A Large-Scale Conformational Change Couples Membrane Recruitment to Cargo Binding in the AP2 Clathrin Adaptor Complex

Summary The AP2 adaptor complex (α, β2, σ2, and μ2 subunits) crosslinks the endocytic clathrin scaffold to PtdIns4,5P2-containing membranes and transmembrane protein cargo. In the “locked” cytosolic form, AP2's binding sites for the two endocytic motifs, YxxΦ on the C-terminal domain of μ2 (C-μ2) and [ED]xxxL[LI] on σ2, are blocked by parts of β2. Using protein crystallography, we show that AP2 undergoes a large conformational change in which C-μ2 relocates to an orthogonal face of the complex, simultaneously unblocking both cargo-binding sites; the previously unstructured μ2 linker becomes helical and binds back onto the complex. This structural rearrangement results in AP2's four PtdIns4,5P2- and two endocytic motif-binding sites becoming coplanar, facilitating their simultaneous interaction with PtdIns4,5P2/cargo-containing membranes. Using a range of biophysical techniques, we show that the endocytic cargo binding of AP2 is driven by its interaction with PtdIns4,5P2-containing membranes.


INTRODUCTION
The regulated movement of proteins and lipids between the many cellular membranes in coated vesicular carriers is important for signaling, homeostasis, defining the interactions of cells with their surroundings, and in controlling the glycosylation and proteolytic processing of lumenal and transmembrane proteins. Clathrincoated vesicles (CCVs) mediate many post-Golgi trafficking routes including internalization from the plasma membrane via clathrin-mediated endocytosis (CME) (Traub, 2009). CCVs have a three-layered structure with an inner membrane layer linked by clathrin adaptors to an outer polymeric clathrin scaffold (Cheng et al., 2007;Fotin et al., 2004). Clathrin adaptors contain a folded membrane-proximal domain, which binds to phosphatidyl inositol polyphosphate (PIP) headgroups and/or Arf GTPases in their membrane-attached, GTP-bound forms, and at least one natively unstructured region, which harbors a clathrin-binding motif (Owen et al., 2004). Transmembrane proteins are generally selected as cargo for incorporation into a CCV through the direct interaction of either widely used, short, linear sequence motifs or covalently attached ubiquitin chains with the membrane-proximal portion of a clathrin adaptor (reviewed in Bonifacino and Traub, 2003;Traub, 2009).
The family of heterotetrameric vesicle coat adaptors comprises APs 1-4 and the b,g,d,z subcomplex of COPI (Schledzewski et al., 1999). AP2 consists of a, b2, m2, and s2 subunits (Figure S1 available online) and is the most abundant endocytic clathrin adaptor (Keen et al., 1981;Pearse and Robinson, 1984). The 70 kDa trunk domains of a and b2, together with the 50 kDa m2 and the 17 kDa s2 subunits, form the 200 kDa membraneproximal core (Collins et al., 2002;Heuser and Keen, 1988). The 30 kDa bilobal C-terminal appendages of a and b2 each possess two sites for binding different motifs found on many other endocytic CCV coat proteins with various accessory/regulatory functions in CCV formation (reviewed in Owen et al., 2004;Schmid and McMahon, 2007). The appendages are connected to the core via long flexible linkers (Zaremba and Keen, 1983) that contain clathrin-binding motifs. The AP2 core is the site of binding to the two widely used endocytic cargo motifs, YxxF (where F is a hydrophobic residue: I, L, M, F, or V) and [ED]xxxL [LI] or ''acidic dileucine'' motifs, as well as to the plasma membrane PIP, phosphatidyl inositol-4,5-bisphosphate (PtdIns4,5P 2 ). The AP2 complex therefore acts as a central hub for CCV formation, binding to clathrin, cargo molecules, accessory proteins, and membranes, and these interactions must be under strict spatiotemporal control.
Two structures of the AP2 core have been solved, one with a PIP headgroup analog (inositol hexakisphosphate [IP 6 ]) but lacking bound YxxF or [ED]xxxL [LI] motifs (Collins et al., 2002), and the other in complex with an [ED]xxxL [LI] motif peptide (Kelly et al., 2008). The a and b2 trunk domains are solenoids of stacked a helices, and the N-terminal m2 domain (N-m2; resi-dues1-135) and s2 domain are globular ''longin domain'' folds. Together, these four domains are arranged into a ''bowl,'' in which the elongated C-terminal m2 domain (C-m2) sits ( Figure S1). Two positively charged sites that can bind PtdIns4,5P 2 , one on a and one on m2, have been identified (Collins et al., 2002;Gaidarov et al., 1996;Rohde et al., 2002) with the basic site on a playing a key role in the initial docking of AP2 onto PtdIns4,5P 2 and cargo-containing membranes (Gaidarov et al., 1996;Hö ning et al., 2005).
The binding site for [ED]xxxL[LI] motifs is located mainly on s2 (Chaudhuri et al., 2009;Doray et al., 2007;Kelly et al., 2008), whereas YxxF motifs bind through a b-augmentation to C-m2 (Ohno et al., 1995;Owen and Evans, 1998). In the motif-free conformation first shown for AP2 and subsequently for AP1 (Heldwein et al., 2004), both cargo-motif binding sites are blocked by portions of the b2 subunit: b2Tyr6 and b2Phe7 block the [ED]xxxL[LI]-binding site and b2Glu364-b2Val406 (especially b2Val365 and b2Tyr405) block the YxxF site (numbering refers to AP2) ( Figure S1), and thus we term this the ''locked'' conformation. In the recent [ED]xxxL[LI] motif-liganded structure, which we term the ''unlatched'' conformation, the N terminus of b2 is displaced, allowing [ED]xxxL[LI] motif binding to occur. Although in the unlatched form the subunits of the bowl have moved a little relative to each other as compared with the locked form, the YxxF motif-binding site remains blocked (Kelly et al., 2008). In both the locked and unlatched structures, the main PtdIns4,5P 2 -binding site on the a subunit and the [ED]xxxL [LI] motif-binding site are adjacent to each other and located on what was consequently proposed to be part of the AP2 membrane-interacting surface, but the YxxF motif-binding site is located on an orthogonal face. It has also been shown that the spacer between the end of a protein's transmembrane helix and a YxxF motif need only be seven residues (around 25 Å for a fully extended peptide) (Rohrer et al., 1996). However, in the locked and unlatched structures, the YxxF-binding site is 60-70 Å distant from the putative membrane-interacting surface. Taken together, these data suggested that AP2 must undergo a large-scale conformational change in order to allow binding to YxxF-containing cargo embedded in a PtdIns4,5P 2 -containing membrane.
Here we present the structure of a form of AP2 in which both YxxF and [ED]xxxL[LI] motif-binding sites are occupied and form a coplanar arrangement with four positively charged patches that are sites for binding the PtdIns4,5P 2 -rich plasma membrane. Achieving this new conformation and relieving the intramolecular inhibition of the cargo motif-binding sites are facilitated by large changes occurring to the structure of the whole AP2 core.

RESULTS
In order to compete effectively with the intramolecular blocking of the YxxF-binding site and so shift the equilibrium in solution to a YxxF-bound conformation, AP2 core was preincubated with a very large (70-fold) molar excess of the TGN38 YxxF peptide DYQRLN (35 mM AP2 with 2.5 mM peptide). Crystals of the AP2 complex with bound peptide grew under a number of conditions, but only those grown using mixtures of ammonium sulfate and lithium sulfate showed useful diffraction. These rhombohedral crystals diffracted anisotropically to 3.1 Å resolution in the best direction but little beyond 5 Å in the worst. Experimental phasing was carried out using the Ta 6 Br 12 2+ cluster compound and cryo-trapped Xe derivatives, followed by density modification. The bowl of the AP2 core and C-m2 could be seen in the resulting electron density ( Figures 1A and 1B). It was immediately obvious that a large conformational change had occurred, as C-m2 was no longer located in the center of the bowl but on an orthogonal face ( Figures 1E and 1F, compare Figures 1C and 1D). Model building began by placing rigidbody components of the unlatched AP2 structure into the electron density. Electron density was visible for the linker connecting the two domains of m2, showing that most of it adopted a helical conformation (Figure 2A), and for the YxxF peptide ( Figure 2B). Surprisingly, electron density was also visible in the [ED]xxxL[LI] motif-binding site on s2, although no appropriate exogenous ligand had been included in the crystallization experiments. Inspection of the electron density ( Figure 2C) suggested that this ''phantom'' [ED]xxxL[LI] motif was part of the myc-tag sequence (MEQKLI) inserted into a surface loop of m2 (residues 218-252) at residue 236, given that at low contour levels almost continuous density could be seen extending from the dileucine-binding site to residue 252. This loop was disordered in the locked and unlatched forms. In this ''open'' form, the myc-tag-containing loop reaches over to a neighboring molecule and forms a vital crystal-packing contact. When the myc-tag was removed from the loop, the resulting protein yielded no diffracting crystals under any crystallization conditions including those used here.

Conformational Change in the AP2 Bowl
The AP2 bowl can be considered as an extended helical solenoid running continuously from the N terminus of a through the interface between the a and b2 subunits' C termini to a dozen residues short of the N terminus of b2, forming a puckered ring like the seam of a tennis ball. When going from locked to the open form, the AP2 bowl collapses inwards, expelling C-m2 ( Figure 3, Figure S2, and Movie S1 and Movie S2). The collapse brings both lobes of the puckered ring closer together, while the ring dislocates between the a and b2 N termini. The ''collapsed'' conformation of the bowl is likely to be its lower energy state as a version of the core lacking C-m2 altogether (m2-truncated core) also adopts the collapsed conformation (see Extended Experimental Procedures).
The collapse of the AP2 bowl is facilitated by rotations about four ''hinge points,'' two each in a and in b2 ( Figure S2). Thus each large subunit can be considered as being composed of essentially three rigid groups. The total buried subunit interfaces for the locked and open forms are 10,040 Å 2 and 9700 Å 2 , respectively. Given that buried surface area is highly correlated with energy and therefore stability (Krissinel and Henrick, 2007), these data suggest that the locked form is more stable and therefore predominates in solution. The major contributions to the buried subunit interfaces come from the interactions between a and s2, a and b2, and b2 and N-m2, which remain largely unaltered during the conformational change (see Table S1 and Figure S3), whereas the major changes in subunit packing are those made by C-m2 (see below) and the N terminus of b2 being displaced from the [ED]xxxL[LI] motif-binding site on s2 (Kelly et al., 2008).

Membrane Binding and the Repositioning of C-m2
The most spectacular subunit rearrangement, and that of the greatest biological significance, is the movement of C-m2 relative to the bowl. This can be described geometrically as a screw rotation of C-m2 by $129 about its long axis, with a translation of 39 Å (about half its length), relative to N-m2 ( Figure 4A). The trajectory followed by C-m2 during the actual conformational switching must be greater than this in order to avoid colliding with b2 (see Movie S3). The subunit contacts made by C-m2 change completely between the locked and open conformers ( Figure 4B). All contacts between C-m2 and a and between C-m2 and s2 are lost, and although the surface area buried between C-m2 and b2 doubles (see below and Figure S3 and Table S1), the total subunit interface area buried by C-m2 in the open form drops by around 1200 Å 2 as compared with the locked form. However, this loss of buried surface area is partly compensated for by the 800 Å 2 surface area buried by the m2 linker binding back onto the rest of the core (see below, Figure 4 and Table S1).
In the locked and unlatched forms, the m2 linker (residues 130-158), which contains the AAK1 (a-appendage binding kinase) catalyzed phosphorylation site at m2Thr156 , is disordered, thus making it ideally suited to recognition by the protein kinase. This phosphorylation is important for AP2 function in vivo because mutating the phosphorylation site inhibits transferrin uptake (Motley et al., 2006;Semerdjieva et al., 2008). In vitro, m2Thr156 phosphorylation enhances the binding of AP2 to endocytic motifs (Hö ning et al., 2005;Olusanya et al., 2001;Ricotta et al., 2002) presumably by driving the equilibrium toward the cargo-binding-competent open form. In the open form the linker forms a four-turn helix, which packs in a trough lined by residues from b2, N-m2, and C-m2 ( Figure 4 and Figure S3). In the locked form this trough was present but unoccupied. It is possible that the m2Thr156 phosphorylation event further stabilizes this helical conformation (C) An overall view of the whole AP2 core in the locked conformation. The bowl is formed by the helical solenoids of the a (blue) and b2 (green) subunits, together with s2 (cyan) and the N-terminal domain of m2 (purple). The C-terminal domain of m2 (C-m2) (magenta) is in the center of the bowl. This coloring scheme is used in all subsequent figures. (D) Experimental electron density for the locked form (Collins et al., 2002), in the same view as (C), showing the C-m2 in the center of the bowl. (E) An overall view of the whole AP2 core in the open conformation. C-m2 is no longer in the bowl, but on the outside of b2, and carries the YxxF motif peptide (gold). The myc loop within C-m2 that reaches into the acidic dileucine motif-binding site on a neighboring molecule in the crystal is labeled, and the site itself is indicated (LI (Myc) peptide). In subsequent pictures this myc loop is omitted for clarity, but the EQKLI sequence is shown in its position on s2. (F) Experimental electron density for the new open form, in the same view as (C), showing no density for C-m2 in the bowl but density in its new position on the outside of the b2 subunit. of the linker by interacting with positively charged residues in the vicinity of the trough, especially b2Arg138 and b2Lys139 ( Figure S3). Unfortunately all attempts to crystallize the phosphorylated form of AP2 have so far failed, and attempts to mutate candidate residues resulted in poorly expressed, aggregationprone complexes. Further, an attempt to mimic the phosphorylation event by mutating m2Thr156 to glutamate was also unsuccessful: m2Thr156Glu core shows 4-fold weaker binding to YxxF/PtdIns4,5P 2 -containing liposomes than wild-type core and 20-fold less binding than phosphorylated AP2 core (S.H., unpublished data).
The result of the relocation of C-m2 is that all three previously biochemically confirmed ligand-binding sites (those for PtdIns4,5P 2 , YxxF motifs, and [ED]xxxL[LI] motifs) become coplanar on a surface of AP2 and are thus suitably arranged for contacting the various signals in the context of the plasma membrane. There are three other regions of highly positive electrostatic potential on this surface in addition to the a subunit PtdIns4,5P 2 -binding site ( Figure 5). The first is formed by basic residues from the N terminus of b2 (Lys5, Lys12, Lys26, Lys27, Lys29, Lys36). A version of the AP2 core in which these residues are mutated to glutamate (b2 PIP-Core) cannot be recruited to PtdIns4,5P 2 -containing membranes, similar to the version of the AP2 core harboring mutations in the PtdIns4,5P 2 -binding site on a (a PIP-Core) ( Figure 5 and Hö ning et al., 2005). The second and third basic regions are on the surface of C-m2. The second basic region (Lys330, Lys334, Lys350, Lys352, Lys354, Lys356, Lys365, Lys367, Lys368, Lys373) was identified as a putative PtdIns4,5P 2 -binding site (Collins et al., 2002;Rohde et al., 2002). Mutation of five lysine residues to glutamate in this region (5K>E) slightly weakens the binding of AP2 cores to PtdIns4,5P 2 -containing membranes (dissociation constant [K D ] 11 mM instead of 7 mM). In the third region (Lys167, Arg169, Arg170, Lys421), the substitution of three basic residues (KRR>E) has little effect on AP2 binding to PtdIns4,5P 2 -containing membranes. However, combining the mutations (m2 PIP-Core) results in an AP2 core with a 4-fold reduction in binding to PtdIns4,5P 2 -containing liposomes ( Figure 5). All mutant forms of the AP2 core were correctly folded as judged by identical levels of expression, incorporation into complexes, and CD spectra (data not shown). These data suggest that initial membrane recruitment of AP2 occurs mainly through the a and b2 PtdIns4,5P 2 -binding sites with the basic surface on C-m2 playing an auxiliary role. However, we propose that the C-m2 basic regions are key to driving the opening of AP2 by the electrostatic attraction of C-m2 to the PtdIns4,5P 2 membrane, as m2 PIP-Core mutant shows strongly inhibited (20-fold) binding to PtdIns4,5P 2 /YxxF liposomes. The binding of the m2 PIP-Core mutant to PtdIns4,5P 2 /[ED]xxxL [LI] liposomes is only reduced by 3-fold, suggesting that acidic dileucine motif binding does The top panels are ''omit'' maps, mFo-DFc difference maps calculated by omitting part of the structure, randomly displacing all the atoms a little and then refining, using the experimental phases as restraints. Omitted residues are colored red (linker) or yellow (YxxF peptide and dileucine peptide mimic). The lower panels show the solvent-flattened experimental map. (A) The m2 linker folds into a helix lying in a groove between N-m2 and b2: the side chain of Thr156, which can be phosphorylated, is shown. (B) The YxxF motif peptide is bound to the C-m2 domain, in the equivalent position to that found on the isolated C-m2 domain (Owen and Evans, 1998). (C) Electron density in the acidic dileucine peptide-binding site on s2 is linked to C-m2 across a crystal contact and has been interpreted as part of the myc-tag EQKLI. The peptide QIKRLL from the unlatched structure (gray) is shown in its position relative to s2.
not strictly need C-m2 to bind to the membrane, only its displacement from the bowl.

Endocytic Motif Binding
Once attached to and stabilized on the membrane through binding multiple PtdIns4,5P 2 molecules, the open form of the complex can be further stabilized by the binding of membraneembedded YxxF and [ED]xxxL[LI] cargo motifs in their respective sites, to which there is now unrestricted access for cargoes embedded in the membrane ( Figure 6). Our motif-liganded structure suggests that both motif-binding sites could be occupied simultaneously. We would therefore predict that AP2 would exhibit tighter binding to a membrane harboring both endocytic motifs than to one with either motif alone, due to the avidity effect of simultaneous binding. This is indeed the case ( Figure 5 and Figure S4). However, whether two different cargoes can bind simultaneously to a single AP2 in vivo will depend on steric clashes between their extracellular domains, the flexibility of their juxtamembrane regions, and the ratios of clathrin:AP2: cargo in a given CCV.
The open form structure suggests that a single cargo protein containing both motifs could, in principle, bind with both motifs engaged by the same AP2 complex, but only if the motifs were separated by at least 65-70 Å (corresponding to a spacing of around 25 residues), but this cannot formally be demonstrated in our assay system. Inspection of a database of type I and II membrane proteins (M. Robinson, personal communication) showed that in only a very few cases do both motifs occur separated by at least 20 residues within their unstructured regions, and even in these cases (CIMPR [Chen et al., 1993], furin [Voorhees et al., 1995], vGlut-1 [Kim and Ryan, 2009], and LRP1 [Li et al., 2001], which have complex trafficking itineraries) it appears that one motif is used mainly for internalization and the other for intracellular trafficking.
A question that naturally arises is, if this open structure is energetically stable, is the locked, non-cargo-binding form biologically relevant, i.e., is it the conformation of AP2 in solution? This seems likely, as in the absence of ligands, the locked conformation buries a larger total surface area and is therefore more stable than the open form; furthermore, the locked conformation has also been seen independently in AP1 (Heldwein et al., 2004). Two further lines of evidence support the hypothesis that AP2 adopts the locked conformation in solution. First, in a surface plasmon resonance (SPR) assay, preincubation of AP2 cores with an excess of YxxF motif-containing peptide has no effect on the association of AP2 with PtdIns4,5P 2 and YxxF motif-containing liposomes, whereas preincubation of isolated C-m2 with an excess of the same peptide is sufficient to abolish binding to the same liposomes ( Figure S5). Second, equilibrium fluorescence anisotropy experiments in which a fluorescein-labeled YxxF motif-containing peptide was titrated with C-m2 or with AP2 core reveal that isolated C-m2 binds the peptide with a K D of $1.9 mM, whereas the AP2 core exhibits no detectable binding ( Figure S5).
The most likely candidate for driving and subsequently stabilizing the conformational change to a cargo-binding-competent open form is interaction with multiple PtdIns4,5P 2 molecules arranged in a roughly planar fashion in the membrane. This is supported by three lines of evidence. First, if no PtdIns4,5P 2 is included in liposomes that contain both YxxF and [ED]xxxL [LI] motifs, there is no detectable binding of AP2 ( Figure 6). Second, mutation of any one of the basic PtdIns4,5P 2 -binding patches on a or b2, or the combined pair on m2, strongly inhibits binding to membranes containing PtdIns4,5P 2 and YxxF sorting motifs ( Figure 5 and Figure S4), i.e., high-affinity binding to YxxFcontaining membranes requires multiple simultaneous PtdIns4,5P 2 -binding events on a, b2, and m2 that can only occur when AP2 is in its open conformation. Finally, in the fluorescence anisotropy assay measuring the equilibrium association of AP2 with fluorescein-labeled YxxF peptide, the binding of AP2 was strongly promoted by the presence of the polyanionic heparin ($50-mer), which, with multiple negative charges disposed along a single large molecule, can mimic the arrangement of charges in a membrane. The binding for AP2 to free YxxF peptide increases from undetectable levels to a K D of $3.4 mM in the presence of heparin, which is a similar strength of binding to that of isolated C-m2 for the same peptide, K D $1.9 mM ( Figure S5).
To explore the nature of AP2 activation further we used polarized fluorescence stopped-flow spectrophotometry to follow the pre-steady-state kinetics. Isolated C-m2 and AP2 preincubated with heparin both display rapid binding to YxxF peptide ( Figure S5), with fast on-rates ($10 6 M À1 s À1 ) and off-rates of $2 s À1 . The kinetically derived dissociation constants of C-m2 and heparin-incubated AP2 (0.5 mM and 3.5 mM, respectively) closely match equilibrium measurements, confirming that interaction occurs in a single step. In contrast, YxxF binding by AP2 in the absence of heparin occurs extremely slowly (Figure S5). Heparin-activated AP2 binds with a relaxation time to the open structure, the puckered ring formed from a and b2 narrows and splits between the N termini, and C-m2 emerges from its bowl and rotates roughly about its long axis. s2 and N-m2 remain fixed to a and b2, respectively.
(a measure of the time in which binding occurs) of 0.056 s, whereas in the absence of heparin this reaction slows $1000fold to >60 s ( Figure 6). The dramatic difference in kinetics can only be explained by a slow rate-determining reaction that precedes the bimolecular binding step. As the stoichiometry of the AP2:YxxF interaction is known to be 1:1, the preceding slow step must represent a pre-equilibrium isomerization to a binding-competent isomer. Calculation of kinetic rate constants (data not shown) gives a derived K D for AP2:YxxF of $4 mM. The difference between the affinities of AP2 in the presence or absence of heparin suggests that, at equilibrium, >99.9% of non-membrane-bound AP2 is in a locked or inactive conformation. These data show that AP2 is thus unable to associate promiscuously in the cytoplasm with proteins containing trafficking motifs.

DISCUSSION
The striking conformational change between the locked and the open, ligand-bound active forms of AP2 described here completely remodels the domain arrangement of the heterotetramer. Based on the known AP2 core structures, we propose the following scenario for the activation of AP2 for cargo binding on the plasma membrane. The model assumes that there are two main conformers of AP2, one locked and one open, that are in equilibrium with each other in solution, but that the equilibrium lies heavily in favor of the closed form in the absence of membrane interaction. By characterizing the pre-steady-state interaction of AP2 with a YxxF peptide, we have provided direct evidence that AP2 isomerizes between inactive (''locked'') and active (''open'') isomer forms and shown that >99.9% of nonmembrane-bound AP2 in the cell will be in the ''locked'' conformation. The ''activated'' form of AP2 binds YxxF peptide with a rapid association rate similar to that of isolated C-m2. This is what we would expect from our structure because the YxxFbinding site on C-m2 is now unobstructed by any part of the complex. Taken together with the coplanar arrangement of all the ligand-binding sites, from which we would predict effects subsequently confirmed by our mutagenesis data ( Figure 5), it seems reasonable to equate our open structure with the ''activated'' form of AP2 on the plasma membrane.
The first step in AP2 activation is its recruitment onto the plasma membrane, primarily by binding through the basic patches on a or b2 to PtdIns4,5P 2 , which in vitro has an apparent K D of 7-8 mM (Hö ning et   The initial dislocation of C-m2 from its site on the bowl causes the bowl to relax toward the lower-energy conformation present in the open and m2-truncated forms of the AP2 core (Extended Experimental Procedures). The most important functional effect of this change in the conformation of the bowl is to cause the b2 subunit to move with respect to the two peptide ligand-binding sites such that it is no longer able to block either. The b2 subunit can therefore be considered as the latch that in solution blocks both peptide-binding sites thus rendering them unusable. Once the motif-binding sites are unblocked, they can then bind to any YxxF or [ED]xxxL[LI] motifs in their vicinity, which, because AP2 is held against the membrane, will be those on transmembrane protein cargo. The energy liberated on cargo binding results in the further stabilization of AP2. The AP2 complex is now tightly attached to the membrane via multiple cargo and phospholipid headgroup interactions (apparent K D around 90 nM; Figure 7; Movie S3).
This model is in agreement with the stabilities of the various AP2 structures as inferred from their buried subunit interfaces (Krissinel and Henrick, 2007). In solution the locked form is more stable (10040 Å 2 of buried subunit interface) than the open form (9700 Å 2 ). However, when cargo binding is taken into account, the buried interface area of the open form rises to 10590 Å 2 (not including bound lipids), and so the open form becomes the more stable in the presence of a PtdIns4,5P 2and cargo-containing membrane. This two-state model does not exclude the existence of intermediates, such as the [ED]xxxL[LI]-liganded ''unlatched'' structure (Kelly et al., 2008). However, such a conformer is likely to be short-lived, as C-m2 will be strongly attracted to the PtdIns4,5P 2 of the membrane and should rapidly complete the final stages of the full conformational change to the open form, which can then bind to any available YxxF-containing cargo.
Recent live-cell imaging studies (Loerke et al., 2009;Saffarian et al., 2009) have revealed the presence of three types of AP2/ clathrin-positive structures at the cell's limiting membrane: those that abort in around 5 s (early abortive), those that abort within 15 s (late abortive), and those that are endocytosed at around 100 s. The model presented here is in line with these findings. The early abortive structures correspond to the situation where AP2 transiently docks to the plasma membrane but fails to undergo the activating conformational change, perhaps because the local concentration of PtdIns4,5P 2 is insufficient (Figure 7, lefthand image). The endocytically productive class of structure is that in which AP2 opens and binds to cargo (Figure 7, righthand image). The late-aborted structures therefore correspond to the situation where sufficient PtdIns4,5P 2 is present to drive the conformational change (Figure 7, center image) but insufficient cargo is available to further stabilize the binding of AP2 to the membrane. As we would predict from this assignment of states to conformations of our structural model, Schmid and colleagues have shown that overexpressing YxxF-containing cargo converts the late-abortive class of clathrin-coated structures to endocytically productive ones. This would be caused by the high concentration of YxxF motifs ''shifting'' the equilibrium between open and closed forms even further toward the open form.
Recent work has suggested that Arf6 plays a role in recruiting AP2 to the plasma membrane (Paleotti et al., 2005), despite the observation that PtdIns4,5P 2 is necessary and sufficient to (B) Equivalent ribbon representations, with mutated patches of Lys and Arg residues on a, b2, and C-m2 that affect membrane binding shown in black. (C) Equilibrium binding constants for the binding of wild-type and PtdIns4,5P 2 -binding site mutants of AP2 cores to liposomes displaying the YxxF motif of TGN38 or the dileucine sorting signal of CD4, or both signals, displayed in PtdIns4,5P 2containing liposomes as determined by SPR. Values were calculated from rate constants obtained using five different concentrations of AP2 ranging from 50 nM to 1 mM. Sample sensorgrams, corrected for nonspecific binding to PC/PE liposomes and obtained using concentrations of AP2 ranging from 100 nM to 20 mM are shown in Figure S4. efficiently recruit AP2 to the membrane. There is, however, little doubt that AP1, AP3, AP4, and COPI are all recruited to their respective membranes primarily through interactions between their large subunits and GTP-bound, membrane-associated Arf1 (Austin et al., 2000;Boehm et al., 2001;Spang et al., 1998). The high degree of structural homology between the four APs and the b,g,d,z subcomplex of COPI (Schledzewski et al., 1999) suggests that the same gross conformational change that facilitates strong membrane attachment and cargo binding in AP2 will occur in all family members. In the case of AP2 the conformational change is driven by PtdIns4,5P 2 (although Arf6 may play a role), but in the case of the other AP family members the change must be driven by Arf1GTP. The most obvious way in which Arf1GTP could shift the equilibrium from the closed to the open state is by binding to and thus stabilizing only an open, cargo-binding-competent conformation very similar to that presented here for AP2. This model predicts that Arf1GTP and cargo binding would be synergistic, and this has indeed been shown to be the case for Arf1 (Baust et al., 2006;Lee et al., 2008).
In summary, we have determined a fully ligand-bound form of AP2, which by comparison with our previous structures shows that the complex has undergone a complicated series of largescale subunit-repositioning events. The functional need for this massive subunit rearrangement is to allow cargo binding to be coupled to PtdIns4,5P 2 -containing membrane attachment so as to prevent inappropriate recognition of YxxF and [ED]xxxL[LI] sequences on cytoplasmic proteins by AP2 when it is free in the cytosol. The unblocking of both motif-binding sites is elegantly coordinated through the use of different parts of the same b2 subunit to block simultaneously the two separate endocytic motif-binding sites.

Structure Determination
Recombinant AP2 cores were made as in Collins et al. (2002). Crystals of the open form were grown from a mixture of AP2 cores (7 mg/ml) with the DYQRLN peptide derived from TGN38 (2 mg/ml) by hanging drop vapor diffusion against a reservoir containing 0.7 M Lithium sulfate, 0.7 M ammonium sulfate, and 200 mM sodium citrate (pH 7.4), 5 mM DTT. Crystals were cryoprotected in mother liquor augmented with 20% glycerol and 3 mg/ml TGN38 peptide, and all data were collected on at 100K at beamline ID29 at ESRF. Crystals were of space group R3 with unit cell dimensions a = 255 Å , c = 157 Å . Diffraction from all crystals was severely anisotropic, extending at best to around 3.1 Å resolution in the a-b plane, but not much beyond 5 Å resolution along c. The structure was solved by multiple isomorphous replacement with anomalous scattering experimental phasing using cryo-trapped Xe (10 atmospheres of pressure for 1 min) and crystals soaked in the Ta 6 Br 12 2+ cluster compound (Table S2). The structure was built using a combination of real-space molecular placement and reciprocal space molecular replacement guided by the experimental electron density. The structure was refined using experimental phase restraints. For a full description of the crystallographic methods and structure validation, see the Extended Experimental Procedures.  (C) AP2 (30 mM orange) or AP2 preincubated with heparin (15 mM green) were rapidly mixed with fluorescein-labeled YxxF peptide in a stopped-flow spectrometer and the change in anisotropy upon binding was measured. The anisotropy change was fitted to a single exponential function to obtain relaxation times.

Surface Plasmon Resonance Biosensor Experiments
Recombinant AP2 cores, mutants thereof, and C-m2 were probed for binding to liposomes captured on the L1 surface of a SPR biosensor (BIAcore 3000 and T100). Recording of the interaction and data evaluation were done as described previously (Hö ning et al., 2005;Kelly et al., 2008). In the competition experiments, AP2 cores and C-m2 were incubated with the indicated concentrations of TGN38 or CD4 sorting signal peptides for 15 min at room temperature prior to the biosensor experiment, during which binding to PtdIns4,5P 2 -YxxF-containing liposomes was recorded.

Equilibrium Fluorescence Anisotropy Measurements
The increase in fluorescence polarization anisotropy upon binding of a large molecule such as AP2 or C-m2 to a fluorescent peptide was employed as a measure of binding. Further details are described in the Extended Experimental Procedures. Briefly, a peptide encoding the TGN38 YxxF motif (sequence ASDYQRL) and modified at its N terminus with fluorescein (Sigma-Genosys) was used in all equilibrium binding titration experiments at a concentration of 20 nM. Fluorescence anisotropy was measured using a PheraStar Plus plate reader (BMG Labtech) with increasing pseudo-first order concentrations of AP2, C-m2, or m2-truncated core. Where used, polymeric heparin ($50 subunits) (Rovi Laboratories) was added to 500 mM concentration. Binding curves ( Figure S7) were fitted to a single-site binding model to estimate K D . The m2-truncated core, lacking the C-m2 subdomain and therefore unable to bind the YxxF motif, was used to determine the level of nonspecific background binding (manifested as a linear increase in anisotropy). Without the addition of polymeric heparin, AP2 showed no specific binding compared to m2-truncated core ( Figure S7B).

Stopped-Flow Polarized Fluorescence Spectrophotometry
Pre-steady-state interaction of a fluorescein-conjugated YxxF-containing peptide with C-m2 and AP2 was performed using a dual-channel fluorescence TgK single-mix SF-61SX2 stopped-flow spectrometer. A collimated excitation beam at 494 nm passed through a calcite prism polarizer was used to excite 1 mM YxxF-containing peptide. Parallel and perpendicular polarized fluorescence was measured on independent photomultipliers fitted with 515 nM glass filters from which the fluorescence anisotropy was calculated. Relaxation times were determined for a range of mM C-m2 and AP2 concentrations at pseudo-first order excess and used to determine kinetic rate constants (further details are given in the Extended Experimental Procedures). For heparin experiments, AP2 was preincubated for 30 min with 200 mM heparin.

ACCESSION NUMBERS
Coordinates have been deposited in the Protein Data Bank with PDB ID 2xa7.   (center) is triggered by the electrostatic attraction of C-m2 to the membrane, which can then bind membrane-embedded cargo motifs (right). An animation is presented in Movie S3.