Targeted manipulation of the sortilin–progranulin axis rescues progranulin haploinsufficiency

Progranulin (GRN) mutations causing haploinsufficiency are a major cause of frontotemporal lobar degeneration (FTLD-TDP). Recent discoveries demonstrating sortilin (SORT1) is a neuronal receptor for PGRN endocytosis and a determinant of plasma PGRN levels portend the development of enhancers targeting the SORT1–PGRN axis. We demonstrate the preclinical efficacy of several approaches through which impairing PGRN's interaction with SORT1 restores extracellular PGRN levels. Our report is the first to demonstrate the efficacy of enhancing PGRN levels in iPSC neurons derived from frontotemporal dementia (FTD) patients with PGRN deficiency. We validate a small molecule preferentially increases extracellular PGRN by reducing SORT1 levels in various mammalian cell lines and patient-derived iPSC neurons and lymphocytes. We further demonstrate that SORT1 antagonists and a small-molecule binder of PGRN588–593, residues critical for PGRN–SORT1 binding, inhibit SORT1-mediated PGRN endocytosis. Collectively, our data demonstrate that the SORT1–PGRN axis is a viable target for PGRN-based therapy, particularly in FTD-GRN patients.


SUPPLEMENTARY FIGURE 4
The PGRN 588-593 binder BVFP inhibits PGRN-SORT1 interactions. (A) Schematic diagram illustrating the components of the PGRN-SORT1 binding assay. M17 cell lysates overexpressing Flag-tagged human SORT1 protein is added to MSD-assay plated coated with anti-Flag antibody. Immunocaptured SORT1-Flag protein on a MSD-assay plate is used as bait to precipitate PGRN. Detection and quantification of precipitated PGRN is then measured by using an anti-PGRN antibody and a sulfo-tag detection antibody. (B) BVFP binding to the PGRN 588-593 motif may prevent PGRN from interacting with SORT1 therefore reducing the amount of PGRN captured. (C) Pre-incubation of BVFP with the N-terminal 6His-tag rPGRN (N-rP) significantly reduced the amount of PGRN that immunoprecipitated with SORT1. Note that a C-terminal 6His-tag rPGRN (C-rP) group was included as a negative control. *** P < 0.001 vs. vehicle control, analysis performed by One-way ANOVA followed by Tukey post-test.

Epic(R) biochemical assay and compound library screen
Essentially, the Epic ® is a spectrophotometric reader that uses a resonant waveguide grating detection mechanism to detect biochemical interactions (Fig.5A). In brief, the waveguide grating biosensor in each well of the 384-well assay plate intercepts broadband light and measures changes in resonant wavelengths (in picometers, pm) when a ligand binds to the target protein immobilized on the biosensor. To identify chemical binders of the PGRN 588-593 peptide, we immobilized the peptide onto the biosensor surface of Epic ® assay plates using an aminecoupling method, and screened 4,800 compounds. For immobilization, 150 g/ml peptide diluted in a 20 mM sodium acetate buffer, pH 5, was added to the assay plate to incubate for overnight at 4 °C. The plate was then washed in binding buffer (PBS with 5% DMSO) for three times and

Molecular modeling of SORT1 with ligand binding
The X-ray structure for SORT1 (PDB code: 3F6K) was imported into the Protein-Preparation-Wizard GUI of Schrödinger with Maestro 2012 version 9.3.5 (Schrödinger, LLC) for adaption to the OPLS2005 force field. Bond orders were assigned, zero-order bonds to metals were determined, disulfide bonds were created as needed, and all hydrogens were re-generated for every residue. Hydrogen-bond assignment was based on sampling water orientations and taking into account crystallographic waters. Protonation states were predicted for pH 7.2 (range +/-2.0) using PROPKA (1,2). Steric clashes were resolved with convergence of RMSD to 0.3Å using SUthe OPLS2005 force field within Schrödinger-2013. >8 Å from the modeled substrate by using harmonic restraints at 100 kcal/mol, and allowed the residues within the 8 Å cutoff to move freely during serial PRCG energy minimization over 500 iterations with repetition as necessary to converge upon a gradient threshold of <0.05.

Grid of SORT1 with human PGRN 588-593 -ALRQLL or mouse PGRN 584-589 -VPRPLL-A
grid was generated for the docking site based upon the X-ray structure for SORT1 (all within grid for 3F6K; forming 216 sites) using superposition algorithm at the terminal leucine residues from Human or Mouse sequence, which then served as a Grid for subsequent docking, consistent with the protocol used for the NTS peptides.
Docking protocol -We have previously described the methodology used for substrate docking(4); briefly, the binding site was generated via overlapping grids based on the X-ray structure with a default rectangular box centered on the target substrate. Substrate peptides were docked into the binding site of SORT1 using Glide extra precision (SP, XP) (Glide, version 5.6, Schrödinger, LLC); molecular conformations were sampled using methods we have described previously (5). A structure-based pharmacophore score was generated from the optimized, best scoring pose for each substrate peptide based on the descriptors from Glide XP score using established approaches (4)(5)(6). The energetic value assigned to each pharmacophore feature was calculated using Phase (Phase, v3.2, Schrödinger, LLC) as the sum of the Glide XP contributions of the atoms comprising the site (SiteMap). Overall dockings at the active site were quantified and ranked on the basis of these energetic terms (6,7). To account for protein flexibility and lessen the effects of minor steric clashes, excluded volumes spheres corresponding to 80% of the VdW atomic radii were created for all SORT1 atoms within 6 Å of each peptide substrate. A minimum of ten poses per peptide with multiple input conformations generated, chosen for a combination of best-scoring features, was selected for visual and energetic comparison(4, 6, 7).
Additionally, docking was enhanced with hydrophobic mapping, flexible docking, and Epik state penalities with OPLS2005 force field. The top poses were compared with results from the InducedFit method of Schrodinger, which accounts for enzyme flexibility upon binding. The pKas for the enzyme were broken into 224 overlapping clusters including waters from the X-ray structure, which allowed sampling of all possible combinations of pKa of 3F6K, was predicted for the given pH range and based on overall lowest energy from minimization of the hydrogens for SORT1 (7)(8)(9).