (Figure 11).

Two classes of integral membrane proteins participate in the transfer of lipids from one leaflet to another within the same membrane: scramblases and flippases/floppases. Technically they are not considered as lipid “transfer” proteins as they do not shuttle lipids across an aqueous space between membranes. Their functional importance is considerable as they abolish (i.e., scramblases) or establish (i.e., flippases and floppases) membrane asymmetry and contribute to lipid fluxes (in particular PS) at MCSs and within the cell through the formation of gradients between leaflets within a membrane and thus support the formation of lateral nanodomains not only within the membrane but also within leaflets (Shin and Takatsu, 2020; Lenoir et al., 2021) (Figures 12A,B). Flippases transport lipids (sterols, glycerophospholipids such as PS, PE or PC and ceramides) from the exoplasmic (external or luminal) to the cytoplasmic leaflet while floppases work in the opposite direction; these processes require ATP hydrolysis to enable the vectorial transport against a concentration gradient. In yeast and Mammals, ABC transporters (Prasad et al., 2016; Kumari et al., 2021; Banerjee et al., 2021) and P4 subtype ATPases (P4-ATPases) (Shin and Takatsu, 2020) are largely represented among lipid translocators (flippases/floppases). ABC lipid translocases play a role in bacterial lipid transport systems such as the Lpt bridge that requires the MsbA LPS-flippase to energize the process in addition to the ATP hydrolysis activity of subunits LptB (Owens et al., 2019; Mi et al., 2017) (Figure 10). P4-ATPases constitute another important group of lipid flippases whose structures and mechanisms of action have been recently elucidated in the case of yeast PS flippase Drs2/Cdc50 (Bai et al., 2019; Timcenko et al., 2019; Timcenko et al., 2021) and human homologues ATP8A-Cdc50 (Hiraizumi et al., 2019) and ATP11C-Cdc50 (Nakanishi et al., 2020a; Nakanishi et al., 2020b).

Scramblases equilibrate lipid distributions between leaflets and thus dissipate lipid asymmetry in membranes; they are ATP-independent and follow energy potential gradients associated with a simple chemical concentration gradient or the membrane electric potential (resulting from charge distributions) (Lenoir et al., 2021). The first scramblase structure belongs to the TMEM16 protein family (Brunner et al., 2014; Brunner et al., 2016) and reveals a dimer where each monomer bears a membrane-spanning hydrophilic cleft facing the lipid environment. Scramblases most likely combine membrane thinning and electrostatic shielding mechanisms to lower the energetic barrier associated with transferring a PL molecule, and in particular its polar headgroup, from one leaflet to the other, by providing a trans-bilayer hydrophilic path for its facilitated diffusion down the concentration gradient (Brunner et al., 2014; Brunner et al., 2016; Theorin et al., 2019) (Figure 12B). Recently, a model proposed that lipid scramblases present in the donor (i.e., ER scramblases VMP1 and TMEM41B) and acceptor (i.e., Golgi scramblase Atg9A) membranes tethered by the mammalian Atg2A tether/LTP support autophagosome expansion by re-equilibrating lipid distributions between leaflets of donor and acceptor membranes and compensate for their respective depletion or enrichment (Figure 9D) (Ghanbarpour et al., 2021). Scramblase Atg9 is the only Atg integral membrane protein (Guardia et al., 2020; Maeda et al., 2020; Matoba et al., 2020); Atg9-vesicles derived from the Golgi are essential as they provide the initial membranous kernel of the isolation membrane (IM or phagophore) (Mari et al., 2010; Yamamoto et al., 2012; Matoba and Noda, 2020; Orii et al., 2021), a cup-shaped membrane structure that further expands into the autophagosome using lipids supplied from the ER by Atg2 (Figure 9C). In this “molecular sink” model, lipids are pushed out at the donor site and pulled in at the acceptor site as a result of synergistic scramblase activities in both organelles.

Compared to scramblases that facilitate lipid movements at rates of ∼10,000 lipids/sec, flippases are much slower (∼2–25 lipids/sec) (Goren et al., 2014); this is due to fundamentally different molecular mechanisms: flippases of the ABC-transporter class use an alternating access mechanism involving large conformational movements and domain rearrangements powered by the energy associated with ATP binding and hydrolysis that has been extensively studied and characterized in transporters belonging to the ABC and the Major Facilitator Superfamily classes (Li et al., 2021). P4-ATPases also undergo ATP-driven large conformational rearrangements (Hiraizumi et al., 2019; Nakanishi et al., 2020a; Timcenko et al., 2021) (Figure 12B).

Osh6 (Moser von Filseck et al., 2015), Osh4 (von Filseck et al., 2015) and OSBP are two examples illustrating how tethered LTPs can simultaneously support the counterflow of two distinct lipids at a membrane contact site using chemical gradient energy. The LTP Osh6 consists solely of an ORP domain (Figure 3) while OSBP contains an ORP domain but also a PH domain and FFAT motif involved in dual membrane tethering through direct interaction with PI4P in the PM and adaptor-mediated attachment to the ER membrane, respectively (Figures 12C,D). Both proteins are capable to bind two different lipids (PS/PI4P and Cholesterol/PI4P, respectively) and transfer them in opposite directions. While Ohs6 functions as an LTP tethered at ER-PM junctions by the Ist2 protein (D'Ambrosio et al., 2020) (Figure 12B), OSBP attaches itself to the Golgi through its PI4P-binding PH domain and also tethers to the ER through its FFAT motif interacting with the ER-anchored vesicle-associated membrane protein-associated protein A (VAP-A) (Figure 12D). Their elegant mechanisms have been dissected in exquisite detail by the Drin (Moser von Filseck et al., 2015) and Antonny (Mesmin et al., 2013; Antonny et al., 2018) groups, respectively. ATP-consuming biosynthesis generates PI4P for its exchange from PM/Golgi to ER while its irreversible hydrolysis into PI by the ER-associated phosphatase Sac1 sets the directionality of these two cycles. OSBP has a higher affinity towards PI4P than the sterol so exchange of sterol for PI4P is facilitated at the Golgi; once the PI4P is shuttled to the ER, its hydrolysis acts as an energy sink, irreversibly driving this branch of the cycle. This counter-exchange mechanism is remarkable as it enables transport of cholesterol from the ER to enrich the Golgi. These latter examples underline the importance and complementarity of LTPs next to vesicular trafficking as LTPs by transferring lipids against their concentration gradients can counteract lipid homogenization resulting from vesicular traffic (Antonny et al., 2018).

In lipid exchange mediated by “basket”-shaped LTPs that shuttle lipids back and forth by diffusion or tethered diffusion, the rate-limiting step for the transfer of a lipid from one membrane to another is the extraction or desorption from the membrane itself rather than the LTP-assisted transport (Dittman and Menon, 2017; Wong et al., 2017). Spontaneous desorption of a lipid into the aqueous phase and henceforth between two membranes is energetically costly (∼10–20 kCal/mol) and extremely slow (several days) (Rogers et al., 2021). Local membrane composition, lipid packing and membrane order also affects the energetics of lipid desorption (Rogers et al., 2021). LTPs lower the desorption energy barrier as they partition the lipid from a bilayer into a hydrophobic binding pocket instead of bulk aqueous solvent. Electrostatics and desolvation/solvation of the lipid-binding site (Lipp et al., 2019) and the membrane reshaping or bending properties of LTPs are thus crucial for desorption at the donor membrane. The curvatures observed in the E-SYT2 dimer and Mdm12-Mmm1 tetramer, might modulate membrane shape (and/or vice-versa) thus lowering the activation energy associated with lipid desorption and transfer to the LTP core.

Redundant yet slow and insufficient? When a lot is not enough. Based mostly on in vitro reconstitution studies and measurements, at the single protein molecule level these systems appear quite slow (∼1 lipid/sec at best) and likely involved in local adjustment/tweaking of the lipid composition of membranes. For example, in vitro, Osh4 transports sterol or PI4P (in absence of their counterpart) at rates ∼ 2 lipids/min; however, when both two lipids are present, Osh4 counter-exchanges them at rates of ∼ 20 lipids/min (von Filseck et al., 2015). CERT transport ceramides at ∼ 4 lipids/min (Hanada et al., 2003). Even when relative abundance and localization of these LTPs are factored in order to calculate the global lipid exchange rate at a specific MCS, the estimated fluxes seem despairingly too slow to support cellular functions. Thus, other factors (either unknown or not easily measurable/quantifiable in the experimental conditions used in vitro or in vivo) have to be considered. In the case of ERMES and lipid exchange at yeast mitochondria-associated membranes, Petrungaro and Kornmann attempt to address these discrepancies (∼500 fold difference) to reconcile Biology with the apparent insufficiency of measured lipid fluxes (Petrungaro and Kornmann, 2019). They propose that the presence of distinct LTPs combined with the existence of bypasses/shunts, well-exemplified by the vCLAMP and ERMES case discussed in Section 4.1, could account for these discrepancies (Figure 13A). They also suggest that membrane “destabilizing” proteins such as Mcp1 (John Peter et al., 2017) could further facilitate lipid desorption; as discussed in Section 4.2, the Mdm10 ERMES porin could also function as a catalyst of lipid desorption by locally distorting/thinning the membrane. Interestingly, another porin the outer mitochondrial protein translocase Tom40 is the receptor for Vps39 in one of the two functionally independent complexes identified at vCLAMPs (González Montoro et al., 2018) (Figure 13B). Tom40 barrels associate into dimers and even tetramers and induce membrane bending (Bausewein et al., 2017; Araiso et al., 2019; Wang et al., 2020). The importance of membrane distortion and thinning induced by transmembrane proteins (either α-helix or β-strand based) has been documented in different systems, in particular protein translocation systems (Li et al., 2021; Wu and Rapoport, 2021). The same argument could be made with LTP-associated scramblases that manipulate the local biophysical properties of membranes to move lipids between leaflets.

What about transfer by lipid bridges/tunnels then? Theoretically they bypass or at least minimize protein diffusion effects leaving us to essentially consider lipid dynamics within the proteinaceous tunnel and lipid desorption as sole apparent limiting factors. Molecular dynamic simulations should prove insightful, although they would require better-defined atomic models of these eukaryotic protein bridges and model their interaction with membrane surfaces. As it stands now, lipid exchange systems such as Atg2 and Vps13, and possibly the more contentious ERMES, would need to sustain much faster rates of transfer, in order to support biological processes requiring fast and/or massive membrane expansion during organelle biogenesis such as phagosome growth and mitochondrial fusion or fission. These considerations beg one question: How complete is our knowledge of all the LTP systems and lipid exchange routes and networks operating in a given organism even seemingly as “simple” and genetically tractable as yeast?