Abstract
SORLA is a neuronal sorting receptor implicated both in sporadic and familial forms of AD. SORLA reduces the amyloidogenic burden by two mechanisms, either by rerouting internalized APP molecules from endosomes to the trans-Golgi network (TGN) to prevent proteolytic processing or by directing newly produced Aβ to lysosomes for catabolism. Studies in cell lines suggested that the interaction of SORLA with cytosolic adaptors retromer and GGA is required for receptor sorting to and from the TGN. However, the relevance of anterograde or retrograde trafficking for SORLA activity in vivo remained largely unexplored. Here, we generated mouse models expressing SORLA variants lacking binding sites for GGA or retromer to query this concept in the brain. Disruption of retromer binding resulted in a retrograde-sorting defect with accumulation of SORLA in endosomes and depletion from the TGN, and in an overall enhanced APP processing. In contrast, disruption of the GGA interaction did not impact APP processing but caused increased brain Aβ levels, a mechanism attributed to a defect in anterograde lysosomal targeting of Aβ. Our findings substantiated the significance of adaptor-mediated sorting for SORLA activities in vivo, and they uncovered that anterograde and retrograde sorting paths may serve discrete receptor functions in amyloidogenic processes.
SIGNIFICANCE STATEMENT SORLA is a sorting receptor that directs target proteins to distinct intracellular compartments in neurons. SORLA has been identified as a genetic risk factor for sporadic, but recently also for familial forms of AD. To confirm the relevance of SORLA sorting for AD processes in the brain, we generated mouse lines, which express trafficking mutants instead of the wild-type form of this receptor. Studying neuronal activities in these mutant mice, we dissected distinct trafficking routes for SORLA guided by two cytosolic adaptors termed GGA and retromer. We show that these sorting pathways serve discrete functions in control of amyloidogenic processes and may represent unique therapeutic targets to interfere with specific aspects of neurodegenerative processes in the diseased brain.
Introduction
Sorting protein-related receptor with A-type repeats (SORLA) is a member of the vacuolar protein sorting 10 protein (VPS10P) gene family, a group of neuronal sorting receptors that traffic target proteins between Golgi, plasma membrane, and endosomes. Earlier work identified central roles for VPS10P domain receptors in control of neuronal viability and function, and in the occurrence of neurodegenerative diseases (for review, see Willnow et al., 2008).
Concerning SORLA, association studies link variants of SORL1 (the gene encoding SORLA) with increased risk of sporadic AD (Rogaeva et al., 2007; Naj et al., 2011). Additionally, recent findings suggest that SORL1 mutations may also underlie autosomal dominant familial forms of this disease (Pottier et al., 2012). According to current models, SORLA impacts AD-related processes by two distinct mechanisms. It acts as a trafficking receptor for APP, which redirects APP from endosomes to the trans-Golgi network (TGN), reducing the extent of proteolytic breakdown into amyloidogenic and nonamyloidogenic products (Offe et al., 2006; Schmidt et al., 2007). SORLA also acts as a sorting factor for Aβ, moving newly produced peptides to lysosomes for catabolism, further reducing the amyloidogenic burden (Caglayan et al., 2014).
Given the importance of SORLA-mediated protein sorting in AD, major attention has been focused on mechanisms governing receptor trafficking at the TGN. Two adaptors received particular attention, the Golgi-localizing, γ-adaptin ear homology domain ARF-interaction proteins (GGAs) and the retromer complex, both of which bind to the cytoplasmic tail of SORLA and are also genetically linked to AD (Small and Gandy, 2006; Wahle et al., 2006; Muhammad et al., 2008). Retromer is a pentameric adaptor complex involved in retrograde sorting of proteins from early endosomes to the TGN (Seaman, 2004; Small, 2008; Fjorback et al., 2012). GGAs mediate anterograde transport of cargo from the TGN to endosomes (Jacobsen et al., 2002; Herskowitz et al., 2012). In established cell lines, disruption of retromer or GGA binding to SORLA impairs trafficking of the receptor and alters APP processing fates (Schmidt et al., 2007; Herskowitz et al., 2012). However, direct proof that these sorting pathways are required for SORLA activity in vivo is lacking.
Here we generated mouse models expressing SORLA mutants defective in binding to retromer or GGAs, and we show that disrupting these trafficking pathways differentially impacts SORLA activities in vivo. Lack of retromer binding impairs the APP sorting function of SORLA, resulting in an accumulation of the receptor in endosomes and in increased APP processing. In contrast, disrupting GGA binding decreases SORLA-dependent lysosomal targeting of Aβ, causing accumulation of the peptide in the brain. Our data document that SORLA-dependent trafficking of APP and Aβ requires distinct steps in anterograde and retrograde sorting, and that these pathways depend on the receptor interaction with GGAs and retromer, respectively.
Materials and Methods
Generation of mouse models.
The generation of mice carrying a cDNA encoding wild-type human SORLA in the Rosa26 locus (Rosa26TgSORLAwt) has been described (Burgert et al., 2013). The same strategy was applied to generate mice carrying constructs for SORLA variants lacking GGA (Rosa26TgSORLAgga) or retromer binding sites (Rosa26TgSORLftaf; Fig. 1A). The constructs were introduced into the murine Rosa26 gene locus using homologous recombination. Mice carrying the transgenes through their germline were crossed with the Cre deleter strain (Taconic) to remove a triple transcription stop site and to enable expression from the endogenous Rosa26 promoter.
Analysis of brain tissue.
Characterization of cortical brain tissues by fluorescence microscopy or Western blotting was performed using standard protocols. APP processing products were determined by multiplex assays (Meso Scale Discovery). Analyses were performed in female mice at 8–10 weeks of age. For differential extraction of Aβ, brain cortices were dissected and homogenized in ice-cold homogenization buffer (0.25 mm sucrose, 2 mm MgCl2, and 20 mm Tris-HCl, pH 7.5). Following incubation for 20 min on ice, nuclei and cellular debris were removed by a centrifugation step (1000 × g, 10 min). The lysate was centrifuged (100,000 × g, 1 h), with the lysate becoming the soluble fraction. The pellet was resuspended in lysis buffer (1% NP-40, 1% Triton X-100, 50 mm Tris-HCl, pH 7.4, 300 mm NaCl) and centrifuged (100,000 × g, 1 h), with the new lysate labeled as the membrane bound fraction. This fraction was used for Western blot analysis. The pellet was subsequently resuspended in 70% formic acid, vortexed for 2 h, and centrifuged at 100,000 × g for 1 h. The resulting supernatant was neutralized in 1 m Tris and labeled the formic acid soluble fraction. All buffers were supplemented with protease inhibitor (05 892 791 001; Roche) and phosphatase inhibitors (78440; Thermo Scientific).
Hippocampal cell cultures.
Primary neurons were generated from P1 mice and analyzed at 5 DIV (Rohe et al., 2008). Pearson's coefficient was determined using Fiji/ImageJ software. At least 20 neurons were imaged per condition. Experiments were replicated at least three times. Studies comparing SORLAGGA with SORLAWT and SORLAFTAF with SORLAWT strains were performed separately, therefore, statistically significant differences between mutant and wild-type lines in each group were evaluated using Student's t test. There was no statistically significant difference in the Pearson's coefficients of the respective SORLAWT condition in both group sets.
Analysis of SH-SY5Y cells.
For pulse chase experiments, SH-SY5Y cells stably overexpressing SORLAWT, SORLAGGA, or SORLAFTAF were incubated in methionine-free medium for 60 min before adding 1.5 mm of the methionine analog l-azidohomoalanine (AHA; Invitrogen). After a 30 min pulse, the cells were chased for various time points and lysed. Using Click-It chemistry, an alkyne-biotin tag was added to AHA in the labeled proteins. Lysates were precipitated with protein A-Sepharose precoated with anti-SORLA IgG, and precipitates were subjected to SDS-PAGE and immunodetection with Streptavidin-HRP. Following densitometric scanning of replicate blots (n = 8 per condition), values were normalized to time point 0 of each experiment (set to 100%). A best-fit, one-phase exponential decay curve (y = Span*e−K*x + Plateau) was determined using GraphPad Prism software, with t(½) = 0.6932/K. For the DAPT experiments, 5 μm DAPT [in the absence or presence of a lysosomal inhibitor cocktail (100 μm leupeptin, 10 μm pepstatin A, and 50 μm chloroquine)] was added to SH-SY5Y cells stably overexpressing APP and SORLAWT, or APP and SORLAGGA (n = 9 replicates). The decay of intracellular Aβ was determined as published previously (Caglayan et al., 2014).
Results
To query the relevance of adaptors for SORLA trafficking in vivo, we generated two mouse models that express receptor variants defective in binding to GGA (SORLAGGA) or retromer (SORLAFTAF). We have applied this strategy before to produce mice expressing wild-type human SORLA (Burgert et al., 2013). In brief, we used PCR-based mutagenesis of the human SORLA cDNA to replace binding site residues for GGA (2207DDVPMV2212) or retromer (2169FTAF2172) in the cytoplasmic receptor domain by alanine residues (Fig. 1A). These modifications eliminate the ability of SORLA to interact with these adaptors in cultured cells (Schmidt et al., 2007; Fjorback et al., 2012). The cDNA constructs were introduced into the Rosa26 locus of mice and activated by Cre-mediated recombination using the Cre deleter strain (Fig. 1A). Mice carrying the activated SORLA transgenes were crossed with animals having a targeted disruption of the endogenous SORLA locus (Sorl1−/−; Andersen et al., 2005) to eliminate expression of the endogenous murine receptor. The corresponding mouse lines (Rosa26TgSORLAgga, Sorl1−/−) and (Rosa26TgSORLAftaf, Sorl1−/−) are referred to as SORLAGGA and SORLAFTAF, respectively. They were compared with(Rosa26TgSORLAwt, Sorl1−/−) animals (SORLAWT) expressing the human wild-type receptor in Rosa26.
Mice expressing human SORLA variants. A, Constructs encoding wild-type SORLA (SORLAWT) or receptor variants lacking the GGA (SORLAGGA) or retromer binding (SORLAFTAF) motifs for targeting the murine Rosa26 locus are shown. Amino acid sequences of the cytoplasmic domains of the receptor variants with residues modified by mutagenesis in the GGA (green) and retromer (red) binding mutants are indicated. The scheme for introducing the targeting vectors between exons 1 and 2 (black squares) of Rosa26 is given below. Removal of the loxP-flanked neomycin resistance cassette (neoR) enables transcription of the SORLA cDNAs from the Rosa26 promoter (activated Rosa26 locus). B, Immunodetection of SORLA (green) in cortices of mice of the indicated genotypes. Sections were costained for the neuronal marker NeuN (blue; left) or astroglial marker GFAP (red) and DAPI (white; right). No immunoreactivity for SORLA is detectable in Sorl1−/− tissue (used as negative control). Scale bar, 10 μm. C, D Western blot analysis (C) and quantification by densitometric scanning of replicate blots (D) documents increased levels for SORLAGGA compared with SORLAWT and SORLAFTAF in brain extracts (n = 3 animals/genotype; *p < 0.05, **p < 0.01, one-way ANOVA, Tukey post hoc analysis). Detection of Na/K ATPase (Na/K) served as loading control in C.
Using immunohistology, we showed that all SORLA variants localized predominantly in the perinuclear region of neurons in the brain, similar to the pattern seen for the murine receptor in Sorl1+/+ mice. No coexpression of the endogenous receptor or of transgenic SORLA variants with GFAP, a marker for astrocytes, was observed (Fig. 1B). This restricted expression of SORLA transgenes from Rosa26 has been seen before and argued for cellular mechanisms that enable expression of this receptor in neurons but not in glia in vivo (Burgert et al., 2013; Caglayan et al., 2014). Surprisingly, SORLAGGA mice showed a stronger SORLA immunoreactivity compared with SORLAWT and SORLAFTAF (Fig. 1B). This notion was confirmed by Western blotting documenting 56% increased levels of SORLAGGA compared with the other receptor variants (Fig. 1C,D). Expression of the human SORLA variants did not alter brain levels (Fig. 2A,B) or cellular localization (Fig. 2C) of retromer subunits VPS26 and VPS35, of GGA1, GGA2, or GGA3, or of other SORLA adaptors, including PACS1, AP2 (μ-subunit), or AP1 (γ-adaptin; Fig. 2A,B). Also, the SORLA binding partners APP and cation-independent mannose-6 phosphate receptor (CI-MPR) showed comparable brain expression patterns in all genotypes (Fig. 2).
Characterization of proteins related to intracellular protein sorting. A, Representative Western blots for proteins related to GGA and retromer activities in the cortices of mice of the various genotypes are shown. Detection of tubulin served as loading control. B, Quantification of the Western blots in A revealed no significant differences in protein expression levels comparing the genotypes (n = 3 animals/genotype). C, Immunodetection of neuronal marker NeuN (blue) and of proteins related to GGA and retromer activities (green) on cortical brain sections of mice of the indicated SORLA genotypes are shown. No discernable differences in neuronal expression patterns in the various genotypes are seen for APP (top row), GGA1 (second row), VPS26 (third row), or CI-MPR (bottom row) are seen. Scale bar, 15 μm.
Next, we asked whether disruption of adaptor binding alters SORLA sorting in neurons as suggested by experiments in non-neuronal cell lines (Nielsen et al., 2007; Schmidt et al., 2007; Herskowitz et al., 2012). Consistent with our observation in vivo (Fig. 1B), all three SORLA variants localized to the perinuclear region in primary hippocampal neurons with expression of SORLAGGA being increased compared with SORLAWT and SORLAFTAF. This increase was seen by immunocytochemistry (Fig. 3A) and by Western blotting (Fig. 3B,C). To test for neuronal protein sorting, we costained primary neurons from the various mouse lines for SORLA and the organelle markers Vti1b (TGN; Fig. 4A), Rab5 (early endosomes; Fig. 4B), or Lamp1 (lysosomes; Fig. 4C). Colocalization of SORLA with these markers was determined by the Pearson's coefficient. Consistent with the established pattern of endogenous SORLA, SORLAWT localized mainly to the TGN, and to a lesser extent to early endosomes and lysosomes. In contrast, SORLAGGA showed increased localization to the TGN and to early endosomes, but a loss from lysosomes. Immunosignals for SORLAFTAF were increased in endosomes, but decreased in the TGN. Localization to lysosomes in SORLAFTAF was not changed compared with SORLAWT. These data confirmed the relevance of GGA and retromer interactions for SORLA sorting in neurons. Specifically, they suggested that loss of anterograde sorting in SORLAGGA retains retrograde trafficking from early endosomes back to the TGN, bypassing lysosomal receptor targeting (Fig. 4D). In contrast, loss of retrograde sorting in SORLAFTAF promotes the anterograde movement of receptors to endosomal compartments, depleting it from the TGN (Fig. 4D).
SORLA levels in primary neurons expressing wild-type and mutant receptor variants. A, Representative images for primary hippocampal neurons (DIV 5) of the indicated genotypes stained for SORLA (green) and DAPI (blue; top), or for neuronal marker MAP2 (white) and DAPI (blue; bottom). Analysis of neuronal cultures from Sorl1−/− mice served as negative control. Scale bar, 10 μm. B, C, Levels of SORLA variants were tested by Western blotting of total cell lysates of primary neurons (B) and quantification by densitometric scanning of replicate blots (C). Levels of SORLAGGA were increased compared with SORLAWT and SORLAFTAF (n = 4 per genotype; *p < 0.05, one-way ANOVA, Tukey post hoc analysis).
Aberrant trafficking of SORLAGGA and SORLAFTAF in hippocampal neurons. A–C, Primary hippocampal neurons of the indicated genotypes stained for SORLA (green), DAPI (blue), and either VTi1b (red; A), Rab5 (red; B), or Lamp1 (red; C). Pearson's coefficients are given in the merged parts. High-magnification insets indicate colocalization of SORLA with the respective marker. At least 20 neurons per genotype were imaged [*p < 0.05;**p < 0.01; ***p < 0.001; Student's t test (as SORLAWT and SORLAGGA were performed separately from SORLAWT and SORLAFTAF experiments]. D, Model of how SORLA localization changes in mutant receptor variants. SORLAWT is found predominantly in the TGN, but also in endosomes and lysosomes (top). Anterograde sorting from TGN to the endosomal/lysosomal compartments is mediated by GGAs (green arrow). Loss of this interaction results in SORLAGGA accumulation in the TGN and depletion from lysosomes (middle). Retrograde sorting of SORLAWT is mediated by retromer (red arrow). Loss of this interaction causes accumulation of SORLAFTAF in early endosomes but depletion from the TGN (bottom).
Having confirmed the necessity of retromer and GGA interactions for receptor sorting, we tested whether loss of these interactions impacts the activities of SORLA in amyloidogenic processes. To do so, we crossed SORLAWT, SORLAGGA, and SORLAFTAF lines with the 5xFAD model of AD (Oakley et al., 2006). As quantified by ELISA, expression of SORLAGGA resulted in an increase in Aβ40 and Aβ42, but not in soluble APPα and APPβ (sAPPα and sAPPβ), compared with SORLAWT (Fig. 5A,B). The increase in Aβ40 and Aβ42 levels was specific for the soluble fraction of the peptides (Fig. 5B) and not seen for the membrane bound (Fig. 5C) or the formic acid soluble (Fig. 5D) fractions that represent different forms of Aβ aggregates. In SORLAFTAF, levels of all APP processing products, including sAPPα and sAPPβ (Fig. 5E) and Aβ40 and Aβ42 (Fig. 5F), were elevated. Similar to what had been seen for SORLAGGA, levels of Aβ40 and Aβ42 were significantly elevated in the soluble (Fig. 5F) but not in the membrane bound (Fig. 5G) or the formic acid soluble (Fig. 5H) fractions. The observation that the SORLA mutants primarily impact soluble Aβ species is in line with a role for this receptor in controlling proximal steps in Aβ production and catabolism. The above data suggested that loss of retromer interaction blocked the ability of SORLA to protect APP from proteolytic breakdown while loss of GGA interaction abrogated its activity to cause lysosomal catabolism of Aβ but did not impact APP processing. In line with these conclusions, levels of full-length human APP were reduced in SORLAFTAF compared with SORLAWT and SORLAGGA indicating enhanced proteolytic breakdown of APP when retrograde sorting was blocked (Fig. 6A,C,D). The expression levels of BACE1 were not impacted by SORLAFTAF or SORLAGGA (Fig. 6A,B).
SORLAGGA and SORLAFTAF differentially impact amyloidogenic processing. SORLAWT, SORLAGGA, and SORLAFTAF mice were crossed with the 5xFAD line. Cortical lysates were collected from females at 8–10 weeks of age and analyzed as indicated. A–D, Quantification by ELISA documents no changes in sAPPα and sAPPβ (swAPPβ). (A), but an increase in Aβ40 and Aβ42 levels (B) in SORLAGGA compared with SORLAWT mice in the soluble brain fraction (n = 12 animals/group). No difference in Aβ40 and Aβ42 levels was seen when the membrane bound (C) or formic acid soluble fractions (D) were analyzed (n = 6–7 animals/group). E–H, ELISA documents increases in sAPPα and sAPPβ (E) and in Aβ40 and Aβ42 (F) in SORLAFTAF compared with SORLAWT animals (n = 14–15 animals/group) in the cytosolic brain fraction. Parallel experiments to E and F were performed, and quantification by ELISA documented no changes in Aβ40 and Aβ42 levels in SORLAFTAF compared with SORLAWT animals in the membrane bound (G) or formic acid soluble (H) fractions (n = 7 animals/group). Statistical analyses were performed using Student's t test (*p < 0.05; **p < 0.01; ***p < 0.001).
SORLAFTAF mice have decreased levels of human APP. A, Immunoblot analyses for APP and BACE1 in cortical brain extracts showing decreased levels of APP but not BACE1 in SORLAFTAF mice. Tubulin served as loading control. B, C, Densitometric scanning of replicate Western blots (as exemplified in A) confirms normal levels of BACE1 (B) but decreased levels of APP (C) in SORLAFTAF compared with SORLAWT and SORLAGGA animals (n = 3/group; one-way ANOVA, Tukey post hoc analysis; ***p < 0.001). D, ELISA substantiating decreased levels of APP in cortical extracts of SORLAFTAF compared with SORLAWT and SORLAGGA animals (n = 3/group; one-way ANOVA, Tukey post hoc analysis; ***p < 0.001).
The loss of SORLA's ability to mediate lysosomal catabolism of Aβ, when GGA interaction was blocked, was confirmed in further cell experiments. Thus we generated SH-SY5Y cell lines overexpressing human APP695 and either SORLAWT or SORLAGGA. Similar to the situation in brain extracts (Fig. 1D) and primary neurons (Fig. 3C), the levels of SORLAGGA in SH-SY5Y were significantly increased compared with SORLAWT. In contrast, APP levels were unchanged, indicating normal rates of processing (Fig. 7A, inset). To study the intracellular catabolism of Aβ peptides, both cell lines were treated with the γ-secretase inhibitor DAPT to block de novo production of Aβ. Subsequently, the turnover of intracellular Aβ40 and Aβ42 was determined in cell extracts as published previously (Caglayan et al., 2014). In line with an impaired cellular catabolism, intracellular levels of Aβ40 and Aβ42 decreased slower in SY5Y-GGA compared with SY5Y-WT cells (Fig. 7A,B). Slower decline of Aβ levels in SY5Y-GGA cells was not from impaired secretion of the peptides, as Aβ40 and Aβ42 levels were higher in this cell line both in the cell lysates and in the cell medium after 3 h of DAPT treatment (Fig. 7C,D). This finding argued for an impaired intracellular catabolism of nascent Aβ peptides in SY5Y-GGA cells that resulted in intracellular accumulation and, consequently, enhanced release of the peptides. To substantiate the involvement of lysosomal pathways in the turnover Aβ in cells expressing SORLA, we treated SY5Y-WT cells with DAPT in the absence or presence of chloroquine, leupeptin, and pepstatin A. Combined application of these inhibitors of lysosomal proteolysis increased the intracellular levels of Aβ40 and Aβ42 compared with SY5Y-WT treated with DAPT only (Fig. 7E,F).
Impaired ability of SORLAGGA to mediate lysosomal catabolism of Aβ peptides. A, B, SH-SY5Y cells overexpressing human APP695 together with SORLAWT or SORLAGGA were treated with 5 μm DAPT. At the indicated time points, the levels of intracellular Aβ40 (A) and Aβ42 (B) were determined by ELISA. Data represent the mean ± SEM of nine replicates expressed as percentage of time point 0 (set to 100%). Where no error bars are visible, they are smaller than the actual symbol shown. The inset in A documents expression of SORLA and APP in the cell extracts by Western blotting. C, D, Levels of Aβ40 (C) and Aβ42 (D) are increased in the cell lysate and the cell media of SORLAGGA compared with SORLAWT cells. Cells were treated with 5 μm DAPT for 3 h, and Aβ levels normalized to levels at time point 0 (n = 3 replicates). E, F, SORLAWT cells were treated with 5 μm DAPT in the absence or presence of 100 μm leupeptin, 10 μm pepstatin A, and 50 μm chloroquine (DAPT + Inhibitor). Levels of Aβ40 (E) and Aβ42 (F) in the cell lysate were determined after 3 h and normalized to levels at time point 0 (n = 6 replicates). Application of the lysosomal inhibitor cocktail reduced catabolism of Aβ40 and Aβ42 compared with the DAPT only condition. Student's t test (*p < 0.05; **p < 0.01; ***p < 0.001).
Neuronal levels of SORLAGGA were elevated compared with SORLAWT and SORLAFTAF (Fig. 1D, 3C), suggesting an increased half-life of SORLAGGA because of the absence of lysosomal targeting. This hypothesis was substantiated by pulse-chase experiments in metabolically labeled SH-SY5Y cells that documented an increased half-life for SORLAGGA (t(½) = 7.17 min) compared with SORLAWT (t(½) = 3.97) and SORLAFTAF (t(½) = 3.43 min; p < 0.05 for SORLA variant, two-way ANOVA; Fig. 8), and further supported GGA-mediated anterograde sorting of SORLA (and its cargo Aβ) to lysosomes.
Prolonged half-life of SORLAGGA compared with other receptor variants. Pulse-chase experiments were performed in SH-SY5Y cells stably transfected with SORLAWT, SORLAGGA, or SORLAFTAF, and the amount of metabolically labeled SORLA at the chase time points determined by Western blotting (as exemplified in the inset). Following densitometric scanning of replicate blots, values were normalized to time point 0 of each experiment (set to 100%). A best-fit, one-phase exponential decay curve (y = Span*e−K*x + Plateau) was determined using GraphPad Prism software, with t(½) = 0.6932/K. Based on the curve fit, the protein half-life was t(½) = 3.97 min for SORLAWT, 3.43 min for SORLAFTAF, but 7.17 min for SORLAGGA (n = 8 replicates per condition). Data represent the mean ± SEM with the best-fit lines overlaid. Two-way ANOVA analysis (interaction between time and SORLA variant, n.s., time, p < 0.001, SORLA variant, p < 0.05) indicated that the SORLAGGA cohort is significantly different from the other SORLA constructs.
Discussion
Considerable evidence implicates GGA and retromer in sorting defects in AD. Impaired activity of the adaptors is proposed to cause SORLA missorting and to enhance the amyloidogenic burden in sporadic AD (Wahle et al., 2006; Tesco et al., 2007; Muhammad et al., 2008). However, GGA and retromer also bind other cargo relevant to AD, including β-secretase (He et al., 2005; Wahle et al., 2005) or the related sorting receptors sortilin (Petersen et al., 1997; Nielsen et al., 2001) and SORCS1 (Lane et al., 2010), complicating the validation of this hypothesis. Using tailored mouse models expressing SORLA variants unable to bind GGA or retromer, we now document the functional interaction of SORLA with retromer and GGA in neurons in the brain. Levels of expression of other components of the neuronal sorting machinery or of alternative adaptor cargo were not impacted in these models. Our findings not only substantiated the relevance of SORLA and adaptor interaction in vivo, but also uncovered that anterograde and retrograde sorting pathways support discrete receptor activities relevant to AD.
Retromer emerges as a key player in sorting of proteins important for neuronal functions (Seaman, 2005; Small and Gandy, 2006; Muhammad et al., 2008; Siegenthaler and Rajendran, 2012). Dysfunction of this adaptor complex is seen in AD and PD (Small et al., 2005; Zavodszky et al., 2014). Binding of SORLA is mediated by the retromer subunit VPS26 (Fjorback et al., 2012), and disruption of this interaction causes altered localization and processing of APP in cells (Schmidt et al., 2007). Also, inactivation of retromer subunits in flies or mice increases Aβ levels (Muhammad et al., 2008) whereas pharmacological stabilization of the adaptor complex decreases APP processing (Mecozzi et al., 2014). In our mouse model, deletion of the retromer binding site in SORLAFTAF results in enhanced proteolytic breakdown of human APP as judged by the decreased levels of the precursor (Fig. 6) and a concomitant increase in all processing products (Fig. 5E,F) in the 5xFAD line. Accelerated loss of full-length APP was not obvious for the murine wild-type protein (Fig. 2A), likely because this APP species is less prone to amyloidogenic processing than the human mutant APPSwe(K670N),Florida(I716V),London(V717I) variant expressed in the 5xFAD line. In primary neurons, SORLAFTAF is depleted from the TGN but accumulates in early endosomes (Fig. 4D) supporting a model in which retrograde sorting results in sequestration of SORLA and APP in the TGN, protecting APP from processing in the secretory compartments (amyloidogenic pathway) and at the cell surface (nonamyloidogenic pathway). An increase in all APP processing products is also seen when the expression of SORLA is moderately reduced by siRNA to levels observed in individuals with SORL1 risk alleles (Schmidt et al., 2012). Jointly, these observations suggest that altered APP sorting is a likely cause of AD in sporadic cases associated with risk alleles encoding SORLA or retromer subunits. A shift of SORLA from the TGN to early endosomes, and enhanced APP processing, is also seen in mice expressing a SORLA variant lacking an acidic tail motif (Burgert et al., 2013). This motif serves as a binding site for AP1 and PACS1, two other adaptors implicated in retrograde sorting. Likely, several adaptor pathways serve a redundant function in retrograde sorting of SORLA/APP complexes, and more severe phenotypes may be expected with receptor variants lacking acidic and FTAF motifs.
Contrary to the situation with SORLAFTAF, loss of interaction with GGAs results in accumulation of SORLAGGA in the TGN and in early endosomes. Most likely, internalization of SORLAGGA molecules from the cell surface provides a sufficient supply of receptors to endosomal compartments to sustain retrograde sorting of APP (Fig. 4D). In line with this assumption, overall processing of APP is not increased in the SORLAGGA mouse (Fig. 5A). BACE, an alternative ligand for GGAs (Wahle et al., 2005), shows normal expression (Fig. 6A) and activity (Fig. 5A) in SORLAGGA mice. Apparently, this function of GGAs is not impacted by the SORLAGGA mutation. This led us to hypothesize that failure of SORLA to interact with GGAs disrupts anterograde movement of SORLA and Aβ to lysosomes, a route that regulates intracellular levels of receptor and ligand. Our studies in the human neuronal cell line SH-SY5Y cells (Fig. 7, 8) confirmed this hypothesis. We used SH-SY5Y cells in these experiments, as primary neuronal cultures generated too low a yield to successfully perform the pulse-chase pull-downs or to measure acute Aβ levels after DAPT treatment. However, our findings in SH-SY5Y cells are supported by studies in primary neuronal cultures demonstrating reduced SORLAGGA localization in Lamp1-positive vesicles compared with SORLAWT (Fig. 4C). The ability to direct Aβ to lysosomes is also blocked in a SORLA mutant identified in a family with an autosomal-dominant form of AD (Pottier et al., 2012). This mutation disrupts the binding site for Aβ in the VPS10P domain of SORLA (Caglayan et al., 2014; Kitago et al., 2015). Although human mutations in GGA or in the GGA binding domain of SORL1 have not been reported yet, our findings suggest that defects in the anterograde sorting pathway for Aβ may underlie some early onset forms of AD not linked to APP or presenilin mutations.
In conclusion, our findings established the causal relationship in a neuronal protein sorting machinery composed of SORLA and adaptors GGA and retromer, and their relevance for amyloidogenic processes in the brain (Fig. 9). Further studies are likely to define additional regulatory components of this sorting pathway central to brain (patho)physiology.
Model of SORLA trafficking pathways in amyloidogenic processes. GGAs mediate anterograde sorting of SORLA and impact SORLA-directed Aβ catabolism. An inherited mutation in SORL1 that disrupts Aβ binding, abrogating SORLA-dependent lysosomal-catabolism of Aβ, is implicated in an autosomal-dominant familial form of AD (FAD). Retromer mediates retrograde sorting of SORLA, and disrupting this pathway reduces APP levels in the TGN and results in overall enhanced APP processing rate. Low levels of SORLA or retromer components in the brain are considered risk factors promoting late-onset forms of AD (LOAD).
Footnotes
Studies were funded by grants from the Helmholtz-Association (iCEMED) and the Fritz-Thyssen-Foundation (to T.E.W.) and the Alexander-von-Humboldt Foundation (to S.B.D.). We are indebted to T. Pantzlaff and M. Schmeisser for expert technical assistance.
The authors declare no competing financial interests.
- Correspondence should be addressed to Thomas E. Willnow, Max-Delbrueck-Center for Molecular Medicine, Robert-Roessle-Strasse 10, D-13125 Berlin, Germany. willnow{at}mdc-berlin.de
