Self assembly of HIV-1 Gag protein on lipid membranes generates PI(4,5)P2/Cholesterol nanoclusters

The self-assembly of HIV-1 Gag polyprotein at the inner leaflet of the cell host plasma membrane is the key orchestrator of virus assembly. The binding between Gag and the plasma membrane is mediated by specific interaction of the Gag matrix domain and the PI(4,5)P2 lipid (PIP2). It is unknown whether this interaction could lead to local reorganization of the plasma membrane lipids. In this study, using model membranes, we examined the ability of Gag to segregate specific lipids upon self-assembly. We show for the first time that Gag self-assembly is responsible for the formation of PIP2 lipid nanoclusters, enriched in cholesterol but not in sphingomyelin. We also show that Gag mainly partition into liquid-disordered domains of these lipid membranes. Our work strongly suggests that, instead of targeting pre-existing plasma membrane lipid domains, Gag is more prone to generate PIP2/Cholesterol lipid nanodomains at the inner leaflet of the plasma membrane during early events of virus assembly.

The retroviral Gag protein drives the assembly process of the Human Immunodeficiency Virus type 1(HIV-1) particles 1,2 . This protein is synthesized as a polyprotein Pr55 Gag, which contains three major structural domains, namely matrix (MA), capsid (CA) and nucleocapsid (NC), as well as two spacer peptides, sp1 and sp2 and an unstructured C-terminus p6 peptide. Each of these domains are known to have a specific and distinct function during the viral assembly process. Importantly, the N-terminal MA domain targets Gag to the plasma membrane and mediates membrane binding, the CA domain is responsible for Gag-Gag interaction and self-assembly and the NC domain recruits the viral RNA that also acts as a scaffold for viral particle assembly 3 . Although it has recently been shown that HIV assembly may be initiated in the cytosol 4 , it is commonly accepted that the formation of large HIV-1 assembly complex mainly occurs at the plasma membrane (PM) of the virus producing cells. Two main features of the N-terminal MA domain of Gag govern the HIV-1 Gag inner leaflet PM binding: the N-terminal myristate and the highly basic region (HBR) that contains a specific binding pocket for the phosphatidyl inositol (4,5) bisphosphate lipid (PI(4,5)P 2 or PIP 2 ) 5 . PIP 2 has been shown to play a major role in HIV-1 assembly in cells since cellular depletion of PIP 2 decrease the efficiency of viral particle release 6 . In addition, PIP 2 is also enriched in the virus envelope relative to the PM of the virus-producing cells 7 . This lipidomic analysis also shows that HIV particles are enriched in sphingolipids and cholesterol 7 supporting the idea that the viral particles are released from the so called "rafts" domains. Historically, Gag has been shown to associate with detergent resistant membrane (DRM) by cell membrane fraction assays 8,9 . Nevertheless, the exact composition of these "rafts" or DRM is controversial 10,11 . Indeed, many different types of "rafts" may exist within the plasma membrane as long as they are enriched in cholesterol and sphingolipids 12 . However, Gag self-assembly occurs at the inner leaflet of the cellular PM where sphingolipids are poorly present. Moreover, natural PIP 2 , the main interacting partner of Gag, has a polyunsaturated fatty acyl residue at the sn-2 position of the glycerol backbone. This acyl chain polyunsaturation is not in favour of PIP 2 partitioning into rafts 13 . To circumvent such a constrain, Saad et al. suggested a model based on NMR data using truncated acyl chains PIP 2 5 . In this model, the sn-2 acyl chain of the PIP 2 is removed from the plane of the membrane and trapped into an hydrophobic pocket of MA. However, coarse grained dynamics studies 14 and new NMR experiments using full length acyl chain PIP 2 15 showed the opposite. Finally, recent experiments using either a multimerizable matrix domain of HIV-1 Gag 16 or RSV Gag 17

Results
Since PIP 2 , Chol and SPM have been shown to be enriched in the virus envelope 7 , the impact of Gag self-assembly on their lateral distribution was tested in different types of model membranes (Fig. 1A) exhibiting different lipid compositions. Large unilamellar vesicles (LUVs) were firstly used in order to easily control the protein over lipid molecular ratio. Given that the plasma membrane of HIV producing cells (~10 μm in diameter) has reduced level of curvature compared to these LUVs (100 nm in diameter), we have also used supported lipid bilayers (SLBs) as membrane models for analyses. Gag is known to bind specifically PIP 2 27 . We then started with a simple PIP 2 containing lipid mixture (PC/PS/PIP 2 hereafter called "basic") in which we introduced Chol or SPM (hereafter called "substituted basic"). Obviously, cellular PM have more complex lipid compositions. We therefore decided to extend our study to compositions mimicking either the inner leaflet cellular plasma membrane 28 (hereafter called "inner leaflet") or the "lipid rafts" domains (i.e. two separated phases, a liquid disordered (L d ) and a liquid ordered (L o ) enriched in cholesterol and sphingomyelin, hereafter called "rafts-mimicking" 16 ). Table 1 gives the exact compositions of the lipid mixtures used in this study. To assess a potential reorganisation of these lipids during Gag self-assembly, we used full-length Gag (FL-Gag) and different mutants. Theses mutants include: (1) a mutant of CA involved in CA-CA interaction and Gag self-assembly (WM) 29 (2) Gag lacking the C-terminus NC-sp2-p6 domains that is involved in NC-RNA association (P39). The membrane binding domain alone of Gag (MA) was also tested (see Fig. 1B). Finally, a cellular PM PIP 2 binding protein, the PH domain of EFA6 (PH-EFA6) 30 and a peptide, MARCKS (151-175) that is known to laterally redistribute PIP 2 24 were used as controls.
HIV-1 Gag efficiently binds model membranes containing PIP 2 . The ability of the different proteins and peptide to bind LUVs and SLBs was analysed using membranes of "basic" composition. For this purpose, we monitored protein binding with two different techniques. On one hand, we performed LUV cosedimentation assays at fixed protein concentration (1 μM) and increasing PIP 2 concentrations. After 15 min of incubation, LUV bound proteins were separated from unbound proteins by ultra-centrifugation. The pellet (P) and the supeRNAtant (S) were deposited on a denaturating gel to quantify the amount of bound (P) and unbound (S) proteins ( Fig. 2A). On the other hand, we used Quartz Crystal Microbalance (QCM) to monitor the association of proteins to SLBs containing a fixed PIP 2 concentration. SLBs were formed onto the quartz crystal. Proteins were then injected at increasing concentrations (from 10 −2 to 10 μM) and the change in the resonance frequency of the quartz crystal was monitored (Fig. 2B). This change in frequency is directly connected to the mass increase at the surface of the quartz crystal. The ratio of membrane bound to unbound protein is linked to an apparent partition coefficients (K p ). K p were determined by fitting, with eq. 1, the plots of membrane bound protein with either increasing concentrations of lipids ( Fig. 2C) or increasing concentration of protein (Fig. 2D). Except for MA, K p obtained for other proteins were less than 1 μM. The results, summarized in Table 2, are in good agreement with the previously obtained K p values for some of the different proteins used in this study. Interestingly, the K p values of MARCKS and PH-EFA6 (0.7 < K p < 0.8 μM) were found to be in the range of the one obtained for Gag (FL) and its mutants (0.2 < K p < 0.8 μM). This shows that the ratio of membrane bound over total protein will be equivalent for all these proteins, allowing, therefore, direct comparisons of their respective role on the lateral sorting of PIP 2 after membrane binding.

FL-Gag is mainly partitioning to L d phases and does not induce micrometer phase separation.
We firstly monitored the possible existence of Gag induced micrometer-range PIP 2 domains by imaging the surface distribution of a fluorescently labelled PIP 2 (namely Top-Fluor PIP 2 or TF-PIP 2 ) and a fluorescently labeled FL-Gag (FL-Gag Alexa 549 or A549-FL-Gag) in "basic" GUVs. As expected, both TF-PIP 2 and A549-FL-Gag showed a uniform distribution over the surface of the GUV (Fig. 3A). The same uniform distribution was also observed with "inner leaflet" GUVs ( Fig. 3A). Spatial auto-correlation of the fluorescence of either PIP 2 or FL-Gag (Fig. 3B) both exhibited fast decorrelation showing that the fluorescent labelling of PIP 2 is not inducing micrometer partitioning per se and that, if they exist, the PIP 2 clusters are less than 200 nm size (i.e. below the diffraction limit). Interestingly, labelled FL-Gag was essentially restricted to L d phases in the "rafts-mimicking" GUVs ( Fig. 3C and D), confirming what was already observed by Keller et al. 16 with a multimerized derivative of MA. This strongly suggested that FL-Gag preferentially binds and remains in non "rafts"/L o lipid phases.
HIV-1 Gag self-assembly induces PIP 2 nanoclustering. The lack of micrometer scale PIP 2 induced domains lead us to monitor, using self-quenching experiments, a possible PIP 2 lateral redistribution upon Gag self-assemby on LUVs or SLBs. As stated in the introduction, self-quenching occurs when two fluorescent molecules are close to each other (3 < r < 10 nm). To perform these experiments, we first controlled that increasing concentrations of MARCKS induced an increasing quenching of the TF-PIP 2 as it was reported before with Bodipy TMR-PIP 2 (BT-PIP 2 ) 24 (See SI Fig. S1). We then monitored the change in TF-PIP 2 fluorescence after addition of Gag or its mutants on "basic" LUVs in order to correctly control the protein to accessible PIP 2 ratio  Table 1 and SI Table S1 for all the different lipid compositions used in this study). (B) Schematic representation of the different Gag mutants used in this study, emphasizing the mutations and the differences in the domains present for each mutant.   Table 2.    reorganizing its lateral distribution, did not produce any effect either ( Fig. 4A and B). A major difference between the Gag matured protein MA and the precursor proteins (FL-Gag, P39 and WM) is their capacity to form large oligomer complexes via the CA-CA interactions 31 , although WM is less efficient (two orders of magnitudes in solution) in supporting the formation of large Gag lattice, due to its inability to dimerize with nearby CA hexamer 29 . Interestingly, this fluorescence unquenching was always less efficient in the case of WM compared to FL-Gag (from 3 to 5 times), indicating that this fluorescence variation depended on the capacity of Gag to multimerize. To support this hypothesis, FL-Gag and WM were incubated with "basic" LUVs for 15 minutes followed by ultracentrifugation. The self-assembly states of WM and FL-Gag bound to LUVs were analysed by performing a non denaturating gel electrophoresis to preserve existing multimers, followed by immunoblot. Figure 4C showed that, in the two P PIP / acc 2 ratio tested, WM was at least two times less efficient than FL-Gag in multimerizing on membranes (Fig. 4B). This ratio was in the range of the relative efficiency of TF-PIP 2 unquenching observed between FL-Gag and WM. This confirmed the role of Gag self-assembly in the observed TF-PIP 2 fluorescence changes. Although LUVs provide a simple, reliable and easy to control (in terms of protein to lipid ratios) model membranes, Gag self-assembly usually occurs on membranes with higher radius of curvature. Therefore, time course fluorescence self-quenching of TF-PIP 2 were also conducted on "basic" SLBs. For FL-Gag and P39 the TF-PIP 2 fluorescence intensity of SLBs decreased (Fig. 4D). This change in intensity was opposite to the one observed with LUVs (Fig. 4A) while still being a function of the protein concentration (Fig. 4E for FL-Gag). As it was the case for LUVs, the TF-PIP 2 quenching induced by addition of the WM mutant was less efficient than in the case of FL-Gag. PH-EFA6 and MA did not induced changes in TF-PIP 2 fluorescence (Fig. 4D).
Interestingly, Fig. 4D shows that MARCKS addition induced the same TF-PIP2 quenching on SLBs and LUVs (see Fig. 4A). This suggest that model membrane curvature (curved LUV vs flat SLBs) is responsible for the opposite quenching observed with Gag and its mutants on SLBs and LUVs (see SI Fig. S2 for detailed explanation). Altogether these results show that the self-assembly of HIV-1 Gag is generating TF-PIP 2 clusters.
Diffusion of Gag is expected to slow down upon self-assembly. Hendrix et al. 4 already observed that Gag multimers exhibited a diffusion coefficient(D) of 0.01 μm 2 .s −1 at maximum. If PIP 2 s are tightly trapped in Gag multimers they will diffuse with similar coefficient as that of these Gag multimers. Consequently, the fluorescence of a 1 μm radius area containing these multimers and labelled PIP 2 , should not totally return to its initial level 15s after photobleaching (Fig. 4F). This will lead the normalized fluorescence recovery value (NFR, defined in eq. 4, 0 < NFR < 1) to be less than 1. We monitored this change in normalized fluorescence recovery in the absence or presence of FL-Gag. For that purpose, in the time course of quenching experiments, we performed Fluorescence Recovery After Photobleaching (FRAP) experiments before (start of the quenching experiment) and after (end of the quenching experiment) addition of FL-Gag on SLBs. In the absence of FL-Gag, normalized fluorescence recovered to 1, as expected for freely diffusing lipids (Fig. 4G, right). On the opposite, addition of increasing FL-Gag concentrations induced a decrease in PIP 2 normalized fluorescence recovery as well as an increase in fluorescence quenching (previously shown in Fig. 4E). We controlled that the decrease in normalized fluorescence recovery was specific of PIP 2 . Indeed, we did not observe any change in the normalized fluorescence recovery of a lipid analogue fluorescent dye (diIC18, see materials in SI) in the presence or in the absence of Gag (Fig. 4G, left).
Our data clearly show that HIV-1 Gag is sorting PIP 2 in the lipid membrane and that Gag self-assembly generates PIP 2 nanoclusters in model membranes.

Cholesterol but not sphingomyelin is sensitive to HIV-1 Gag self-assembly. Since Chol and SPM
are supposed to be the main lipid components of "raft" domains, their enrichment into the Gag self-assembly induced PIP 2 nanoclusters was also assessed. 2% mol of egg phosphatidyl choline (EPC) present in our "basic" lipid composition, were substituted either by SPM or by Chol with half of them being labelled with TF derivatives (see Table 1 for detailed composition). This substitution allowed the net surface charge and the PIP 2 content to be maintained, limiting any drastic change in the partitioning constant K p . Time course fluorescence change of either TF-SPM or TF-Chol were then monitored after FL-Gag addition. Figure 5A shows that FL-Gag addition had no effect on TF-SPM fluorescence whereas Fig. 5B shows that increasing concentrations of FL-Gag induced increasing quenching of TF-Chol, as it was the case for TF-PIP 2 (Fig. 4E). This shows that FL-Gag self-assembly is able to generate Chol-enriched lipid nanodomains whereas it is not changing the SPM lateral distribution. Co-clustering of these molecules was also assessed by simultaneously monitoring the change in fluorescence upon addition of FL-Gag with either TF-SPM and BT-PIP 2 (Fig. 5C) or TF-Chol and BT-PIP 2 (Fig. 5D) labelled SLBs (see SI Table 1 for exact dual labelled lipid membranes compositions). TF-Chol and BT-PIP 2 exhibited simultaneous quenching after addition of various concentration of FL-Gag, suggesting that TF-Chol and BT-PIP 2 are co-clustered in Gag self-assembly induced nanodomains. In contrast, while BT-PIP 2 was quenched as previously, TF-SPM exhibited a surprising apparent fluorescence unquenching after Gag addition. This was an unexpected result regarding the TF-SPM quenching induced by FL-Gag self-assembly in the presence of unlabelled PIP 2 (Fig. 5A). Since this apparent unquenching of TF-SPM upon addition of FL-Gag only occurred when BT-PIP 2 was present (see Fig. 5A vs Fig. 5C), we checked if TF-SPM (as a donor) and BT-PIP 2 (as an acceptor) were able to exhibit FRET. Donor FRET efficiency has been shown to be a function of acceptor concentration in a lipid bilayer 32 . This means that the FRET efficiency will decrease if acceptor (BT-PIP 2 ) local concentration decreases and that, as a consequence, the fluorescence of the donor (TF-SPM) will increase. We tested this by performing acceptor photobleaching experiments in order to decrease the acceptor concentration, locally and reversibly. We observed that, in the area where the acceptor (BT-PIP 2 ) was bleached, the fluorescence of the donor (TF-SPM) increased. This effect was abolished when concentrations were re-equilibrated by diffusion (see SI Fig. 3). In our TF-SPM/BT-PIP 2 co-clustering assay, we observed a decrease in BT-PIP 2 due to Gag self-assembly induced quenching and concomitantly, a non reversible TF-SPM fluorescence increase due to loss in FRET efficiency induced by decreasing concentration of acceptor (TF-PIP 2 ). This can only be explained by considering that Gag self-assembly clusters BT-PIP 2 without TF-SPM.
HIV-1 Gag is driving PIP 2 and cholesterol nanoclustering independently of surrounding lipids. Because cellular PMs are complex lipid mixtures, we examined a possible role of this complexity on the lateral sorting of PIP 2 , Chol and SPM during HIV-1 Gag self-assembly. For that purpose, we used the two other different lipid mixture mimicking either the inner leaflet of cells PM ("inner-leaflet") or the "rafts" lipid mixture ("raft-mimicking") (see Table 1 and SI Table 1 for detailed lipid composition). We first compared the effect of FL-Gag self-assembly on PIP 2 clustering for these three lipid compositions. No characteristic change in TF-PIP 2 fluorescence unquenching with increasing concentrations of FL-Gag (LUVs, Fig. 6A) or fluorescence quenching (SLBs, Fig. 6B) could be clearly detected amongst the three different lipid compositions. We then tested again Chol and SPM ability to partition into these FL-Gag induced PIP 2 nanoclusters in the complex lipid mixtures.   These results show that, as in the "basic" composition, in complex lipid mixtures the PIP 2 nanoclusters induced by Gag self assembly are enriched in Chol but not in SPM. This clearly suggest that Gag is sorting these lipids during self-assembly independently of the surrounding lipids chemical nature and acts as the driving force for lipid reorganization during HIV-1 assembly.

Discussion
Although the MA domain is primary responsible for HIV-1 Gag binding to the PM, the ten times higher apparent affinity for membrane models observed here in the case of FL-Gag, P39 and WM confirms that the NC domains and CA-CA self-assembly of Gag are involved in membrane binding efficiency. Indeed, the NC domain of Gag alone has recently been described to bind to PIP 2 containing lipid membranes with an apparent K p as high as 7 μM 33 . It has also already been shown that the driving force for membrane association stems largely from ionic interactions between multimerized Gag and negatively charged phospholipids 34 .
Using micrometer scale phase separated GUVs, we have observed that Gag was mainly partitioning in L d phases, i.e., more likely outside of lipid "rafts" (namely Chol and SPM enriched) domains. The lack of myristate in the different Gag variants tested here could explain the Gag L d phase localization 35 . Nevertheless, our results are consistent with the L d phase GUV localization of multimerizable myr(+ )MA protein observed by Keller et al. 16 .
However, it is known that cholesterol is crucial for virus infectivity 36,37 and that Gag can sense cholesterol 38,39 . It was recently described that PIP 2 could form clusters in the presence of cholesterol alone 40 . Here we show that PIP 2 and cholesterol are laterally redistributed upon Gag self-assembly on membranes and participate to the formation of cholesterol/PIP 2 /Gag enriched nanodomains. Importantly, we also show that sphingomyelin is not sorted during HIV-1 Gag self-assembly and is excluded from the Gag/PIP 2 /Chol nanoclusters. Taken together, these results suggest that binding and self-assembly of Gag protein does not occur in pre-existing lipid domains (such as "rafts") but that this self-assembly is more likely to induce lipid nanodomains.
It is known that plasma membrane lipid organization is not only driven by lipid-lipid interactions. For example, cytoplasmic proteins such as ezrin 41,42 , syntaxin-1 43 have also been described to induce or interact with PIP 2 nanoclusters. As a general mechanism, proteins with basic interfaces can recruit acidic lipids that, in turn, can facilitate recruitment and clustering of these proteins into nanodomains 44,45 . A similar cooperative mechanism could also happen when HIV-1 Gag binds to PIP 2 during viral assembly at the plasma membrane. Interestingly, mesoscale organized structures such as cortical actin network has also been shown to play an important role in the lateral organization of not only transmembrane proteins, but inner leaflet plasma membrane lipids such as PIP 2 46 and more interestingly outer leaflet components such as GPi-anchored proteins 47 . Moreover, this cortical actin network has also been shown to play a role in HIV-1 Gag assembly at the PM of Jurkat T-cells 48 .
Since HIV-1 Gag has been often found in DRMs 8 , pre-existing outer leaflet "rafts" could be trapped by these nascent PI(4,5)P 2 -Gag nanodomains through transbilayer coupling -as we already proposed 18 or as it has been recently shown for lipid domains 49 and for outer leaflet GPi anchored proteins 50 . Moreover, a recent study of the SPM dynamic in the cellular PM revealed that instead of clearly partitioning into nanodomains, SPM was mainly transiently trapped 51 . In the case of the ganglioside GM1, another lipid described as partitioning into L o phases, its transient trapping was shown to depend on molecular pinning and interleaflet coupling between lipid tail domains 52 . This suggest that Gag could not only generates his own lipid nano-domains at the inner leaflet of the cellular PM but also induce the formation of domains in the outer leaflet of the cellular PM by transiently trapping SPM or other components.
Finally, we also observed that the matrix domain of Gag is not able to induce this PIP 2 clustering, suggesting that, after maturation and particle release, the inner leaflet lipids of the virus envelope might be free to diffuse again.

Conclusion
We have shown using simple and cell mimicking lipid composition model membranes that Gag self-assembly is inducing nano-clusters enriched in PIP 2 and cholesterol, instead of partitioning to pre-existing ones. This lipid nano-clustering does not require sphingomyelin and mainly occurs out of the L o phases in GUVs. Further, the different lipid composition tested here does not strikingly affect the capacity of Gag to induce these lipid nano-clusters suggesting that Gag is able to sort its own lipids independently of the surrounding lipids.

Methods
All experiments were performed at room temperature (RT). Experimental buffer was Hepes 10 mM, pH = 7.4, KCl 150 mM, EDTA 2 mM except for QCM-D binding experiments. All images were acquired with a Zeiss LSM 780 microscope (Carl Zeiss, Inc.) using a 63x NA 1.4 oil objective and quantified using Image J software (NIH, MD, USA). Detailed materials, methods for model membranes (LUV, GUV, SLB) preparation and for protein purification are in SI.
Binding experiments. K p were determined on basic lipid composition (EPC 68%, BPS 30% & PI(4,5)P 2 2% for LUVs or POPC 68%, POPS 30% & PI(4,5)P 2 2% for SLBs). Methods used were either cosedimentation assays for LUVs or QCM-D experiments on SLBs. Co-sedimentation assays were made at 1 μM protein concentration with varying concentrations of total accessible lipids from 0.07 to 2250 μM in 100 μl of the same buffer that the one used in quenching experiments, according to the protocol in ref. 53. After 15 min of incubation at room temperature, samples were centrifuged at 220,000 g for 1 h at 4 °C using a Beckman Coulter's TLA 100 rotor. The top 80 μl was considered as supeRNAtant (S) and the remaining 20 μl diluted with 60 μl of working buffer as pellet (P). Pellet and supeRNAtant were analyzed on a 10% SDS-PAGE and stained using coomassie blue. Quantifications were made using Image J software (National Institutes of Health, MD, USA). SLBs were prepared with 0.1 mg.mL −1 liposomes flowing at 10 μL.min −1 for 10-20 min on a UV-treated SiO 2 surface of Q-sensor fixed in a Q-Sense Flow module, QFM 401 Biolin Scientific, Sweden). Stable SLBs were rinsed with citrate buffer (NaCitrate 10 mM, 100 mM NaCl, and 0.5 mM EGTA, pH 4.6) and then with injection buffer (5 mM Tris & 100 mM NaCl pH 7.4). At equilibrium, 200 μL of increasing protein concentration was successively injected into the flow chamber followed by rinsing steps in between. The same was repeated with increasing protein concentrations until saturation. Sensorgrams were normalized to third harmonic in the case of varying harmonic curves. Δ F (plateau values) were used to measure the lipid SLB surface fraction of protein bound. The fraction of protein bound is related to an apparent association constant K (i.e, the reciprocal of the apparent partitioning constant, K p ) following the equation: Image analysis of GUVs. GUVs were deposited on coverslips coated with casein (2 mg.mL −1 ) fitted in an Attofluor cell chamber (Thermo Fisher Scientific, Inc.). Labeled FL-Gag was added to the buffer. Spectral images were acquired at constant excitation intensity at 488 nm, 561 nm and 633 nm using an emission spectral range between 499 and 690 nm with a 8 nm resolution. Linear unmixing of these images was done to avoid potential bleed-through due to fluorophore emission overlapping. To establish the spatial auto-correlation of the fluorescence intensity of TF-PIP 2 or A594-FL-Gag on basic composition GUVs, the change in intensity was plotted along the GUV using the following integration: ∑ ∑ Fluorescence quenching measurements. For LUV experiments, fluorescence was monitored using a spectrometer (Photon Technologies InteRNAtional, Inc.) with λ exc = 485 ± 2 nm and λ em = 520 ± 10 nm. Excitation lamp intensity was calibrated using the Raman spectrum of pure water and its fluctuations corrected every second. For every interaction assay on LUVs, the obtained intensity curve was corrected for both bleaching and dilution effect and then normalized to the mean intensity before injection. In the case of SLBs, fluorescence was monitored by acquiring images using the Zeiss definite focus system to avoid any z-drift with a 2-photons excitation (930 nm) every 5 s (in order to strongly reduce photobleaching). TF-lipids fluorescence was acquired at 520 ± 30 nm and BT-PIP 2 at 600 ± 40. Mean intensity of each image was normalized to the intensity before injection.
Self-assembly assay on LUVs. 100 μL of the basic LUVs was incubated at room temperature for 15 min with either 0.9 μM or 4.5 μM of FL-Gag or WM. After centrifugation at 10,000 g for 5 min at 4 C, 20 μL of the supeRNAtant was loaded on a 10% native-PAGE and proceed for immunoblot as in ref. 53. FL-Gag and WM were detected by a primary anti-capsid antibody (HIV-1 p24 NIH AIDS Reagents) followed by secondary antibody goat anti-mouse HRP conjugated. Membrane was revealed by Femto substrate (Thermo-scientific) and imaged by a G:Box (Syngene).
Fluorescence Recovery After Photobleaching. The image sequence was acquired at 20 Hz using the 488 nm line of an Ar + laser at a very low power to avoid photobleaching. After 2.5 s, 3 regions of interests (ROI), of 1 μm radius each, were rapidly photobleached (t < 60 ms) at maximal laser power. Fluorescence recovery was monitored for 15 s. The recovery curves were obtained as in ref. 54. In order to correctly estimate F 0 (the fluorescence intensity immediately after the end of the bleach) and F 15s , the curves were fitted as in ref. 54. The normalized fractional recovery (NFR), is defined as: