Date Published: March 01, 2016
Publisher: International Union of Crystallography
Author(s): Bernard T. Kelly, Stephen C. Graham, David J. Owen.
A selenomethionine marker strategy allowed the identification of a region of disconnected electron density at low resolution and despite poor selenomethionine incorporation, thereby building a structural framework for understanding how the clathrin adaptor AP2 regulates clathrin binding in mammalian cells.
Eukaryotic cells contain a plethora of specialized lipid membrane-enclosed organelles. Transmembrane proteins (and often their luminal cargo) are transported between these organelles in a controlled fashion to ensure the correct functioning of the cell. For example, activated cell surface receptors are often downregulated by internalization from the plasma membrane and delivery to lysosomes, where they are degraded. Transmembrane-protein ‘cargo’ is moved between organelles by incorporation into small membrane-bound transport containers termed ‘vesicles’ that bud off one organelle and are transported to and fuse with a second (destination) organelle. This process has to be tightly regulated to ensure that proteins are delivered in a timely and accurate manner. Thus, eukaryotes have evolved a modular trafficking system in which transmembrane proteins and organelles are marked with signals that interact with the cytosolic proteins that control inter-organelle traffic (Traub, 2009 ▸).
To investigate this observation, we attempted to crystallize a form of the AP2 complex comprising the whole of the β2 subunit (and thus both clathrin-interacting sites) along with the μ2 and σ2 subunits and the trunk subdomain of α. Unfortunately, we were unable to crystallize this complex. Next, we constructed a version of the AP2 core complex extended to include a 68-residue part of the unstructured β2 hinge (including the clathrin-box motif; Fig. 2 ▸f), which we termed βhingeHis6.AP2. The extended β2 subunit in our βhingeHis6.AP2 construct ended at Met650, whereas the β2 subunit that we had previously used to determine the core structure ended at residue Lys591 and the last ordered residue discernible in the core structure was Val582. We were concerned that the unstructured segment of the β2 hinge might be prone to proteolysis. We therefore moved the hexahistidine tag to the C-terminus of β2 in an attempt to ensure that only AP2 complexes containing full-length β2 (i.e. trunk plus hinge fragment) are bound during the Ni–NTA purification step. This extended β2 subunit construct was successfully crystallized in the same conditions that were previously used to grow crystals of the AP2 core in the locked (inactive) conformational state (Collins et al., 2002 ▸).
To begin to narrow down the buried region, we constructed mutants of βhingeHis6.AP2 truncated after Gln619 and Leu636 (Fig. 2 ▸f) and lacking the C-terminal hexahistidine tag to avoid the possibility of the tag interfering with binding in the bowl. These mutants were expressed and crystallized as described above, yielding crystals that were isomorphous to those of the nontruncated complex. Refinement of the AP2 core complex structure against these data revealed that the Leu636 truncation mutant retained the unmodelled difference density in the bowl (Fig. 2 ▸d), whereas the Gln619 truncation mutant did not (Fig. 2 ▸e). This suggested that the buried sequence was N-terminal to Leu636 and might lie between Gln619 and Leu636. On this basis, we prepared preliminary models that placed the region between residues 619 and 636 into the difference electron density visible in the bowl. Secondary-structure prediction using the JPred server (Cole et al., 2008 ▸) suggested the presence of a short region of helix spanning Asp626–Leu631. At low contour levels, a 2mFo − DFc map hinted at a possible helical region in the buried electron density; as a result, our first model was built on this basis. The occupancy of the buried fragment when refined with fixed B factors in REFMAC5 was ∼0.8. We then prepared a series of models sequentially shifted by one residue at a time. The quality of the electron density was, however, insufficient to differentiate between these models. Similarly to all AP2 structures determined to date, the β2 subunit is less well ordered than the σ subunit or the N-terminal regions of the α subunit abutting σ, probably because β2 acts as a ‘latch’ to hold the complex shut and is thus poised to swing away from σ and μ2 in order to reveal the cargo-binding sites (Jackson et al., 2010 ▸). It is therefore not surprising that the buried portion of the β2 hinge is likewise comparatively poorly ordered, rendering definitive identification of the residues problematic. It remained possible that the buried region lay partly or wholly N-terminal to Gln619 and that the removal of residues 619–636 destabilized the hinge–bowl interaction perhaps owing to a loss of weaker, secondary interactions. Thus, we sought a way to identify the buried residues definitively.
Others have successfully used methionine point mutants incorporating selenomethionine (SeMet) to identify regions of structure in low-resolution maps (Pomeranz Krummel et al., 2009 ▸; Oubridge et al., 2009 ▸) or for chain tracing (Evans, 2003 ▸). We decided to pursue a similar strategy to identify the residues buried in the bowl of AP2. Apart from two methionine residues at the extreme C-terminus, the β2 hinge fragment in our construct lacks endogenous methionines (Fig. 2 ▸f). We therefore constructed a series of point mutants in which single residues were substituted with methionine (Fig. 3 ▸a). Initially, we chose hydrophobic residues (valine, isoleucine and leucine) together with glutamine or glutamate residues (which contain an aliphatic side chain similar in length to methionine) to mutate. We subsequently mutated a single aspartate in order to bridge a gap of three residues between neighbouring mutation sites. By crystallizing each mutant and pinpointing selenium sites, we hoped to determine the position of the single introduced methionine in each case and thereby trace the residues buried in the bowl.
In almost all cases the anomalous signal was quite weak, with useful signal generally not extending beyond ∼6 Å resolution (as judged by the resolution at which the ratio of anomalous differences to their estimated standard deviations drops below ∼1.3; Schneider & Sheldrick, 2002 ▸; Fig. 4 ▸). In the absence of any other phase information, this would make substructure solution difficult, and indeed attempts to solve the substructure with SHELXD (Schneider & Sheldrick, 2002 ▸) failed with all but one of the mutant data sets (D626M). Given the low incorporation of selenomethionine (∼45%) and the large number of selenomethionine sites (38 in the core), this is not surprising. However, our goal was not to solve the structure using experimental phases, but rather to identify selenium marker sites. Therefore, we could make use of this weak anomalous data to find sites by using phases calculated from our existing AP2 locked-core model. Our strategy was to identify anomalous scatterers (i.e. selenium sites) by iterative substructure completion using anomalous log-likelihood gradient maps with Phaser-EP, where starting phases were provided by an AP2 model refined against the new data and including a ‘best-guess’ model of the buried hinge fragment. In this approach, SAD log-likelihood gradient maps are searched for sites where the addition of an anomalous scatterer would improve the fit of the anomalous scattering model to the experimental data and, after new sites have been identified, the process is iterated until the map is ‘flat’ (Read & McCoy, 2011 ▸). The likelihood formulation has the advantage of increased sensitivity compared with simple difference Fouriers (de La Fortelle & Bricogne, 1997 ▸). We used custom scripts to automate the substructure completion with Phaser-EP. The Z-score cutoff for addition of new sites was set at the default level of 6.
The model was refined by iterative rounds of rebuilding in Coot (Emsley et al., 2010 ▸) and TLS and restrained refinement in REFMAC5. MolProbity (Chen et al., 2010 ▸), accessed via the PHENIX interface (Echols et al., 2012 ▸), and the validation tools within Coot were consulted throughout the refinement process. The final model had R and Rfree residuals of 0.203 and 0.259, respectively, and good stereochemistry (r.m.s.d.s of 0.013 Å for bond lengths and 1.54° for bond angles; Table 1 ▸). In common with the original AP2 core structure (Collins et al., 2002 ▸), the helical solenoid of the β2 trunk is followed by a stretch of extended peptide and a trio of short helices that pack against each other and against the β2 trunk; after Val582 the hinge becomes disordered. Our new structure (Fig. 6 ▸a) reveals that after 35 disordered residues, the β2 hinge then loops back in towards the bowl of AP2, forming a short stretch of β-sheet with a loop between two helices of the α-subunit solenoid; there follows a turn and an α-helix that includes the first few residues of the clathrin-box motif before the electron density is lost.
Our studies have shown that useful information can be obtained from partial selenomethionine-incorporation strategies when full incorporation is prohibited owing to problems with protein production. Although it was necessary to screen multiple crystals or sites on large crystals in order to obtain the best diffraction and anomalous signal, this is now a practical approach because of improvements in synchrotron beamlines and X-ray diffraction detectors that have dramatically increased the speed of data collection. Our crystallographic studies provided a structural framework to design biochemical experiments that elucidated how AP2 keeps its clathrin-binding motif hidden from clathrin until it is correctly localized at the plasma membrane and bound to cargo (Kelly et al., 2014 ▸).