Sortilin-Mediated Endocytosis Determines Levels of the Fronto-Temporal Dementia Protein, Progranulin

SUMMARY

The most common inherited form of Fronto-Temporal Lobar Degeneration (FTLD) known stems from Progranulin (GRN) mutation, and exhibits TDP-43 plus ubiquitin aggregates. Despite the causative role of GRN haploinsufficiency in FTLD-TDP, the neurobiology of this secreted glycoprotein is unclear. Here, we examined PGRN binding to the cell surface. PGRN binds to cortical neurons via its C-terminus, and unbiased expression cloning identifies Sortilin (Sort1) as a binding site. Sort1−/− neurons exhibit reduced PGRN binding. In the CNS, Sortilin is expressed by neurons and PGRN is most strongly expressed by activated microglial cells after injury. Sortilin rapidly endocytoses and delivers PGRN to lysosomes. Mice lacking Sortilin have elevations in brain and serum PGRN levels of 2.5- to 5-fold. The 50% PGRN decrease causative in FTLD-TDP cases is mimicked in GRN+/− mice, and is fully normalized by Sort1 ablation. Sortilin-mediated PGRN endocytosis is likely to play a central role in FTLD-TDP pathophysiology.

INTRODUCTION

The FTLDs are characterized clinically by dementia with prominent behavioral alterations and distinct syndromes including progressive aphasia and semantic disorders (Cairns et al., 2007; Mackenzie et al., 2009). A subset of cases exhibit Tau-positive neurofibrillary tangles, but the majority show TDP-43 and ubiquitin positive inclusions in the brain, termed FTLD-TDP. TDP-43 aggregates and TDP-43 mutations are also observed in ALS (Kabashi et al., 2008; Neumann et al., 2006; Sreedharan et al., 2008). Mutations in GRN are the most common inherited cause of FTLD known (Baker et al., 2006; Cruts et al., 2006; Gass et al., 2006). Such families exhibit autosomal dominant inheritance of FTLD-TDP and their GRN mutations result in loss-of-function. Levels of PGRN protein are reduced by 50% in GRN-mutant FTLD-TDP cases (Baker et al., 2006; Cruts et al., 2006; Finch et al., 2009; Gass et al., 2006; Ghidoni et al., 2008; Sleegers et al., 2009).

PGRN is an evolutionarily conserved, secreted glycoprotein with 7 granulin (GRN) repeats. It has been implicated in wound healing, and is associated with malignancy, although the basis for these effects is not fully defined (He et al., 2003; Zhu et al., 2002). Effects of PGRN on neurite outgrowth or cell survival in different assays have been reported (Van Damme et al., 2008). However, in our preliminary studies, PGRN had no detectable effects on the survival or neurite outgrowth from cerebral cortical or hippocampal neurons (Suppl. Fig. S2). While human haploinsufficiency results in FTLD-TDP, mice homozygous for a GRN null mutation exhibit little of no FTLD phenotype (Kayasuga et al., 2007; Yin et al., 2010). As yet, there is no consensus regarding PGRN action in the nervous system or the mechanism whereby 50% reduction might lead to FTLD-TDP. A fraction of PGRN can be proteolytically processed to smaller GRN peptides (Zhu et al., 2002), but the relative role of these fragments in FTLD-TDP is not known.

Elucidation of PGRN action and the control of PGRN levels may have broad relevance for both FTLD and ALS. In order to advance understanding of secreted PGRN biology, we searched for high affinity cell surface binding sites in an unbiased screen. We report that Sortilin is such a site and mediates rapid endocytosis and lysosomal localization of PGRN. Moreover, this mechanism has pronounced effects on brain and serum PGRN levels that are as great as the haploinsufficiency causing FTLD-TDP. Thus, a Sortilin-dependent pathway is likely to play a central role in FTLD-TDP, and possibly in ALS.

RESULTS

PGRN Binds with High Affinity to Sortilin

We hypothesized that the key step in PGRN action is binding to neurons. We created an alkaline phosphatase (AP) tagged PGRN ligand to assess binding (Suppl. Fig. S1A). Fusion of the amino terminus of PGRN to AP yields protein with high affinity for 21 DIV cultured neurons (Fig. 1A). Binding is saturable with an apparent K_D of 15.4±2.5 nM (Fig. 1B), and can be displaced by unlabeled PGRN (Suppl. Fig. 1B).

Figure 1. Cellular Progranulin binding is mediated by a direct high affinity interaction with Sortilin.

(A) Cultured E18 rat hippocampal neurons at 21 days in vitro (21 DIV), were incubated with conditioned medium containing 50 nM AP-PGRN at 4C. Binding was visualized with AP substrate BCIP/NBT.

(B) AP–PGRN binding to cortical neurons measured as a function of ligand concentration. Inset, Scatchard plot. K_D, average ± sem, n = 4 separate experiments.

(C) COS-7 cells transfected with Sortilin or vector control were incubated with conditioned medium containing 50 nM AP-PGRN.

(D) AP–PGRN binding to Sortilin expressing COS-7 cells measured as a function of AP–PGRN concentration.

(E) Binding of PGRN to Sortilin on COS-7 cell surface. Purified Flag-tagged PGRN was incubated with COS-7 cells transfected with myc-tagged Sortilin on ice for 2 hours. After washing, live cells were co-stained with mouse anti-Flag (green) and rabbit anti-myc (red) antibodies on ice for 1 hour. Cells were then washed, fixed and stained with secondary antibodies, and counter stained with DAPI (Blue). Note that PGRN binds only to Sortilin expressing cells. In the high magnification view of a Sortilin-expressing cell in the bottom panels, there is extensive colocalization of the two proteins at the cell surface.

(F) Co-immunoprecipitation of Flag-PGRN with myc-tagged ecto domain of Sortilin (Sort1-ecto). Conditioned media containing Flag-PGRN or Flag-Reg2 were mixed together with conditioned media containing myc-tagged Sort1-ecto protein. Immunoprecipitation (IP) was then carried out using anti-myc or anti-Flag antibodies as indicated. The presence of Sort1-ecto or Flag-tagged protein in the IP was detected in the immunoblot using anti-myc or anti-Flag antibodies, respectively.

(G) Binding of PGRN to immobilized Sortilin ectodomain by surface plasmon resonance analysis (BIAcore). PGRN (10-100 nM) was applied to a microchip containing immobilized Sortilin (0.0952 pmol/mm²). The calculated K_D value is indicated, mean ± sem.

(H) Purity of PGRN ligand for surface plasmon resonance studies. His-PGRN protein was purified by nickel affinity chromatography from transfected cell conditioned medium, and then analyzed by SDS-PAGE with Coomassie Blue staining. Migration of M_r standards, in KDa, at left.

Scale bars, 50 μm for all, except 5 μm in the bottom panels of E.

Using this ligand binding assay, we searched for the molecular identity of this site in an unbiased fashion by expression cloning in COS-7 cells (Fig. 1C). Untransfected COS-7 cells do not bind AP-PGRN. After screening 225,000 clones at low stringency in pools of 100 clones, and 352 transmembrane proteins at high stringency as single clones, we identified only one clone that supported high affinity for AP-PGRN. This cDNA encodes Sortilin (Sort1), a single-pass type I transmembrane protein of the Vps10 family which is localized to the cell surface, secretory and endocytic compartments of eukaryotic cells (Willnow et al., 2008). The affinity of AP-PGRN binding to Sortilin-expressing COS-7 cells is indistinguishable from AP-PGRN binding to neurons (Fig. 1D, 2C). Binding to Sortilin-expressing COS-7 cells is displaced by excess unlabelled PGRN (Fig. 2D).

Figure 2. Specificity of Progranulin interaction with Sortilin1.

(A) PGRN does not bind to Sortilin homologues SorLA or SorCS1, or to neurotensin receptor 2 (NTR2) in transfected COS-7 cells.

(B) The C-terminus of PGRN is the Sortilin binding domain in PGRN. AP-PGRN or AP-PGRN-E (aa 494-593) or AP-PGRNΔE (aa 18-493) was allowed to bind to COS-7 cells transfected with a Sortilin expression vector. While PGRN-E exhibits strong binding to Sortilin, PGRNΔE does not bind to Sortilin.

(C) Scatchard plot of AP-PGRN-E or AP-PGRN binding to Sortilin expressing COS-7 cells. K_D, mean ± sem, n = 4.

(D) Displacement of PGRN binding to Sortilin. COS-7 cells transfected with Sortilin vector were pre-incubated with 100 nM purified His tagged PGRN, 100 nM purified Sort1-Ecto domain, 20 μM GST, 2 μM neurotensin (NTS) peptide, 50 μM neurotensin (NTS) peptide residues 1-6, 2 μM neurotensin (NTS) peptide residues 8-13 or 20 μM purified GST tagged Pro domain of NGF before addition of AP-PGRN at 4 nM. Cells were further incubated for another 2 hours before washing, fixation and visualization of binding using AP substrates.

(E) Binding from experiments as in panel D is quantitated. Data are mean ± sem form 3-4 independent measurements. *, P <0.05 relative to No Addition, one-way ANOVA.

(F) p75NTR does not affect PGRN binding to Sortilin. AP-PGRN at the indicated concentration was allowed to bind COS-7 cells transfected with Sortilin alone, p75NTR alone or co-transfected with Sortilin and p75NTR, as indicated.

(G) Binding analysis of PGRN to all five Vps10p family members by surface plasmon resonance (SPR) analysis demonstrated strong PGRN binding to Sortilin, very weak binding to SorLA and no detectable binding to SorCS1-3. PGRN (200 nM) was included in the perfusate during the interval from 100-550 sec.

(H) Competitive SPR analysis of PGRN and proNGF to Sortilin. Between time points 1 and 2, Sortilin chips were exposed to saturating proNGF (1000 nM) or buffer, as indicated. Between points 2 and 3, the chips were exposed to PGRN (200 nM) or proNGF (1000 nM). Binding of PGRN during the 2-3 interval is not substantially altered by prior incubation with proNGF.

Scale bars, 50 μm.

The binding of PGRN to Sortilin-expressing cells appears to be direct by several criteria. Bound PGRN localizes to Sortilin-expressing cells and, at high magnification, PGRN colocalizes extensively with cell surface Sortilin (Fig. 1E). PGRN and the ecto-domain of Sortilin co-immunoprecipitate from the conditioned medium of transfected cells, whereas an unrelated secreted protein (Reg2) does not associate with the ecto-domain of Sortilin (Fig. 1F). Furthermore, PGRN binding to Sortilin-ectodomain coated surface plasmon resonance chips exhibits an affinity similar to cellular binding (Fig. 1G, H). Thus, Sortilin is identified as a direct high affinity binding site for PGRN.

Specificity of the Sortilin Interaction with PGRN

Sortilin and related proteins have been reported to bind several ligands (Jansen et al., 2007; Nykjaer et al., 2004; Willnow et al., 2008). SorLA, but not Sortilin, is implicated in APP processing and Aß levels in Alzheimer’s Disease (Andersen et al., 2005; Rogaeva et al., 2007). Therefore, we expressed Sortilin-related proteins, SorLA and SorCS1, in COS-7 cells and examined PGRN binding. PGRN binds only to Sortilin (Fig. 2A, Suppl. Fig. S1C). By surface plasmon resonance, SorLA, SorCS1, SorCS2, SorCS3 exhibit much lower affinity for PGRN than does Sortilin (Fig. 2G). Sortilin was originally identified as a low affinity neurotensin (NT) binding protein and termed NT receptor 3, NTR3 (Willnow et al., 2008). Two other NT binding sites are G protein-coupled receptors, but NTR1 and NTR2 do not bind PGRN (Fig. 2A, Suppl. Fig. S1C).

Neurotensin has recently been shown to complex with the Sortilin ß-propeller domain via its extreme carboxyl terminus (Quistgaard et al., 2009), and the last 6 residues of NT displace PGRN binding to Sortilin (Fig. 2D, E). We considered a similar model for PGRN and compared the Sortilin association of the C-terminal 100 residues of PGRN, containing the GRN-E domain plus the extreme C-terminal residues (termed PGRN-E), with that of the N-terminal 80% of the protein (termed PGRNΔE). The carboxyl PGRN-E segment fully accounts for PGRN interaction with Sortilin. In fact, PGRN-E displays slightly higher affinity for Sortilin than does full length PGRN (Fig. 1D, 2B, 2C) while the majority of the protein, contained in PGRNΔE, has no detectable affinity for Sortilin (Fig. 2B).

Pro-neurotrophins, such as pro-NGF are also reported ligands of Sortilin, but they do not depend on the C-terminus of the ligand for binding to Sortilin, as do NT and PGRN (Domeniconi et al., 2007; Jansen et al., 2007; Nykjaer et al., 2004; Teng et al., 2005). While these ligands can reduce AP-PGRN binding to Sortilin, inhibition requires very high levels of pro-NGF, in the μM range (Fig. 2D, E), with no inhibition by nM concentrations (data not shown). In surface plasmon resonance competition assays, proNGF and PGRN show additive binding signals for immobilized Sortilin, consistent with distinct binding sites (Fig. 2H). Pro-NGF is reported to interact with a p75NTR/Sortilin complex with enhanced bipartite affinity (Nykjaer et al., 2004). Therefore, we examined AP-PGRN interaction with cell expressing p75NTR, Sortilin or both (Fig. 2F). PGRN shows no affinity for p75NTR and the Sortilin/p75NTR co-expressing cells show no enhancement of affinity for PGRN. Thus, the binding of PGRN’s carboxyl terminus to Sortilin resembles a high affinity version of NT binding, but may be distinct from that of pro-neurotrophin to Sortilin.

PGRN Binding to Neuronal Sortilin

A key issue is the extent to which Sortilin accounts for neuronal binding sites for PGRN. Several criteria validate Sortilin as a major PGRN binding in brain tissue. First, the development of Sortilin parallels that of AP-PGRN binding sites. For embryonic cortical neuron cultures at 7 DIV, both Sortilin expression and PGRN binding sites are low, but both increase substantially by 14 DIV (Fig. 3A, B). PGRN-E binds with equal to or greater affinity than does full length PGRN to cortical neurons, while PGRNΔE does not bind to cortical neurons at 10 nM (Fig. 3C, Suppl. Fig. S1D). Recombinant immobilized GST-PGRN-E, but not GST, pulls down Sortilin from whole brain extracts of wild type, but not Sort1−/− mice (Fig. 3D). Sort1−/− mice produce low levels of a misfolded Sortilin deletion mutant that lacks a portion of the beta propeller region, and does not bind PGRN-E. Most critically, high affinity AP-PGRN-E binding to cortical neuron cultures from Sort1−/− mice is significantly less than from wild type mice (Fig. 3E, F). This reduction is prominent even after any potential compensatory up-regulation of alternate binding sites in the constitutive Sort1−/− knockout strain. Thus, Sortilin is a major high affinity PGRN binding site in cortical neurons.

Figure 3. Sortilin is an endogenous brain PGRN binding site.

(A) Sortilin expression in cortical neuron culture lysate by immunoblot.

(B) AP-PGRN binding to cortical neuronal cultures grown for indicated times. Scale bar, 50 μm.

(C) AP-PGRN-E exhibits strong binding to cortical neurons (21 DIV), while PGRNΔE binding is minimal at 10 nM. Scale bar, 50 μm.

(D) Brain lysate of wild type (+/+) or Sort1 knockout (−/−) mice were incubated with purified GST or GST-PGRN-E (aa 494-593) protein. After further incubation with glutathione beads, the pull down products were washed and blotted with mouse anti-Sortilin antibodies. The trace immunoreactivity in the lysate of the Sort1−/− sample is derived from a truncated, non-functional protein ((Jansen et al., 2007) and data not shown). M_r markers are shown to the right, 100 kDa in the top panel, and 37 kDa plus 25 kDa in the bottom panel.

(E) AP-PGRN-E (1 nM) binding to cultured mouse neurons (21 DIV) from Sort1 +/− and Sort1 −/− mice. Scale bar, 50 μm.

(F) Quantification of AP-PGRN-E binding to cultured mouse neurons from Sort1 +/− and Sort1 −/− mice. The gray box represents non-specific binding assessed in the presence of 200 nM His-PGRN as in Suppl. Fig. S1B. Mean ± sem, **, P < 0.01, one-way ANOVA.

Sortilin Regulates Extracellular PGRN Level

As a first step to explore possible effects of PGRN/Sortilin interaction, we co-expressed the two proteins in HEK293T cells. PGRN is secreted from transfected cells but, when co-expressed with Sortilin, PGRN levels in conditioned medium are dramatically reduced, to 15% of control levels (Fig. 4A, B). It has been reported that Sortilin can be a substrate for regulated intramembranous proteolysis by gamma-secretase (Hermey et al., 2006; Nyborg et al., 2006). From HEK cells, there is limited secretion of a Sortilin proteolytic fragment into conditioned medium, and this cleavage is not altered by PGRN co-expression (Fig. 4C).

Figure 4. Levels of Extracellular Progranulin are Reduced by Cellular Sortilin.

(A) HEK 293T cells were co-transfected with human PGRN expression construct plus pcDNA vector control, SorLA or various Sortilin expression constructs. The conditioned media (CM) and cell lysates (cell) were collected 5 days after transfection and immunoblotted for PGRN and Sortilin.

(B) Quantification of PGRN level in the conditioned media (CM) or in the cell lysate (Cells) for experiments in A expressed as percentage of pcDNA control. Mean ± sem, **, P < 0.01, one-way ANOVA.

(C) HEK 293T cells were co-transfected with human full length Sortilin expression vector plus pcDNA control vector or PGRN expression vector. Immunoblot for Sortilin protein shows that the majority is cell-associated and not shed, regardless of PGRN co-expression.

The pronounced reduction in extracellular free PGRN caused by Sortilin co-expression might be attributed to impaired secretory trafficking within the cell, sequestration at the cell surface via binding, or endocytosis and clearance from the medium. Sortilin family proteins have been shown to play a role in both trafficking and endocytosis for other ligands (Nielsen et al., 2001; Willnow et al., 2008). To consider these possibilities, we expressed mutants of Sortilin that lack the cytoplasmic tail of the protein with an alternative C-terminal domain (pDisplay Sort1) or that carry a mutation disrupting endocytosis and possibly sorting (Sort1 mut)(Nielsen et al., 2001). While these mutants traverse the secretory pathway and reach the cell surface to support binding of AP-PGRN to transfected cells (Fig. 6C, Suppl. Movie 3, and data not shown), they do not alter PGRN levels in conditioned medium (Fig. 4A, B). A version of Sortilin consisting only of the ectodomain (Sort1 ecto) is secreted efficiently from transfected cells, but does not alter PGRN level in the medium (Fig. 4A, B). Furthermore, inhibition of endocytosis by addition of 0.45 M sucrose to the culture medium for 12 hours increases the level of PGRN in the medium of Sortilin+PGRN expressing cells to that of PGRN only cells (data not shown). Thus, these initial studies suggested that the reduced PGRN levels in conditioned medium from Sortilin co-expressing cells are due to endocytosis of PGRN.

Figure 6. Rapid Endocytosis of PGRN by Sortilin.

(A) COS-7 cells expressing Sortilin were incubated with conditioned media containing mCherry tagged PGRN at 4C and confocal images were acquired after staining with anti-Sortilin antibodies. A magnified view of the boxed region is provided by the inset. Colocalization at the cell surface was seen at 4C.

(B) Sortilin mediates endocytic trafficking of PGRN. GFP-Sortilin expressing COS-7 cells were incubated with mCherry-PGRN at 4C and then shifted to 37C for live imaging by TIRF microscopy (see Supplemental Movie 1). The insets provide magnified view of the boxed regions. After 4 min at 37C, mCherry-PGRN is localized to puncta with GFP-Sortilin (arrow). By 8 min, most mCherry-PGRN diverges into vesicles that lack GFP-Sortilin (arrow).

(C) Endocytosis of PGRN by Sortilin requires the cytoplasmic domain of Sortilin. COS-7 cells expressing Sortilin or pDisplay-Sort1 were incubated with media containing mCherry-PGRN washed and imaged by TIRF microscopy for the indicated times. Near complete internalization of mCherry-PGRN is seen in the Sortilin expressing cells by 18 min at 37C (Supplemental Movie 2), whereas pDisplay-Sort1 expressing cells show persistent mCherry-PGRN signal near the cell surface after 60 min (Supplemental Movie 3).

(D) GFP-sortilin expression cells were incubated without mCherry-PGRN at 4C and the GFP-sortilin signal was imaged using TIRF for the indicated times (Supplemental Movie 4). The GFP-sortilin distribution near the cell surface is stable compared to cells with ligand in B.

(E) The background-corrected mCherry fluorescence intensity relative to time zero is plotted as a function of time from experiments as in B-D. The mCherry-PGRN signal from Sortilin expressing cells is mean ± sem for n = 9 and the apparent half-life is 4 min.

(F) Endocytosed PGRN localizes to lysosomes. Sortilin-expressing COS-7 cells were incubated with mCherry-PGRN (red in merge) on ice for 4 hours before washing and shifting to 37°C. Cells were fixed at indicated times and stained with mouse anti-Lamp1 antibodies (green in merge). Images were obtained with a laser-scanning confocal microscope.

(G) Endogenous brain PGRN is localized to lysosomes. Sections of adult frontal cortex were doubled stained for anti-PGRN (green) and anti-Lamp1 (red). Extensive co-localization demonstrates that many PGRN-reactive puncta are lysosomes in these two examples.

Scale bar, 10 μm.

Localization of PGRN to Activated Microglia and Sortilin to Neurons

The potential roles of endocytic versus secretory pathways on PGRN levels depend on whether PGRN and Sortilin are expressed in the same cells (cis) or different cell types (trans) in brain. We examined frontal cortex for Sortilin and PGRN protein localization. Both proteins exhibit intracellular granular immunoreactivity in cortical neuronal soma, but there is little colocalization (Fig. 5A). Sortilin and other members of the Vps10 family are known to be enriched in recycling endosomes (Willnow et al., 2008), consistent with this pattern. A vacuolar pattern for PGRN has been reported in neurons, and may reflect lysosomal localization (see below). In Sort1−/− brain, a similar pattern of PGRN immunoreactivity is observed, but there is a trend towards reduced vacuolar staining and increased diffuse neuropil staining (Fig. 5B).

Figure 5. Localization of Sortilin and PGRN in Different CNS Cell Types.

(A) Double immunohistochemistry for PGRN (green) and Sortilin (red) in adult mouse frontal cortex. The boxed area in the merge is shown at the right. Scale bar, 7 μm.

(B) Anti-PGRN staining of mouse frontal cortex from WT and Sort1−/− mice. In both cases vacuolar staining is prominent in neurons, and there is a trend to decreased vacuolar localization and increased diffuse neuropil staining in the Sortilin null samples. Scale bar, 7 μm.

(C) Anti-Sortilin staining of human frontal cortex from neurological healthy control and an FTLD-TDP case with PGRN mutation. Scale bar, 30 μm for top panels and 10 μm for bottom panels.

(D) Sortilin immunohistochemistry of the ventral horn in the mouse L5 spinal cord shows significant expression of Sortilin in motor neurons. Scale bar, 50 μm.

(E, F) Immunohistochemistry for PGRN in the ventral horn (VH) of L5 spinal cord transverse sections from mice that had unilateral sciatic nerve resection 7 days previously. Injury-induced PGRN expression is detected on the axotomized side, but not on the uninjured side in E. Significant co-localization of injury-induced PGRN with the microglial marker (Iba1) and is observed on the axotomized side (blue arrows in F), but not on the uninjured side. Less intense vacuolar PGRN immunoreactivity is detected in motoneurons (white arrows in F). Scale bars, 50 μm in E and 18 μm in F.

(G) The conditioned medium and cellular fraction of C13-NJ microglial cell line cultures were immunoblotted for PGRN and for Sortilin. PGRN secretion is prominent, and there is little Sortilin immunoreactivity.

Previous studies have documented an unaltered PGRN distribution in FTLD-TDP brain with GRN mutations (Mackenzie et al., 2006). We examined Sortilin histologically in such cases (Baker et al., 2006) (Fig. 5C). The granular pattern of Sortilin inmmunoreactivity within frontal cortex neurons is similar in healthy human brain to that seen in mouse brain. This distribution is not altered in FTLD-TDP due to GRN mutation. Thus, these steady state localization studies of mouse and human frontal cortex demonstrate that PGRN and Sortilin are present, but do not provide mechanistic insight as to their roles.

To provide an experimental model that might mimic CNS stress, repair and regeneration of relevance to FTLD and ALS, we examined the ventral horn of the lumbar spinal cord of mice after sciatic nerve injury. After axotomy, the protein that aggregates in FTLD-TDP, TDP-43, shifts from a nuclear to a cytoplasmic localization (Moisse et al., 2009; Sato et al., 2009)(Suppl. Fig. S4). Sortilin is strongly expressed by spinal motoneurons (Fig. 5D)(Domeniconi et al., 2007), but not by microglia (Suppl. Fig. 7B). In contrast, PGRN is strongly induced in activated microglial cells that surround motor neurons after peripheral axonal injury but not by astrocytes (Fig. 5E, 5F, Suppl. Fig. S3), consistent with human FTLD-TDP pathology (Baker et al., 2006; Mackenzie et al., 2006). In naïve tissue, PGRN expression is much lower and includes a neuronal component (Fig. 5E, F and (Ryan et al., 2009)). These histological studies, as well as previous expression surveys (Suppl. Fig. S5), are most supportive of the hypothesis that PGRN is secreted by activated microglial cells and then interacts in trans with Sortilin on motoneurons to be endocytosed. Indeed, the C13-NJ microglial cell line secretes PGRN robustly but expresses little Sortilin (Fig. 5G).

Rapid Sortilin-Mediated Endocytosis of PGRN to Lysosomes

Given the separate cells of origin, we focused on endocytosis of PGRN by Sortilin in controlled systems. To visualize endocytosis we expressed Sortilin in COS-7 cells and applied mCherry-PGRN ligand (Fig. 6A). At 4C, binding of fluorescent PGRN ligand is detected at the cell surface and is extensively colocalized with Sortilin. Cells were then shifted to 37C and the region of the cell within 400 nm of the cell attached surface was imaged by total interference fluorescence microscopy (TIRF) (Fig. 6B, C, Suppl. Movie 1, 2). Within 5 minutes, numerous mobile puncta enriched for both mCherry-PGRN and GFP-Sortilin become apparent, consistent with endocytic vesicles (Fig. 6B, inset). Over the ensuing 10 minutes, nearly all PGRN is removed from the cell surface of Sortilin-expressing cells (Fig. 6B, C, E, Suppl. Movie 1, 2). By 18 min, no diffuse plasma membrane PGRN signal is visible, and limited signal remaining near the cell surface is concentrated in mobile puncta consistent with endosomes. The half-life for PGRN bound to Sortilin at the cell surface is 4 min (Fig. 6E). After 30 min, the mCherry-PGRN signal visualized by confocal microscopy further from the cell surface colocalizes extensively with the lysosomal marker, Lamp1 (Fig. 6F).

The clearance of PGRN from the cell surface is dependent on the cytoplasmic domain of Sortilin, because a truncated mutant binds mCherry-PGRN, but there is little change in PGRN TIRF signal over 60 min (Fig. 6C, Suppl. Movie 3), consistent with a half-life at the cell surface at least 5 times longer than for ligand bound to intact Sortilin. A portion of the GFP-Sortilin protein initially colocalizing with PGRN ligand at the cell surface later becomes clustered in mCherry-PGRN-positive puncta (Fig. 6B, Suppl. Movie 1), and there is a net decrease in GFP-Sortilin signal within 400 nm of the plasma membrane. Clearance is not as complete as for mCherry-PGRN (Fig. 6B). As the mCherry-PGRN ligand signal is cleared over 10 min, the GFP-Sortilin receptor signal stabilizes and there is a recovery of the diffuse plasma membrane signal, consistent with recycling of GFP-Sortilin from endosomes to plasma membrane (Fig. 6B, Suppl. Movie 1). The decreased TIRF signal for GFP-Sortilin at 37C after exposure to PGRN requires the presence of bound ligand, since there is little signal decrease in the absence of PGRN (Fig. 6D, Suppl. Movie 4). Thus, there is rapid endocytosis of extracellular PGRN by cell surface Sortilin to COS-7 lysosomes.

We considered whether the vacuolar pattern of PGRN staining observed in neurons of the frontal cortex and spinal motoneurons (Fig. 5) might reflect lysosomal accumulation of the PGRN protein. Double immunohistochemistry for the lysosomal marker Lamp1 and for PGRN, reveals that a substantial fraction of PGRN in frontal cortex is present in Lamp1-positive lysosomes (Fig 6G).

Sortilin Controls PGRN Levels in vivo

To consider the impact of this endocytic mechanism in vivo, we examined PGRN levels in Sort1 −/− mice. PGRN immunoreactivity in brain exhibits two species of 73 and 78 kDa due to differential glycosylation (Fig. 7, Suppl. Fig. S6A), while PGRN in serum is largely of 78 kDa (Fig. 7D). In 7-month-old mice lacking Sortilin, PGRN protein levels are strongly upregulated (Fig. 7B-E). The increase in total brain PGRN level is 2.5-fold (Fig. 7B, C), while the increase of 78 kDa PGRN in brain and in serum is 5-fold (Fig. 7D, E). Because FTLD derives from PGRN haploinsufficiency with a 50% decrease in PGRN protein levels, the absence of Sortilin may fully normalize PGRN levels. Indeed, GRN+/− mice lacking Sortilin expression exhibit levels of PGRN protein that equal or exceed wild type levels (Fig. 7F, G, Suppl. Fig. S6B, C). The protein changes are specific for PGRN, in that levels of another Sortilin ligand, prosaposin (Lefrancois et al., 2003; Zeng et al., 2009), are not altered (Suppl. Fig. S6E). While Sortilin deficiency increases PGRN level, ablation of PGRN does not alter Sortilin level or proteolysis in brain tissue (Suppl. Fig. S6D). The rapid endocytosis of PGRN by Sortilin (Fig. 6) is the likely cause for increased PGRN in Sort1−/− mice. To consider an alternative transcriptional mechanism, we assessed the level of PGRN mRNA in brain by quantitative RT-PCR. No difference in PGRN mRNA levels occurs in Sort1−/− versus wild type mice (Suppl. Fig. S6F), supporting the hypothesis that Sortilin endocytosis determines PGRN level in brain and serum.

Figure 7. PGRN levels are increased in mice lacking Sortilin.

(A) WT and GRN−/− mouse brain lysate was analyzed by anti-PGRN immunoblot.

(B) Tissue lysate collected from the cerebral cortex of 7-month-old mice was subjected to SDS-PAGE and anti-PGRN immunoblot. Two PGRN bands are seen at 78KDa (band A) and at 73KDa (band B); both bands increase in the Sort1−/− samples compared to WT.

(C) Quantification of PGRN level (bands A, B and A+B) in the cortical lysate, normalized to GAPDH immunoreactivity. Sort1−/− n = 4 mice, WT n = 5. Mean ± sem, **, P < 0.01, one-way ANOVA.

(D) Serum samples from 7-month-old Sort1−/− and WT mice were collected, stripped of albumin and IgG and immunoblotted for PGRN and transferrin. Serum levels of PGRN were increased in the Sort1 −/− compared to littermate WT.

(E) Quantification of PGRN level in the serum normalized to transferrin level. For both genotypes, n = 4 mice. Mean ± sem, **, P < 0.01, one-way ANOVA.

(F) Serum samples from 2-month-old mice were obtained from the indicated genotypes and analyzed by anti-PGRN immunoblot, with anti-thrombin III immunoblot as a loading control.

(G) Quantification of PGRN level in the serum normalized to anti-thrombin III level. For all genotypes, n = 4 mice. Mean ± sem; *, P < 0.05; **, P < 0.01; one-way ANOVA.

DISCUSSION

The major finding of this work is that Sortilin is a principal neuronal binding site for the FTLD protein, PGRN. In a stressed nervous system, after production by activated microglial cells, PGRN binds to Sortilin expressed on the neuronal cell surface. Binding occurs via the carboxyl terminus of PGRN, in a manner that may resemble NT binding to Sortilin (Quistgaard et al., 2009). A dramatic consequence of such binding is the rapid endocytosis of PGRN by Sortilin. In vivo, the absence of Sortilin raises PGRN levels by 2.5- to 5-fold in different compartments. Given that FTLD is caused by a 2-fold reduction of PGRN levels (Baker et al., 2006; Cruts et al., 2006; Finch et al., 2009; Gass et al., 2006; Ghidoni et al., 2008; Sleegers et al., 2009), the magnitude of this change has clear pathophysiological consequences, and PGRN deficiency in GRN+/− mice is fully normalized by deletion of Sortilin expression (Fig. 7F and 7G, Suppl. Fig. S6).

There are several mechanisms whereby PGRN/Sortilin interaction might alter neuronal function and contribute to FTLD-TDP. Sortilin may function upstream to titrate PGRN levels and/or downstream of PGRN to mediate its effects on cell function. To distinguish the relative importance of Sortilin upstream versus downstream of PGRN will require the creation of FTLD-TDP relevant models in which altering PGRN function changes the outcome. Unfortunately, such an experimental model is not yet established. The changes in brain and serum PGRN levels in Sort1−/− mice observed here demonstrate that Sortilin can have an upstream effect to regulate levels of PGRN. However, there is ample reason to believe that Sortilin’s primary effect is downstream of PGRN, as a mediator or receptor for functional effects in neurons. Sortilin is reported to possess signalling function in complex with p75NTR for pro-neurotrophin apoptotic signalling (Domeniconi et al., 2007; Jansen et al., 2007; Nykjaer et al., 2004; Teng et al., 2005). While we did not observe modulation of PGRN binding to Sortilin by p75NTR, pro-neurotrophin action may be altered in a more complex in vivo setting. Specifically, since pro-NGF and PGRN can both bind to Sortilin (Fig. 2D, E, H), PGRN might rescue cells otherwise subject to pro-neurotrophin-induced cell death through Sortilin/p75NTR. Displacement of PGRN by pro-NGF is of low potency so it may be indirect or allosteric, and therefore highly dependent on cellular context. Further study of pro-neurotrophin-related effects under a range of conditions may be revealing. However, it remains at least equally likely that PGRN signals via Sortilin independently of pro-neurotrophins.

Separate from signalling at the cell surface, PGRN may have intracellular functions dependent on endocytosis by neuronal Sortilin and delivery to lysosomes. Importantly, immunologically intact PGRN accumulates within the lysosomal compartment of COS-7 cells and neurons. In healthy or FTLD-TDP human brain PGRN is most prominent in microglial cells, but is also present in puncta within the cytoplasm of neurons (Mackenzie et al., 2006), and these appear most consistent with lysosomes. These observations raise the possibility that PGRN functions within autophagosomal/lysosomal pathways. Endocytosis and lysosomal delivery of another Sortilin ligand, prosaposin, serves to enhance lysosomal function rather than simply degrading the ligand (Lefrancois et al., 2003; Zeng et al., 2009).

Under this hypothesis, Sortilin binding is crucial to provide neuronal competence for TDP-43 clearance via autophagosomal/lysosomal mechanisms. Sortilin is a member of the Vps10 family (Willnow et al., 2008), members of which are required for sorting to the lysosomal vacuole in yeast (Cooper and Stevens, 1996). Sortilin is known to participate in the delivery of extracellular prosaposin and intracellular Golgi-derived cathepsins to lysosomes (Canuel et al., 2008; Lefrancois et al., 2003; Ni and Morales, 2006; Zeng et al., 2009). PGRN is co-regulated with lysosomal genes (see Suppl. Data in Reference (Sardiello et al., 2009)). We and others (Ahmed et al., 2010) have observed accelerated brain lipfuscinosis in mice lacking PGRN. Of relevance to FTLD, the clearance of cytoplasmic TDP-43 depends at least in part on autophagy (Caccamo et al., 2009; Ju et al., 2009; Urushitani et al., 2010; Wang et al., 2010). Ubiquitin co-accumulation with TDP-43 aggregates may reflect disrupted balance between proteosomal pathways, macroautophagy and chaperone-mediated autophagy pathway.

In this regard, it may be relevant to consider those mutations in Valosin-Containing Protein (VCP), which are known to cause FTLD plus myopathy and Paget’s disease (Custer et al., 2010; Ju et al., 2009; Watts et al., 2004). The accumulation of TDP-43 and ubiquitin aggregates in brain of these rare cases is similar to that in more common PGRN-mutant cases. Since VCP has a prominent role in autophagy, the PGRN pathway and the VCP pathway may overlap as intracellular constituents are sorted to lysosomes. Both PGRN and VCP might function to clear TDP-43 aggregates via autophagy.

As noted above, PGRN can be converted proteolytically into smaller GRN peptides (Zhu et al., 2002). In the extracellular space, this conversion can be mediated by elastase, and inhibited by secreted leukocyte protease inhibitor (SLPI). Lysosomal localization of PGRN may also influence conversion from PGRN to GRN, and therefore the balance between PGRN and GRN function. Such conversion may have downstream effects on PGRN biology and the pathophysiology of FTLD. In order to separate these possibilities further in vivo neuronal studies of GRN versus PGRN activity will be required.

It is clear from the present data that the interaction of PGRN with Sortilin-mediated endocytosis significantly reduces steady-state PGRN levels. The magnitude of this reduction is at least as great as that of those PGRN mutations causing FTLD-TDP. Thus, Sortilin binding provides a potential therapeutic site to alter PGRN-dependent pathways and alleviate TDP-43 pathology.

It is clear that microglial cells are a major source of PGRN production. For example, activated microglia in the vicinity of axotomized motoneurons strongly induce PGRN after sciatic nerve injury. From this observation, it follows that characterizing the mechanisms of microglial PGRN induction or the selectivity of PGRN for subsets of microglia may reveal additional means to modulate the course of FTLD-TDP.

Because neither GRN +/− nor GRN−/− mice exhibit an FTLD-like phenotype (Kayasuga et al., 2007; Yin et al., 2010) (data not shown), this human inherited disease cannot be modelled simply in rodent. Although peripheral axotomy shifts TDP-43 from the nucleus to the cytoplasm of motoneurons (Moisse et al., 2009; Sato et al., 2009) and might sensitize to FTLD/ALS-like pathology, neither GRN−/− nor Sort1−/− mice have detectable motoneuron loss or ventral root axonal degeneration after this perturbation (Suppl. Fig. S4). Future functional studies of Sortilin in PGRN biology will require development of robust rodent models for PGRN-dependent neurodegeneration. Nevertheless, our work implicates Sortilin-mediated PGRN endocytosis as a key pathway for further study in FTLD pathophysiology.

EXPERIMENTAL PROCEDURES

Protein chemistry, expression cloning and binding assays

AP-PGRN fusion proteins were expressed in HEK293T cells and purified as described for other proteins (Fournier et al., 2001; Hu et al., 2005; Lauren et al., 2009). GST-PGRN-E was produced in E. coli. AP-PGRN binding assays and cDNA library screening were performed with transfected COS-7 or mouse cortical neurons, as described for other ligands (Fournier et al., 2001; Hu et al., 2005; Lauren et al., 2009). Immunoblot and affinity chromatography methods have been described (Fournier et al., 2001; Hu et al., 2005; Jansen et al., 2007; Lauren et al., 2009; Nykjaer et al., 2004).

Sort1 −/− and GRN −/− mice

The Sort1−/− mouse line has been described (Jansen et al., 2007) and studies here utilized mice backcrossed for more than 9 generations to C57BL6. The Sort1−/− mice are generated by deletion of a critical exon encoding a portion of the ß propeller domain. Although trace levels of immunoreactivity of smaller size are detectable in the Sort1−/− strain, none of the mutant protein folds appropriately or reaches the cell surface (data not shown). The GRN−/− mouse has been described (Kayasuga et al., 2007).