Functional analyses of phosphatidylserine/PI(4)P exchangers with diverse lipid species and membrane contexts set unanticipated rules on lipid transfer

Several members of the oxysterol-binding protein-related proteins (ORPs)/oxysterol-binding homology (Osh) family exchange phosphatidylserine (PS) and phosphatidylinositol 4-phosphate (PI(4)P) at the endoplasmic reticulum/plasma membrane (PM) interface. It is unclear whether they preferentially exchange PS and PI(4)P with specific acyl chains to tune the features of the PM, whether they use phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2) instead of PI(4)P for exchange processes and whether their activity is influenced by the association of PS with sterol in the PM. Here, we measured in vitro how the yeast Osh6p and human ORP8 transported diverse PS and PI(4)P subspecies, including major cellular species, between membranes. We established how their activity is impacted by the length and unsaturation degree of these lipids. Surprisingly, the speed at which they individually transfer these ligands inversely depends on their affinity for them. To be fast, the transfer of high-affinity ligands requires their exchange for a counterligand of equivalent affinity. Besides, we determined that Osh6p and ORP8 cannot use PI(4,5)P2 for intracellular lipid exchange, as they have a low affinity for this lipid compared to PI(4)P, and do not transfer more PS into sterol-rich membranes. This study provides insights into PS/PI(4)P exchangers and sets unanticipated rules on how the activity of lipid transfer proteins relates to their affinity for ligands.

Crystallographic data has revealed that Osh6p consists of one domain -called ORD (OSBP-related domain)with a pocket that could alternately host one molecule of PS or PI(4)P, a lipid belonging to the class of phosphorylated phosphoinositides (PIPs) 12,13 . The pocket is closed by a molecular lid once the lipid is loaded.
These structural data along with in vitro analyses and cellular observations have revealed the following mechanism: Osh6/7p extract PS from the ER and exchange it for PI(4)P at the PM; then they deliver PI(4)P into the ER and take PS once again. This PS/PI(4)P exchange cycle is propelled by the synthesis of PI(4)P from phosphatidylinositol (PI) in the PM and its hydrolysis in the ER membrane, which maintains a PI(4)P concentration gradient between the two compartments.
The PS/PI(4)P exchange mechanism is evolutionarily conserved 14 . Human cells express ORP5 and ORP8 that tether the ER membrane to the PM and exchange PS and PI(4)P between these membranes. They include an N-terminal pleckstrin homology (PH) domain, an ORD resembling Osh6p, and a C-terminal transmembrane segment 15,16 . They are anchored to the ER by this segment and associate with the PM via their PH domain that targets PI(4)P but also PI(4,5)P 2 [17][18][19] chains. This seems critical for the maintenance of a PI(4,5)P 2 pool and a PI(4,5)P 2 -dependent signaling process in the PM via the so-called PIP cycle 32 . Consequently, it is worth analyzing how ORP/Osh proteins transfer PS and PI(4)P species with different acyl chains in order to define to what extent they can contribute to the tuning of lipid homeostasis in the PM.
A second issue concerns the links between the ORP/Osh-mediated PS transfer process and the regulation of PI(4,5)P 2 levels. It has been reported recently that ORP5/8 use PI(4,5)P 2 rather than PI(4)P as a counterligand for supplying the PM with PS 17 . This would mean that the PI(4,5)P 2 level in the PM is directly lowered by the consumption of PI(4,5)P 2 during exchange cycles. Yet this conclusion is disputed and remains surprising in view of the very first structural analyses that suggest that the polar head of PI(4,5)P 2 , unlike that of PI(4)P, cannot be accommodated by the ORD due to steric constraints 33 . The structures of the ORD of ORP1 and ORP2 in complex with PI(4,5)P 2 have been solved 34,35 but they revealed that the PI(4,5)P 2 molecule is only partially buried in the binding pocket. All these observations raise doubts about the existence of functional PI(4,5)P 2 -bound forms of ORPs, including ORP5 and ORP8, in the cell.
Third, as mentioned above, unsaturated PS and sterol preferentially associate with each other in the PM 30,31 . Recently, it has been proposed that unsaturated PS and PI(4)P co-distribute into sterol-rich areas in the PM of yeast to form nanodomains that are preferentially targeted by PI(4)P 5-kinase and are the place of active PI(4)P-to-PI(4,5)P 2 conversion 22 . Osh proteins, such as Osh6/7p but also Osh4/5p which are sterol/PI(4)P exchangers, control the formation of these domains. One might wonder whether the tight association of sterol with PS acts as a thermodynamic trap that aids PS/PI(4)P exchangers to accumulate PS in the PM and thus contribute to the coupling between PS transfer and PI(4,5)P 2 synthesis.
Here, we addressed these three interrelated questions by conducting in vitro functional analyses of Osh6p and ORP8, combined with simulations and cellular observations. Using a large set of PS subspecies, we found that these LTPs transfer unsaturated PS more slowly than saturated PS between liposomes. In contrast, in a situation of PS/PI(4)P exchange, only the transfer of unsaturated PS species is largely accelerated, and efficient exchange occurs with certain unsaturated PS and PI(4)P species that are prominent in cells. Unexpectedly, by measuring the affinity of Osh6p and ORP8 for PS and PI(4)P subspecies and correlating these data with transfer rates, we established that the simple transfer of high-affinity ligands is slower than that of low-affinity ligands. Next we found that high-affinity ligands are rapidly transferred only if they can be exchanged for ligands of equivalent affinity. Furthermore, we determined that, if PI(4)P and PI(4,5)P 2 are both accessible to Osh6p and ORP8, PI(4,5)P 2 cannot be transferred or exchanged for PS because PI(4,5)P 2 is a low-affinity ligand. This suggests that PI(4,5)P 2 cannot be transported by ORP/Osh proteins in cells. Finally, we found that the activity of PS/PI(4)P exchangers barely changes on sterol-rich membranes. Our study provides insights into PS/PI(4)P exchangers but also sets general rules on how the activity of LTPs relates to their affinity for lipids, which improves our knowledge of lipid transfer.

Osh6p and ORP8 transfer saturated and unsaturated PS species differently
We first measured in vitro the speed at which Osh6p and the ORD of ORP8 (ORP8   14   ). Like Osh6p, ORD8 transferred saturated PS species more rapidly than unsaturated ones when PI(4)P was absent. In a situation of PS/PI(4)P-exchange, the transfer of unsaturated PS species (except for 18:2/18:2-PS) was much more rapid (up to 29-fold) whereas the transfer of saturated PS was slightly enhanced or inhibited. Collectively, these data did not point to a monotonic relationship between PS transfer rates and the length of PS acyl chains or the degree of unsaturation of these chains. However, they indicated that PS species were transported and exchanged with PI(4)P quite differently depending on whether or not they had at least one double bond.

Coupling between the transfer rate of PS species and PI(4)P under exchange conditions
Next, we determined whether the rate of 16:0/16:0-PI(4)P transfer was different depending on the nature of the PS species under the exchange conditions. Using a fluorescent PI(4)P sensor (NBD-PH FAPP ) and a FRET-based strategy akin to that used to measure PS transfer, we measured the speed at which PI(4)P, at 5% in L B liposomes, was transported by Osh6p and ORD8 to L A liposomes devoid of PS or containing a given PS species (at 5%) (Figure 1b, Figure 1 -Figure Supplement 2). With PS-free L A liposomes, the initial PI(4)P transfer rate was 8.4 ± 1.3 PI(4)P.min -1 for Osh6p and 4.2 ± 0.5 PI(4)P.min -1 for ORD8 (Figure   1d). In a situation of lipid exchange, these transfer rates increased to a different degree when L A liposomes contained a PS species other than 18:0/18:0-PS and, in experiments with ORD8, 16:0/16:0-PS (Figure 1d).
For each PS species, we calculated an acceleration factor corresponding to the ratio (expressed as a log value) of the PI(4)P transfer rate, measured in the presence of this PS species, to the PI(4)P transfer rate measured in the absence of counterligand. Also, acceleration factors based on PS transfer rates reported in Figure 1c were determined. Then for each LTP, these two factors were plotted against each other, allowing saturated and unsaturated PS species to cluster in two groups (Figure 1e). With Osh6p, the group including saturated PS species was characterized by null or negative acceleration factors for PS (down to -0.49) associated with low or moderate acceleration factors for 16:0/16:0-PI(4)P (from 0.20 to 0.58). In contrast, the group corresponding to unsaturated PS species was characterized by higher acceleration factors for both PS (from 0.30 to 1.30) and PI(4)P (from 0.49 to 0.79). With ORD8, the acceleration factors for saturated PS were negative, null or moderate (from -0.15 to 0.47) and associated with null or moderate acceleration factors for PI(4)P (from 0 to 0.48). For unsaturated PS, the acceleration factors were higher, ranging from 0.44 to 1.47 for PS and from 0.40 to 0.55 for PI(4)P. The observation of high acceleration factors for both unsaturated PS and 16:0/16:0-PI(4)P, and much lower or even negative values for saturated PS species, suggests that LTPs exchange unsaturated PS for PI(4)P much more efficiently than saturated PS.
Osh6p slowly transferred the two PS species from L A to L B liposomes in the absence of PI(4)P but ten Interestingly, the 16:0/18-1-PS and 16:0/18:1-PI(4)P transfer rates were both similar and high in a situation of lipid exchange, suggesting that Osh6p can execute an efficient one-for-one exchange of these major yeast PS and PI(4)P species.
In the absence of PI(4)P, ORD8 slowly transferred 16:0/18:1-PS and 18:0/18:1-PS between membranes and much faster in a situation of exchange, by a factor of 4.8-6.8 and 11.5-16.6, respectively, depending on the PI(4)P species used as counterligand (Figure 2b, Figure 2 - Figure Supplement 2c, d). Like Osh6p, ORD8 barely transferred unsaturated PI(4)P under non-exchange conditions (< 0.26 lipids.min -1 ), compared to 16:0/16:0-PI(4)P. Under exchange conditions, ORD8 transferred these PI(4)P species more rapidly but far less than PS in the opposite direction. This suggests that ORP8 cannot efficiently exchange unsaturated PS for PI(4)P. We conclude that the acyl chain composition of PI(4)P, and primarily its unsaturation degree, impacts how Osh6p and ORP8 transfer and use this PIP in exchange for PS.
The opposite transfer of 16:0/18:1-PI(4)P was inhibited if 18:0/18:0-PS was tested as counterligand but enhanced using the two unsaturated PS forms (Figure 3a). Similar results were obtained with ORD8 except that the transfer of 18:0/18:0-PS was slightly more rapid (by 2.2-fold) in exchange conditions (here 18:1/18:1-PI(4)P was used as counterligand (Figure 3c)). However, the rate of acceleration was low compared to that measured with 18:0/18:1-PS and 18:1/18:0-PS (by 24.0 and 8.4-fold, respectively). Jointly these results indicate that only one double bond, in one or the other acyl chain of PS, is enough to dramatically change how the LTPs transports and exchanges this lipid for PI(4)P.

Osh6p and ORD8 have a higher affinity for unsaturated than saturated PS and PI(4)P species
Our results suggest that Osh6p and ORD8 transport and exchange PS and PI(4)P at a different speed depending on the unsaturation degree of these lipids. To further analyze why, we devised assays to determine the relative affinity of these LTPs for each PS and PI(4)P species. We established that the intrinsic fluorescence of both proteins (from tryptophan, with a maximum intensity at λ = 335 and 340 nm, respectively), was quenched by ~25% when mixed with liposomes doped with 2% NBD-PS, a PS species whose C12:0 acyl chain at the sn- . These results suggest that Osh6p is more prone to capture and hold unsaturated PS and PI(4)P than saturated PS, corroborating the results from competition assays.

The affinity of Osh6p and ORP8 for PS and PI(4)P species dictates how they transfer and exchange them
Next, we analyzed how the affinity of Osh6p and ORD8 for lipid ligands was related to their capacity to transfer them. To do so, we plotted the PS and PI(4)P transfer rates measured in non-exchange contexts (reported in Figure 1c 50 ). Collectively, these analyses reveal an inverse correlation between the affinity of an LTP for a ligand and its ability to simply transfer it down its concentration gradient. Moreover they indicate that in a situation of lipid exchange, the acceleration of the PS and PI(4)P transfer rate is more pronounced with high-affinity PS species.

Simulation of transfer and exchange activity of the ORD as a function of its affinity for PS and PI(4)P
To understand why the affinity of Osh6p and ORD8 for PS and PI(4)P species governed how they transferred and exchanged these lipids, we built a simplified kinetic model (Figure 5a). It was assumed that the ORD interacts similarly with A and B membranes during a transfer process (k ON-Mb= 10 s -1 and k OFF-Mb =0.1 s -1 ) with an equal ability to capture and release a given lipid (similar k ON-lipid and k OFF-lipid ). We simulated initial PS transfer rates for k ON-PS values ranging from 10 -2 to 10 4 µM -1 .s -1 to evaluate how the affinity of the ORD for PS Next, using the same range of k ON-PS values, we simulated PS and PI(4)P transfer rates in a situation of lipid exchange between A and B membranes that initially contained 5 µM PS and PI(4)P, respectively ( Figure   5b, left panel, pink dots for PS and orange dots for PI(4)P). Acceleration factors were determined from rates established for each lipid in exchange and non-exchange contexts. The PS transfer rate found to be maximal at k ON-PS = 3.7 µM -1. s -1 when PI(4)P was absent, slightly increased in the presence of PI(4)P (acceleration factor = 0.13). If k ON-PS > 3.7 µM -1 .s -1 the PS transfer rates were lower in a non-exchange situation but considerably higher if PI(4)P was present as counterligand. In contrast, for k ON-PS < 3.7 µM -1 .s -1 , the PS transfer rate decreased toward zero, even if PI(4)P was present, and acceleration factors were almost null. In parallel, the PI(4)P transfer rates were found to be systematically higher in the presence of PS yet to a degree that depended on k ON-PS values ( Figure 5b, left panel). Finally, we noted that an ORD efficiently exchanges PS and PI(4)P if it has a higher affinity for PS than PI(4)P (k ON-PS > k ON-PI(4)P , Figure 5b, left panel, grey area). These simulations again corroborated our experimental data. In non-exchange situations, saturated PS species, which are low-affinity ligands compared to 16:0/16:0-PI(4)P, are transferred at the fastest rates; yet these rates barely or marginally increase once PI(4)P is present. In contrast, unsaturated PS species, which are globally better ligands than 16:0/16:0-PI(4)P, are slowly transferred in non-exchange situations, but much faster when PI(4)P is present. In all cases, the PI(4)P transfer rate is unchanged or higher in the presence of PS.
To consolidate this analysis, we plotted the acceleration factors for PS and PI transferred and exchanged for PS while PS transfer was enhanced by these PI(4)P species (Figure 2). Overall our model showed how variations in the capacity of the ORD to extract and deliver PS and PI(4)P could modify how it transfers and exchanges these lipids.

Osh6p and ORD8 cannot use PI(4,5)P 2 if PI(4)P is present in membranes
ORP5/8 have been suggested, notably based on in vitro data, to use PI(4,5)P 2 instead of PI(4)P as a counterligand to supply the PM with PS 17 but this conclusion is disputed 19 . To address this issue, we measured how ORD8 transferred PI(4)P and PI(4,5)P 2 (with 16:0/16:0 acyl chains) from L B liposomes that contained only one kind of PIP or, like the PM, both PIPs, to L A liposomes. NBD-PH FAPP , which can detect PI(4,5)P 2 in addition to PI(4)P 17 , was used as sensor. ORD8 transported PI(4,5)P 2 more swiftly than PI(4)P (Figure 6a,b), as previously shown, but surprisingly, when both PIPs were in L B liposomes, the transfer kinetics was comparable to that measured with PI(4)P alone.
To understand why, we specifically measured the transfer of PI(4,5)P 2 with a sensor based on the PH domain of the phospholipase C-δ1 (PH PLCδ1 ), which has a high affinity and specificity for the PI(4,5)P 2 headgroup 42 . This domain was reengineered to include, near its PI(4,5)P 2 -binding site 43 , a unique solventexposed cysteine (C61) to which a NBD group was attached ( Fluorescence assays also indicated that NBD-PH PLCδ1 bound to PI(4,5)P 2 -containing liposomes, as its NBD signal underwent a blue-shift and a 2. These data indicate that ORP8 and Osh6p preferentially extract PI(4)P from a membrane that contains both PI(4)P and PI(4,5)P 2 , suggesting that they use PI(4)P rather than PI(4,5)P 2 in exchange cycles with PS at the PM.
To address this possibility in vitro, we devised an assay with three liposome populations (Figure 6d, e) to examine whether ORP8 delivers PS in a PI(4)P-containing membrane or in a PI(4,5)P2-containing membrane. First, L A liposomes doped with 5% PS were mixed with NBD-C2 Lact . Then L B liposomes, containing 5% PI(4)P and 2% Rhod-PE, and L C liposomes, only made of PC, were successively added.
Injecting ORD8 provoked a quenching of the NBD signal, indicating that the C2 Lact domain moved onto the L B liposomes. The signal normalization indicated that ~1 µM of PS was transferred to L B liposomes.
Equivalent data were obtained with L C liposomes doped with PI(4,5)P 2 , suggesting that this lipid has no influence on the PI(4)P-driven transfer of PS to L B liposomes mediated by ORD8. We performed mirror experiments with L B liposomes that contained PI(4,5)P 2 and L C liposomes with or without PI(4)P.
Remarkably, PS was transferred to L B liposomes but not if L C liposomes contained PI(4)P. We concluded that ORP8 selectively delivers PS in a compartment that harbors PI(4)P if PI(4,5)P 2 is present in a second compartment. This suggests that PI(4)P, and not PI(4,5)P 2 , is used by ORP5/8 to transfer PS intracellularly.
These observations suggest that ORD8 and Osh6p have a lower affinity for PI(4,5)P 2 than for PI(4)P.
Confirming this, the NBD-PS-based competition assay showed that each protein barely bound to PI(4,5)P 2 compared to PI (4) Osh6p(noC/S190C), which has a unique cysteine at position 190; this residue is solvent-exposed only if the molecular lid that controls the entrance of the binding pocket of the protein is open 44 . This construct was added to liposomes devoid of PIPs or containing 2% PI(4)P or PI(4,5)P 2 . Then, 7-diethylamino-3-(4'maleimidylphenyl)-4-methylcoumarin (CPM), a molecule that becomes fluorescent only when forming a covalent bond with accessible thiol, was added to each sample. After a 30-min incubation, a high fluorescence signal was measured with Osh6p(noC/S190C) mixed with PC liposomes, indicating that the protein remained essentially open over time (Figure 6g). In contrast, almost no fluorescence was recorded when the protein was incubated with PI(4)P-containing liposomes, indicating that it remained mostly closed, as previously shown 44 . Remarkably, a high signal was obtained with liposomes doped with 2% PI(4,5)P 2 , indicating that Osh6p remained open as observed with pure PC liposomes. Altogether, these data suggest that Osh6p/7p and ORP5/8 have a low affinity for PI(4,5)P 2 compared to PI(4)P, likely as they cannot form stable and closed complexes with this lipid.

Sterol abundance in membrane does not enhance PS delivery and retention
Like PS, sterol is synthesized in the ER and enriched in the PM, where it constitutes 30-40% of all lipids 45 . PS is thought to associate laterally with sterol, thus retaining sterol in the inner leaflet of the PM and controlling its transbilayer distribution 30,31 . However, it was not known whether sterol stabilizes PS and thereby aids ORP/Osh proteins to accumulate PS in the PM. To explore this possibility, we measured in vitro the speed at which Osh6p transported PS from L A liposomes to L B liposomes, containing or not 30% cholesterol or ergosterol, and doped or not with 5% PI(4)P. These assays were performed using 16:0/18:1-PS and 16:0/18:1-PI(4)P, as in yeast, these predominant PS and PI(4)P species are thought to preferentially populate sterol-rich nanodomains in the PM 22 . However, we observed that the transfer of PS was not markedly impacted by higher We conclude that the presence of sterol in the PM is not essential for PS retention and thus for PS/PI(4)P exchange.

Discussion
LTPs have been discovered and studied for more than 40 years, yet few studies have explored how their activity depends on the nature of the lipid acyl chains. A few kinetic studies have shown that a nonspecific-LTP transfers shorter PC more easily than long PC 48 , that a glycolipid transfer protein (GLTP) preferentially transports short glucosylceramides 49 and that the ceramide transfer protein (CERT) is active with ceramide species whose length does not exceed the size of its binding pocket 50,51 . Moreover, the link between the activity of LTPs and their affinity for lipid ligands remained largely obscure. Here, we measured how fast LTPs transfer saturated vs unsaturated lipids between membranes, both in a situation of simple transfer and in a situation of exchange with a second ligand, and we measured their relative affinity for these lipids. Our investigations used approaches that detect the transfer and binding capacities of LTPs with unmodified lipid ligands (i.e., without bulky fluorophores), thus with an unprecedented level of accuracy. This study offers novel insights into PS/PI(4)P exchangers and, by identifying how the activity of LTPs relates to their affinity for lipid ligands, provides general rules that serve to better analyze lipid transfer processes.
Our kinetic measurement indicated that overall ORD8 transfers PS and PI(4)P more slowly than Osh6p does. This might be due to structural differences between the two proteins or to the fact that ORD8 only functions optimally in the context of the full-length ORP8 or between closely-apposed membranes. Apart from this, Osh6p and ORD8 responded similarly to the same changes in the PS and PI(4)P acyl chain structures. In particular, they transferred these lipids quite differently depending on whether they were saturated or not. This seems to be mainly because these LTPs have a lower affinity for saturated than unsaturated lipids. Remarkably the presence of only double bond in one of the acyl chains of the ligand is sufficient to significantly change their behavior. Our kinetic model suggests that the affinity of PS/PI(4)P exchangers for lipid ligands and its capacity to transfer them is governed by the extraction step (reflected by the k on values). An early study of large series of PE and PC species showed that increasing the degree of unsaturation in the acyl chains of phospholipids increases the rate at which they spontaneously desorb from the membrane 52 . One might therefore posit that the intrinsic propensity of PS species to move out of the membrane determines how they are captured and transferred by ORP/Osh proteins. However, several data suggest that the intrinsic tendency of PS species to leave the membrane cannot explain, or only very partially, why these lipids are more or less easily captured by Osh6p and ORD8. processes. An equivalent analysis with PI(4)P remains difficult to conduct as much fewer PI(4)P species were assayed. The fact that polyunsaturated PI(4)P, the major brain PI (4) . This suggests that the PS transfer activity of ORP5/8 intimately depends on PI(4)P but is likely not influenced by its unsaturation degree, a parameter that varies between cells in tissues and cultured transformed cells. When exchange is possible, the transfer of 18:0/20:4-PI(4)P and 18:1/18:1-PI(4)P is faster but remains, intriguingly, much slower than the PS transfer, suggesting a weak coupling between the two transfer processes. This presumably arises from the high affinity of the ORD for these PI(4)P species, as suggested by our kinetic model.
Possibly, the hydrolysis of PI(4)P by Sac1 at the ER enhances through mass action the transfer of PI(4)P from the PM to the ER by ORP5/8 while facilitating the extraction of PS from this organelle, and thus improves the coupling between PS and PI(4)P transfer at the ER/PM interface. Of note, PI(4)P hydrolysis is mandatory for OSBP to execute the sterol/PI(4)P exchange as OSBP has a much higher affinity for PI(4)P than for sterol 56 . Last we found in vitro that cholesterol does not promote the delivery of 18:0/18:1-PS in an acceptor membrane and the retention of PS in the PM of human cells.
Corroborating previous observations 17 , we measured that Osh6p and ORD8 transferred PI(4,5)P 2 between liposomes more rapidly than PI(4)P. However, when PI(4)P and PI(4,5)P 2 both resided in the same donor membrane, only PI(4)P was transferred to acceptor membranes, for all tested acyl chain compositions.
Moreover, only PI(4)P was used as counterligand for the transfer of PS when both PIPs were present. In fact, Osh6p and ORD8 show a much lower affinity for PI(4,5)P 2 than for PI(4)P. This likely relates to the fact that PI(4,5)P 2 , contrary to PI(4)P (or PS), cannot be entirely buried in the pocket and capped by the lid. This hypothesis was suggested by structural analyses of ORP1 and ORP2 in complex with PI(4,5)P 2 34,35 and confirmed by our in vitro assays. Because PI(4)P and PI(4,5)P 2 co-exist in the PM, this strongly suggests that only PI(4)P is used by ORP5 or ORP8 for the exchange with PS in the cell. Our data also suggest that ORP1, ORP2 and ORP4L 57 cannot trap PIPs other than PI(4)P in cells or only if they operate on organelle membranes devoid of PI(4)P.
Quite interestingly, the comparison between the transfer rates and affinity determined with Osh6p and ORD8 for various lipid species allows us to infer general rules that can serve to better understand the cellular activity of LTPs. A first lesson is that, in a simple transfer process between membranes, a lowaffinity ligand can be transferred more rapidly down its concentration gradient than a high-affinity ligand.
This was observed when comparing saturated PS with unsaturated PS or PI(4,5)P 2 with PI(4)P. Presumably, as suggested by our kinetic model, it lies in the fact that a low affinity recognition process can be detrimental when the ligand is extracted from a donor membrane but an advantage in the transfer process, notably by preventing any re-extraction of the ligand from the acceptor membrane.
The picture is different with a membrane system of higher complexity that reconstitutes a cellular context more faithfully. In exchange conditions, the ability of Osh6p and ORD8 to saturated PS was poorly enhanced or often inhibited when PI(4)P was present as a counterligand, because these LTPs have a higher affinity for the latter. When PI(4)P and PI(4,5)P 2 were present in the same donor membrane, these LTPs preferentially transferred PI(4)P, for which they have the highest affinity, to acceptor membranes. In an even more complex system where PS, PI(4)P and PI(4,5)P 2 were present each in distinct liposome populations, mimicking three cellular compartments, ORD8 used PI(4)P as a counterligand to transfer PS between two membranes.
These observations have important implications. They suggest first that some caution must be exercised when analyzing in vitro data using membranes of low compositional complexity: measuring a fast transfer rate for a given lipid ligand and an LTP does not necessarily mean that this ligand is the true cellular cargo of the LTP. When considering a cellular context, one can assume that a mere lipid transporter preferentially recognizes and transfers its high-affinity lipid ligand. This potentially implies, as suggested in vitro, a lower speed of transfer but at the benefit of a higher accuracy as no fortuitous ligand can be taken.
However, this limitation in terms of speed is lifted if the LTP can exchange this high-affinity ligand for a second one, as measured with Osh6p and ORP8 using unsaturated PS and PI(4)P. This can primarily be explained by the fact that this second ligand prevents the re-extraction by the LTPs of the other ligand from its destination compartment, which improves its net delivery. These observations on Osh6p and ORP8 confirm very first data on the sterol/PI(4)P exchange capacity of Osh4p 33 . Of note, our experiment and models suggest that an optimal exchange occurs, i.e. with similar PS and PI(4)P transfer rates along opposite directions, when a lipid exchanger has a similar affinity for each ligand. Interestingly, the experiments with PS, PI(4)P and PI(4,5)P 2 even suggest that this exchange process can channel the lipid flux between two membrane-bound compartments if there are more than two compartments, such as in a cell. Collectively, our study supports the notion that lipid exchange processes are mechanisms that ensure fast, accurate and directional transfer of lipids between organelles.

Acknowledgments
We wish to thank Pr. Pietro de Camilli for providing the plasmid coding for the GST-ORD8 construct, Dr.
Enrique Castano for the plasmid coding for the GST-PH PLCδ1 and Dr. Fabien Alpy for the plasmid coding for the mCherry-D4-His 6 . We thank Dr. Frédéric Brau for his help in image analysis. We are grateful to Ms. Y.

Competing interests
The authors declare no competing interests.

Figure 5. Analysis of the relation between ORD's affinity for PS and PI(4)P and its ability to transfer these lipids between membranes.
(A) Description of the kinetic model. Osh6p or ORD8 (ORD) interacts with the same affinity with two distinct membranes A and B, each harboring a PS and PI(4)P pool, and can extract and release PS and PI(4)P. ORD-PS and ORD-PI(4)P correspond to the ORD in 1:1 complex with PS and PI(4)P, respectively. All k ON and k OFF rates were set to 10 µM -1 .s -1 , and 1 s -1 , respectively, unless otherwise specified. (B) Initial PS transfer rate (grey dots) as a function of k ON-PS values (ranging from 0.01 to 10,000 µM -1 .s -1 ), under the condition where the A membrane initially contained 5% PS and B membrane was devoid of PI(4)P (non-exchange condition). Initial PS (pink dots) and PI(4)P transfer rates (orange dots) were also calculated as a function of k ON-PS , considering that PS and PI(4)P were initially present at 5% in the A and B membranes, respectively (exchange condition). PI(4)P transfer rate simulated with A membrane devoid of PS (non-exchange condition) was indicated by a dashed line. The grey areas correspond to regimes where k ON-PS > k ON-PI(4)P i.e., the ORD has more affinity for PS than PI(4)P. The acceleration factors, calculated for PS and PI(4)P, correspond to the ratio (in log value) between the transfer rates derived from simulations performed in exchange and non-exchange conditions. (C)  nm. Incremental amounts of liposome, containing 5% PI(4)P or PI(4,5)P 2 were injected to the reaction mix.

Acceleration factors of PS and PI(4)P transfer in exchange conditions
The signal was normalized considering the initial F max fluorescence, prior to the addition of NBD-PScontaining liposomes, and the dilution effect due to liposome addition. Data are represented as mean ± s.e.m.       (A) Initial PS (pink dots) and PI(4)P transfer rates (orange dots) were calculated as a function of k ON-PS for different k ON -PI(4)P values (10 µM -1 .s -1 , as shown in Figure 5, but also 40, 75 and 100 µM -1 .s -1 ), considering that PS and PI(4)P were initially present at 5% in the A and B membranes, respectively (exchange condition). The   Experiments were performed at 30°C as described in Figure 6A
Their concentration was determined by UV spectrometry.
To prepare NBD-labelled PH PLCδ1 , an endogenous, solvent-exposed cysteine at position 48 of a GST- Biosciences.

Liposomes preparation
Lipids stored in stock solutions in CHCl 3 or CHCl 3 /methanol were mixed at the desired molar ratio. The solvent was removed in a rotary evaporator under vacuum. If the flask contained a mixture with PI(4)P and/or PI(4,5)P 2 , it was pre-warmed at 33°C for 5 min prior to creating a vacuum. The lipid film was hydrated in 50 mM HEPES, pH 7.2, 120 mM K-Acetate (HK) buffer to obtain a suspension of multi-lamellar vesicles. The multi-lamellar vesicles suspensions were frozen and thawed five times and then extruded through polycarbonate filters of 0.2 µm pore size using a mini-extruder (Avanti Polar Lipids). Liposomes were stored at 4°C and in the dark when containing fluorescent lipids and used within 2 days.

Lipid transfer assays with two liposome populations
Lipid transfer assays were carried out in a Shimadzu RF 5301-PC or a JASCO FP-8300 spectrofluorometer.

NBD-PS based competition assays
In a cylindrical quartz cuvette, Osh6p or ORD8 was diluted at 240 nM in a final volume of 555 µL of freshly degassed and filtered HK buffer at 30°C under constant stirring. Two minutes after, 30 µL of a suspension of DOPC liposomes containing 2% NBD-PS was added (100 µM total lipid, 1 µM accessible NBD-PS). Five minutes after, successive injections of 3 µL of a suspension of DOPC liposomes enriched with a given PS or PI(4)P species (at 5%) were done every 3 minutes. Tryptophan fluorescence was measured at λ = 340 nm (bandwidth 5 nm) upon excitation at λ = 280 nm (bandwidth 1.5 nm). The signal was normalized by dividing F, the signal measured over time, by F 0 , the signal measured prior to the addition of the NBD-PS-containing liposome population, and corrected for dilution effects due to the successive injections of the second population of liposome. The signal between each liposome injection was averaged over 2 min to build the binding curve as a function of concentration of accessible non-fluorescent PS or PI(4)P species (from 0 to 1.25 µM).

Thermal shift assay
The

CPM accessibility assay
The day of the experiment, 100 µL from a stock solution of Osh6p(noC/S190C) construct was applied onto an illustra NAP 5 column (Cytivia) and eluted with freshly degassed HK buffer, according to the

Microscopy and image analysis
One day after transfection, the cells were observed in live conditions using a wide-field microscope