Membrane protein sequestering by ionic protein-lipid interactions

Phosphoinositides are lipids that contain an inositol headgroup conjugated to 1–3 phosphate groups. With ~1% of total lipids in the inner leaflet of the plasma membrane⁴, PIP2 is the most abundant phosphoinositide. Earlier studies identified PIP2 as a second messenger in the phospholipase-C pathway. However, the list of cellular functions of PIP2 is rapidly growing, and PIP2 is also involved in membrane targeting, cytoskeletal attachment, endocytosis and exocytosis⁴. PIP2 interacts with many different proteins, either via unstructured basic residue-rich regions or via more structured domains^4,5.

Neuronal exocytosis requires plasma membrane PIP2^2,6-8. PIP2 levels at the plasma membrane determine the rates of vesicle priming, the size of the readily releasable pool, and the rates of sustained exocytosis in stimulated cells^2,6,8. This regulation is probably mediated by interactions of PIP2 with proteins involved in docking and fusion such as rabphilin, CAPS, synaptotagmin, SCAMP2 and Mints^6,9. In docking, PIP2 clusters may act as molecular ‘beacons’ that target synaptic vesicles to the fusion sites. Indeed, PIP2 is locally enriched at the sites of docked vesicles and colocalizes with at least 5–10% of the microdomains of syntaxin-1A (Supp. Fig. 1)^2,3,9, the membrane-anchored t-SNARE of neuronal exocytosis¹.

The amount of PIP2 at the sites of membrane fusion in PC12 cells has been estimated at 3–6% PIP2 of surface area (Supp. Fig. 2)⁹. In these experiments, membrane sheets were specifically stained for PIP2 with the PH domain of protein lipase C delta fused to GFP⁹ or citrine (a YFP analog¹⁰; PH_PLCδ-citrine, Supp. Fig. 1–2), and the fluorescence from the punctuated PIP2-microdomains was quantified. However, as explained in reference⁹, this approach underestimates the fraction of PIP2 if the size of the PH_PLCδ-microdomains is smaller than the ~200 nm diffraction limited resolution of conventional fluorescence microscopy. To obtain a more accurate estimate, we re-analyzed PC12 membrane sheets labeled with PH_PLCδ-citrine or an antibody raised against PIP2 using super resolution STED (stimulation emission depletion) microscopy¹¹ (Fig. 1a–c). These experiments revealed that the PIP2 stained clusters are much smaller than anticipated, with an average diameter of only 73 ± 42 nm (s.d.). Although this is still a higher estimate since it represents the microdomain size convoluted with the resolution of the STED microscope (~60 nm), it is in good agreement with the size of the syntaxin-1A microdomains¹². Using this value, we re-calculated the surface density of PIP2 (Supp. Methods). For this calculation, we first estimated the total amount of PIP2 in a microdomain when sampled with the diffraction limited resolution of our epi-fluorescence microscope (Supp. Fig. 2; Fig. 1d, black curve). We then calculated the peak concentration when this PIP2 was concentrated into 73 nm microdomains (Fig. 1d, red curve). Here, we assumed a Gaussian distribution of PIP2 in the microdomains. A peak surface density of 82% PIP2 was obtained (Fig. 1d). It needs to be kept in mind that (i) at these high PIP2 concentrations, molecular crowding might hinder binding of PH_PLCδ-citrine, (ii) relatively small errors in microdomain size and resolution of the microscopes result in substantial errors, and (iii) PH_PLCδ-citrine and antibody binding may alter PIP2 localization and is only indicative of PIP2 microdomains. Nevertheless, our values are much higher than any previous estimate and it seems safe to conclude that PIP2 is the dominant inner-leaflet lipid in the microdomains. The question then arises by which molecular mechanism such high concentrations of PIP2 are achieved.

PIP2 has a net negative charge of -3–5⁴ and interacts with polybasic stretches of amino acids^4,5,13,14. Proteins with such stretches can sequester PIP2 even in excess of monovalent anionic lipids, such as MARCKS, spermine and even pentalysine (Lys₅)^5,14. Similar to these proteins, syntaxin-1A also possesses a stretch of basic amino acids. These residues are adjacent to the transmembrane domain and are in contact with the head-groups of the phospholipids (Supp. Fig. 3a)^15,16. Indeed, it is well established that this conserved stretch with 5 positive residues (260-KARRKK) interacts with PIP2^9,15-17. Removal of charge diminishes this interaction (Supp. Fig. 3b–c), but syntaxin-1A remains capable of fusing membranes even upon removal of all 5 charges^9,16. Because PIP2 colocalizes with at least a fraction of syntaxin-1A microdomains (Supp. Fig. 1)², we speculated that their interaction might drive domain formation similar to various soluble lipid binding proteins^5,14. Two independent approaches were used to test this hypothesis: (i) reconstitution in giant unilamellar vesicles (GUVs)¹⁸, and (ii) hydrolysis of PIP2 in PC12 cells using a membrane-targeted variant of the PIP2 phosphatase synaptojanin-1⁷.

Murray and Tamm¹⁷ showed syntaxin-1A clustered in a non-raft way in neutral cholesterol:phosphatidylcholine (PC) membranes. Here, cholesterol clusters syntaxin-1A by competing for solvation by PC. Indeed, a synthetic C-terminal peptide of syntaxin-1A (residues 257–288; 3 mol%; Fig. 2, Supp. Fig. 4a) clustered in domains in > 50% of GUVs composed of DOPC (1,2-dioleoyl-sn-glycero-3-PC) with 20 mol% cholesterol. This peptide contained both the polybasic juxtamembrane linker and transmembrane region and was N-terminally labeled with either rhodamine red or Atto647N. Analysis of fluorescence showed a 1.6 ± 0.2 (s.d.; n = 18) fold enrichment of syntaxin-1A_257–288 in these clusters, but this lower estimate is limited by the optics. Negatively charged PIP2 or DOPS (1,2-dioleoyl-sn-glycero-3-phosphatidylserine) dispersed these clusters (Fig. 2)¹⁷. Thus, while cholesterol competition might explain syntaxin-1A clusters that are not enriched in PIP2 (Supp. Fig. 1), they cannot explain the high accumulation of PIP2 at the sites of docked vesicles. However, 1.5 mol% (total lipids) PIP2 also clustered syntaxin-1A in 1–10 μm domains in 1–5% of the GUVs (Fig. 2, Supp. Fig. 4b–c). These domains did not depend on cholesterol or DOPS. In these domains, PIP2 was 1.9 ± 0.2 (s.d.; n = 13; Supp. Fig. 5) and syntaxin-1A_257–288 5.5 ± 1.4 (s.d.; n = 27) fold enriched based on fluorescence. Importantly, no domains were observed without peptide or when the PIP2 concentration exceeded 5 mol%. Divalent cations can act as bridges between two adjacent lipids and induce aggregation of PIP2 into clusters^19-21, but even 1 mM Ca²⁺ was not sufficient to compete syntaxin-1A away from the microdomains. Domains were present with both synthetic dioleoyl-PIP2 and with PIP2 extracted from pig brain (Supp. Fig. 4b). Thus, syntaxin-1A can be clustered in the membrane both by cholesterol and PIP2.

These cholesterol- and PIP2-mediated clusters both differ from ‘rafts’. They also differ from each other. First, PIP2-domains are always round and only 1–2 per vesicle, whereas cholesterol generally (but not always) induces many small domains (Supp. Fig. 4). Second, fluorescence recovery after photobleaching showed that syntaxin-1A remained mobile in the PIP2-domains while syntaxin-1A was essentially immobile in the cholesterol-dependent clusters (Supp. Fig. 6). Syntaxin-1A thus diffuses in the PIP2-domains and forms large circular domains for minimizing boundary energy²¹. Third, 6-dodecanoyl-2-dimethylaminonaphthalene (Laurdan)²² showed a high hydration of the PIP2-domains, whereas the cholesterol domains were much denser packed (Supp. Fig. 7). Fourth, phase contrast microscopy showed a thickening of the cholesterol-dependent clusters, but not of the PIP2-domains (Supp. Fig. 8). Thus, even though no saturated lipids are present, the cholesterol-dependent domains show behavior that essentially resembles the Lo phase. In contrast, the PIP2-domains seem much more disordered and resemble the Ld phase. Ca²⁺ demixing of polyanionic amphiphiles showed that electrostatic interactions can indeed lead to liquid-like domains²¹.

The transmembrane helix of syntaxin-1A has been reported to homodimerize. However, introducing the M267A C271A I279A mutations that prevent homodimerization of the syntaxin-1A peptides²³ did not prevent cholesterol or PIP2 mediated clustering (Supp. Fig. 9). In contrast, no PIP2-domains were observed when two charges (K264A K265A) from the polybasic linker were removed, but cholesterol-dependent clusters were still observed (Supp. Fig. 5). Overexpression of the C-terminal part of syntaxin-1A fused to GFP²⁴ in PC12 cells also showed 4–8-fold loss of clustering of the K264A K265A mutant (Supp. Fig. 10–11). These data show that electrostatic interactions between PIP2 and the juxtamembrane helix of syntaxin-1A are sufficient for domain formation.

We then set out to investigate to what extent PIP2 is required for syntaxin-1A clustering in PC12 cells. For this purpose, we expressed a RFP-tagged construct containing the phosphatase domain of synaptojanin-1 fused to a CAAX-box, resulting in its targeting to the plasma membrane⁷. Synaptojanin-1 is a polyphosphoinositide 5-phosphatase, and the expression of the construct completely removed PIP2 from the plasma membrane (Supp. Fig. 12)^6,7. Importantly, synaptojanin-1 expression 3.7-fold reduced the punctuate distribution of endogenous syntaxin-1A (Fig. 3; Supp. Fig. 13). Thus, this provides evidence that PIP2 is indeed required for at least part of syntaxin-1A microdomain formation.

We performed molecular dynamics simulations to gain insight in the precise conformation of the PIP2-syntaxin-1A microdomains. In these coarse-grained simulations, several atoms are represented by one simulation bead^25,26 (Supp. Fig. 14). This allowed for simulations of relatively large lipid bilayers of ~2,500 copies of a 4:1 molar ratio of DOPC:DOPS and 40–64 copies of syntaxin-1A_257–288 and PIP2. Within 10 μs simulation time, up to 10 copies of syntaxin-1A_257–288 clustered with PIP2 into microdomains (Supp. Fig. 15). Equal amounts of PIP2 and syntaxin-1A were present in the bulk-phase of those domains, while more PIP2 and DOPS associated transiently to the periphery. We used this information to construct a domain with 64 copies of syntaxin-1A (Fig. 4, Supp. Movie 1), which is comparable to the syntaxin-1A content in the microdomains in PC12 cells¹². These domains were stable over 6 μs simulation time and contained <10% residual DOPC or DOPS. Together, we conclude that syntaxin-1A and PIP2 can form dynamic, amorphous networks with PIP2 acting as a ‘charge-bridge’ and spanning the distance between the various syntaxin-1A molecules (Fig. 4c).

In summary, our findings show that electrostatic interactions between the membrane lipid PIP2 and the SNARE syntaxin-1A suffice to induce membrane sequestering and microdomain formation without the need for high local PIP2 production or a (complex) ‘molecular fence’ restricting PIP2 and protein diffusion²⁷. This does not exclude an additional role for protein-protein interactions between either transmembrane helices or soluble domains. In fact, these seem essential for segregation of proteins in similar structure and size, such as syntaxin-1A and syntaxin-4 (both have polybasic regions and cluster separately)^12,24. The mutual enrichment of syntaxin-1A and PIP2 at the fusion sites by electrostatic interactions has clear advantages. First, accumulation of syntaxin-1A may facilitate SNARE interactions and thereby increase the membrane fusion efficiency^3,28. Second, the lipid environment modulates the energetic requirements for fusion^13,16. Third, both PIP2 and syntaxin-1A function as molecular docking sites and facilitate assembly of the complete fusion machinery^1,2,6-9. Our findings that electrostatic protein-lipid interactions are sufficient for membrane sequestering constitute a novel mechanism for the formation of protein microdomains in the membrane that is clearly distinct from the well-established lipid phases^18,29.