To investigate the mechanisms that enable ABCA1 to export PL, we inserted the cryo-EM structure of ABCA113 (wild type and mutated forms) into a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) membrane bilayer. Note the key structures of wild-type ABCA1: the outward-open transmembrane cavity, extracellular domains (ECDs), TMDs, and an elongated hydrophobic tunnel13 (Fig. 1A). We then used coarse-grained molecular dynamics (CGMD) to study interactions between PL and protein. This approach allowed us to monitor both the movements of individual PL molecules and changes in the structure of ABCA1 during CGMD simulations of wild type and two mutated ABCA1 structures.
POPC molecules extracted from the outer leaflet of the membrane diffuse through the opening in the outward-open transmembrane cavity of ABAC1
We performed three 10 µs replica CGMD simulations of wild-type ABCA1 structures inserted into a POPC bilayer. The final frames after each 2 μs simulation of the three replicas, converted to all-atom models, demonstrated that 3–5 POPC molecules extracted from the outer leaflet of the membrane had diffused through the opening in the outward-open transmembrane cavity of ABCA1—between TMD-1 and TMD-2—and into the interior (shown schematically in Fig. 1A). After each 2 µsec simulation, the extracted POPC molecules had assembled into a complex, termed the membrane mound, ~10 Å above the plane of the lipid bilayer (Fig. 1B). At the end of each of the three 10 µs replica CGMD simulations, the mound had collapsed into a single POPC molecule, indicating that the mound was transient.
Inspection of the cryo-EM structure revealed a loop of ABCA1 (residues 564–592) at the base of the elongated hydrophobic tunnel (Figs. 1A and 2A). This loop was contiguous with the outward-open transmembrane cavity on one side and the transverse hydrophobic tunnel on the other13. The large open loop contained 29 amino acid residues of which 5 were basic, 6 were acidic, 5 were aromatic and 7 were hydrophobic. Several of these charged residues were in close contact with the head groups of the POPCs that had diffused into the interior of the protein to form the membrane mound (Fig. 1B).
Extracted POPC molecules diffuse through the gateway toward the elongated hydrophobic tunnel of ABCA1
We term the loop formed by the residues the gateway for three reasons. First, POPC molecules from the membrane diffused partway into the loop toward the elongated hydrophobic tunnel of ABCA1. Second, four point mutations that cause Tangier disease (two shown not to affect cell-surface expression of ABCA1) are located in the loop (D571G, R587W, W590S/L) (Fig. 1C). Third, mutating positively and negatively charged amino acid side chains in the loop blocked the movement of membrane PL into the interior of ABCA1 in our simulations. They also inhibited PL and cholesterol export by ABCA1 from cultured cells (Fig. 3B).
The annulus regulates the movement of POPC into the elongated hydrophobic tunnel
The position of the elongated hydrophobic tunnel relative to the gateway is shown in Fig. 1A. A structure, termed the annulus (residues 69, 71–80, 363, 368–379), forms a compact domain at the tunnel’s base (Fig. 1A). By definition, this domain contains all residues at the base of the elongated hydrophobic tunnel within 10 Å of gateway residues (Figs. 1D and 2C). By mutating key residues, we determined that the central portion of this domain, termed the annulus orifice (Fig. 2D; residues 73–74, 77, 371, 375), helps regulate PL export by ABCA1.
To test if the charged residues of the gateway are important in formation of the lipid mound, we analyzed the 3 replicas of the 10 µs CGMD simulation of wild-type ABCA1 to quantify how often the nearest four charged residues of the gateway (K568, K570, D571, and D575) form salt bridges with POPC in the membrane mound. K568 formed salt bridges with the PO4– group of POPC 77% of the time, about twice as often as D571; D571 formed salt bridges with the choline group 39% of the time, 1.6 times as often as D575 (Supplementary Table 1). These observations suggest that K568 and to a lesser extent D571 are the major sites for initial salt-bridge formation between the gateway and POPC.
When the gateway/annulus complex (Fig. 2C) was rotated 90˚ around its x-axis, residues of the gateway could be easily visualized through the annulus orifice (Fig. 2D). Amino acid residues 73–74, 77, 371, and 375 formed the perimeter of the orifice (Figs. 1D and 2D). When the structure in Fig. 2D was rotated 180° around its y-axis (Fig. 2E) and the gateway was no longer visible (Fig. 2F), the annulus orifice faced the outward-open transmembrane cavity. The amino acid composition of the gateway–annulus orifice complex is illustrated in Fig. 2G.
Salt-bridge formation by the gateway mediates the extraction of POPC from the membrane mound
These observations raise the possibility that salt-bridge formation in the gateway of ABCA1 helps extract PL molecules from the outer membrane leaflet to create the membrane mound. We tested this hypothesis with CGMD simulations that determined how neutralizing charged residues in the gateway affected the extraction of POPC. For this experiment, we used single and multiple alanine substitutions (K568A and 11CmutA—mutation of all 11 charged residues of the gateway to alanine). The wild type and two mutated ABCA1s were then inserted separately into a membrane bilayer of POPC and subjected to 3 replica sets of CGMD simulation for 10 µs. After 10 µs CGMD simulation, the number of POPC molecules in the lipid mound was reduced about 50% by the K568A mutation and about 70% by the 11CmutA mutation (Supplementary Fig. 2D).
Examination of the frames in the last 5 µs of the three 10 µs CGMD simulations of wild-type ABCA1 revealed that a single POPC molecule was extracted from the outer surface of the plasma membrane (Fig. 4A–C). The PO4− group of the POPC formed salt bridges with K568 in 77% of the frames, with R565 in 3% of the frames, with K570 in 7% of the frames, and with R579 in 5% of the frames; no salt bridges were formed with R587 (Supplementary Table 1). During the simulation, POPC progressively moved 10, 12, 15, and 17 Å above the outer monolayer surface as it formed sequential salt bridges with D571, D575, E584 and D585, respectively (Fig. 4A–C). The Y573/W574 cluster at the bottom of the gateway also rearranged to shield the extracted POPC from the outer membrane monolayer as it moved toward the annulus (Fig. 4D–F).
ABCA1 with the K568A mutation also extracted a single POPC molecule from the outer leaflet of the membrane but had a remarkably different pattern of salt bridges from the wild type; the PO4− group of the POPC formed salt bridges with R565 in 5% of the frames, with K568 in 0% of the frames, with K570 in 0.2% of the frames, with R579 in 11% of the frames, and with R587 in 0% of the frames (Supplementary Table 1). The 11CmutA mutation exhibited a flat membrane surface with no salt bridges to POPC and no lipid extraction during the entire CGMD simulation (Fig. 3A, right panel).
To determine if charged amino acids in the gateway also help export PL from cells, mutations K568A and 11CmutA were created in constructs of human ABCA1. When wild-type and mutant ABCA1 proteins were expressed in BHK cells, the K568A mutation reduced PL and cholesterol efflux by APOA1 by ~65% and ~60%, respectively (Fig. 3B). The 11CmutA mutation caused ~90%–95% reduction. Similar results were observed with L-4F, an APOA1-mimetic peptide18 (Fig. 3B). Further, a control mutation, Y573F, had little effect on cholesterol or PL efflux. It is important to note that none of the mutations affected cell-surface expression of ABCA1 (Fig. 3B).
To test the possibility that the gateway’s charged and aromatic residues help translocate POPC to the annulus orifice, we conducted two additional mutagenic experiments. Mutation of a cluster of three negatively charged residues (D581K/E584K/D585K) to lysines in the gateway near the annulus reduced cholesterol and PL efflux by 80%–85%. The double mutation F583K/W590E in the upper gateway reduced cholesterol efflux by 95% (Fig. 3B). Neither mutation affected cell-surface expression of ABCA1 (Fig. 3B). Thus, the interactions between the charged residues and hydrophobic residues with POPC that we identified in our simulations fit the extracellular phospholipid translocase model well (Fig. 4).
POPC movement unwinds the annulus
There are potentially three energy barriers to spontaneous movement of POPC through the annulus orifice into the elongated hydrophobic tunnel. The first is the energetic cost of exposing the acyl chains of POPC to solvent as they move up the gateway to the annulus. The second is the movement of the charged headgroup of POPC through the hydrophobic outward-closed transmembrane cavity and into the annulus orifice. The third is the barrier that would be posed by Van der Waals forces if POPC tried to pass through the initially narrow annulus orifice. Overcoming these possible energy barriers would require some form of power stroke, such as closing the outward-open transmembrane cavity.
Using steered molecular dynamics (SMD), we simulated the movement of a POPC molecule from the membrane mound to the annulus and the elongated hydrophobic tunnel (Fig. 1D–F and Supplementary Fig. 1). The PO4– moiety of the POPC initially formed a salt bridge with residue K568 of the gateway, while the N(CH3)+ moiety formed a salt bridge with D571 (Figs. 1D and 2B). The POPC monomer then migrated through the gateway (Figs. 1D–F and 2D–F) to the annulus at the mouth of the elongated hydrophobic tunnel (Supplementary Fig. 1), contacting residues W590 and R587 of the gateway along the way.
We then used SMD to determine how the annulus changed its structure as extracted POPC moved from the outward-open transmembrane canal through the annulus and into the elongated hydrophobic tunnel. A single POPC molecule (Fig. 4G, H) initially interacted with the gateway (Fig. 4A) and was then translocated through the orifice as its N(CH)3+ headgroup made Van der Waals contact with the annulus. Two points are noteworthy: (i) two aromatic residues, F583 and W590, created a surface for aligning the POPC acyl chains parallel to the two helices of the annulus (residues 69–80 and 367–379); and (ii) charged residues formed salt bridges with each other to shield the POPC acyl chains on two sides (Fig. 4H).
The conformational changes in the annulus orifice seen during SMD simulations are summarized in Supplementary Fig. 3 and Supplementary Movie 1. We used two pulling velocities (fast, 50 Å/ns and slow, 0.5 Å/ns) for the simulations. The force versus time profile for the two pulling velocities are shown in Supplementary Fig. 4. In both cases POPC passage required input of force. In all of the simulations the choline headgroup of POPC made initial Van der Waals contact with the annulus orifice; then its acyl side chains entered the elongated hydrophobic cavity. As the POPC molecule moved through the orifice the helical region near residue I371 in the annulus partially unwound. The unwound segment regained its helical conformation as the POPC moved through the orifice into the elongated hydrophobic tunnel. The same partial uncoiling of the annulus domain was seen transiently in each of the SMD simulations with each pulling velocity. Hydrogen bonds formed between the PO4− group and the exposed helical NH backbone groups at the 2.0 and 7.5 ns timeframes (Supplementary Fig. 5).
These observations suggest that PL can move from the outward-open transmembrane cavity into the elongated hydrophobic tunnel with conformation changes in the annulus requiring energy input from ATP. To test this hypothesis, we generated three different double mutations in ABCA1 that were predicted to prevent the annulus from opening because of crosslinking by disulfide bond (I371C/L375C, I74C/I371C) or salt-bridge formation (I74K/I371E). We then expressed the wild-type and mutated ABCA1s in BHK cells, and quantified PL and cholesterol efflux. All three double mutations reduced lipid efflux by ABCA1 by 85–90% compared with wild-type ABCA1. Importantly, none of the three mutations affected cell-surface expression (Fig. 5).
Our molecular modeling of ABCA1 suggested that a single I371C mutation or a double i and i + 4 V304C/V308C mutation in the residues adjacent to the annulus orifice would not significantly affect lipid efflux (Fig. 5). When expressed in BHK cells, neither of those mutations in ABCA1 significantly impacted either PL or cholesterol export. Unlike C74–C371 or C371–C375, residue C371, though hydrophobic, is unable to form a disulfide bond to blockade the annulus. C304-C308, which are immediately adjacent to—but not part of—the annular orifice, can form a disulfide bond, but that had little effect on efflux.
The ATP-free structure of ABCA1 and the homology model of the ATP-bound ABCA1 are respectively in the outward-open and outward-closed transmembrane cavity conformation
Our simulations suggest that ABCA1 transports PL by a mechanism distinct from the alternating-access mechanism and that of other known ABC transporters. To further explore the mechanism of PL export, we constructed a homology model of ATP-bound ABCA1 based on recently published cryo-EM structures of ABCA419,20. We used this approach because the amino acid sequence of ABCA1 is 51% identical and 66% similar to the sequence of ABCA421. Moreover, phylogenetic modeling22 shows that the two proteins are among the most closely related ABCA transporters23. Finally, it is well established that ABCA4 extracts its substrate from the outer plasma membrane monolayer19,20, which is in excellent agreement with our proposed model of ABCA1.
In the cryo-EM structure of ATP-free ABCA1, the transmembrane cavity is outward-open (Fig. 6A, B). In contrast, our homology model of ATP-bound ABCA1 resembles the cryo-EM structure of ABCA4 because it shows no significant substrate-binding transmembrane cavity on either side of the membrane leaflet (Fig. 6C), a conformation we term outward-closed. This homology model suggests that ATP-binding changes the conformation of ABCA1 while providing the energy for translocating POPC through the annulus orifice and into the elongated hydrophobic tunnel.
We hypothesized that the change from the outward-open to outward-closed conformation of ABCA1 might also open the transporter’s annulus orifice. Examination of the two conformations revealed that the orifice’s two helices moved 6 Å further apart, providing space for one or more POPC molecules to pass through (compare Fig. 6H, I with 6J, K). Our observations suggest that ATP-driven changes in the conformation of ABCA1 from the outward-open to the outward-closed form provides the energy for translocating PL from the outward-facing cavity, into the annulus, through the annulus orifice, and into the elongated hydrophobic tunnel.
As noted by Xie et al.20 the outward-open conformation of ABCA4 exhibits a well-ordered S-loop (residues ~337–353) that is located just above the plane formed by the membrane outer leaflet of the outward-open structure. In their nomenclature, S stands for substrate; they show the substrate, N-retinylidene-phosphatidylethanolamine, is coordinated to aromatic residues W339, Y340, Y345, and F348 in a small helical portion of the S-loop. Close inspection reveals that this region is completely disordered in all cryo-EM structures of ABCA1, including ATP-free13, such that Qian et al. did not attempt to model that region.
When we aligned the ATP-free ABCA1 structure with the ATP-free ABCA4 structure, the gateway domain in ABCA1 (residues 564–592) aligns with an analogous domain in ABCA4 (residues 579-607) (Supplementary Fig. 6). The most significant compositional difference between the two aligned gateway domains is the addition of an extra basic residue by replacement of G572 in ABCA1 with R587 in ABCA4. We postulate that this additional basic residue is responsible for creation of the S-loop by formation of a salt bridge between the gateway and E341 in the S-loop (Supplementary Fig. 6).
ABCA1 and ABCA4 are known to transport substrate in opposite directions, toward and away from the extracellular space, respectively. Not only does the location of the S-loop in the outward-facing transmembrane cavity block contact of PL with the analogous gateway in ABCA4 upon diffusion from the outer membrane monolayer into the outward-open transmembrane cavity (Fig. 7) but the location provides an interesting additional confirmation of our model. The S-loop/gateway complex forms a salt bridge with K583—the residue analogous to the key PL-extracting residue in ABCA1, K568 (Supplementary Fig. 6). In future studies, it will be important to further clarify the structural features of ABCA1 and ABCA4 that permit them to transport substrates in opposite directions.
ABCA1 lacks clusters of charged amino acid residues in the middle of its transmembrane domain (Fig. 8A)
Our findings contrast with the alternating-access mechanism of substrate export used by ABC transporters with floppase activity. Those transporters switch between an open, inward-facing transmembrane cavity and a closed, outward-facing transmembrane cavity17. Moreover, most of their structures reveal multiple basic amino acid residues (Fig. 8C) or, in the case of ABCB424, a single histidine residue in the middle of the TMD. In contrast, the cryo-EM structure of ABCA1 shows only hydrophobic amino acids in the middle of the TMD. In the ATP-free conformation, there is only one histidine residue 5 Å or so toward the cytoplasmic side of the membrane and not in the outward-open transmembrane cavity.