Disruption of amyloid precursor protein ubiquitination selectively increases amyloid β (Aβ) 40 levels via presenilin 2-mediated cleavage

Amyloid plaques, a neuropathological hallmark of Alzheimer's disease, are largely composed of amyloid β (Aβ) peptide, derived from cleavage of amyloid precursor protein (APP) by β- and γ-secretases. The endosome is increasingly recognized as an important crossroad for APP and these secretases, with major implications for APP processing and amyloidogenesis. Among various post-translational modifications affecting APP accumulation, ubiquitination of cytodomain lysines may represent a key signal controlling APP endosomal sorting. Here, we show that substitution of APP C-terminal lysines with arginine disrupts APP ubiquitination and that an increase in the number of substituted lysines tends to increase APP metabolism. An APP mutant lacking all C-terminal lysines underwent the most pronounced increase in processing, leading to accumulation of both secreted and intracellular Aβ40. Artificial APP ubiquitination with rapalog-mediated proximity inducers reduced Aβ40 generation. A lack of APP C-terminal lysines caused APP redistribution from endosomal intraluminal vesicles (ILVs) to the endosomal limiting membrane, with a subsequent decrease in APP C-terminal fragment (CTF) content in secreted exosomes, but had minimal effects on APP lysosomal degradation. Both the increases in secreted and intracellular Aβ40 were abolished by depletion of presenilin 2 (PSEN2), recently shown to be enriched on the endosomal limiting membrane compared with PSEN1. Our findings demonstrate that ubiquitin can act as a signal at five cytodomain-located lysines for endosomal sorting of APP. They further suggest that disruption of APP endosomal sorting reduces its sequestration in ILVs and results in PSEN2-mediated processing of a larger pool of APP-CTF on the endosomal membrane.

more synapto-and neurotoxic (1). In familial AD (FAD), overall levels of A␤ increase, or there is a shift in A␤ production that increases the ratio of A␤42/40. However, in the majority of AD cases, which occur sporadically as late-onset AD (LOAD), mechanisms of A␤ accumulation remain poorly understood, although they likely also involve reduced A␤ clearance (15).
It is now well-established that endosomes represent a key intracellular station for APP processing and that endosomal dysfunction is a pathophysiological hallmark of AD (16 -18). Several susceptibility genes associated with LOAD from genome-wide association studies (GWAS) are endosomal regulators (19 -22), and specific defects in endosomal sorting pathway components have been associated with LOAD (8,23). For example, the retromer pathway, which traffics protein cargo from the endosome to the trans-Golgi network or the cell surface, has been implicated in APP trafficking and processing and is found deficient in vulnerable brain regions from patients with LOAD (24 -27). APP and its proteolytic enzymes are transmembrane proteins that traffic through the secretory pathway and endolysosomal system, and thus the sorting and compartmentalization of these proteins dictates processing of APP into A␤ and other fragments.
In normal neuronal metabolism, A␤ is produced and secreted from the cell at low levels. However, intraneuronal increases in A␤ are toxic and precede the gross pathological changes of AD (28,29). AD-affected neurons accumulate A␤, particularly in multivesicular endosomes (MVEs) (30,31), which consist of a limiting membrane containing intraluminal vesicles (ILVs). The endosomal sorting complex required for transport (ESCRT) is a series of protein complexes (ESCRT-0 to ESCRT-III) that form ILVs by invagination of the MVE limiting membrane, sequestering specific protein cargo inside. Several studies implicate APP as a protein cargo that can be recognized and sorted by ESCRT (7,32,33). In ILVs, cargo can undergo several fates, including lysosomal degradation and extracellular secretion as exosomes. Underscoring the pathophysiological significance of ILV sorting for APP, silencing of CD2-associated protein (CD2AP), an endosomal regulator associated with increased risk for LOAD, has recently been shown to lead to an increase in intracellular A␤42 in neuronal dendrites by interfering with sorting of APP from the endosomal limiting membrane to the ILVs and thereby preventing APP degradation in lysosomes (34).
Sorting of transmembrane proteins can be controlled by ubiquitination, a post-translational modification that can signal multiple fates, including proteasomal degradation, endoplasmic reticulum-associated degradation (ERAD), endocytosis, or sorting in the MVE via ESCRT (35). Ubiquitin is a 76-amino acid peptide that is covalently attached to lysine residues of a protein cargo via sequential enzymatic reactions of E1, E2 and E3 ubiquitin ligases (35). The E3 ubiquitin ligase(s) specific for APP remains uncertain, but some studies suggest promising candidates (36,37), and a recent GWAS of Caribbean Hispanics identified F-box/leucine rich-repeat protein 7 (FBXL7), a subunit of an E3 ligase, to be associated with LOAD (38). Several reports identify ubiquitination sites on APP (37,39,40) or report alterations in A␤ levels due to changes in APP ubiquiti-nation (7,36,41), but they do not investigate in depth of the mechanism underlying this phenomenon or the downstream effects of reduced ubiquitination. APP contains five C-terminal lysines that could potentially be ubiquitinated: Lys-724, Lys-725, Lys-726, Lys-751, and Lys-763. Two studies identified APP in screens of the ubiquitome. Kim et al. (39) showed ubiquitination of APP Lys-751 and Lys-763 in a screen of HCT116 cells after proteasomal inhibition and affinity purification using an antibody that recognizes the di-glycine remnant of ubiquitinated proteins. Similarly, Wagner et al. (40) identified ubiquitination of APP Lys-751 and Lys-763 in a screen of the ubiquitome in unperturbed HEK-293T cells. Several studies have looked specifically at APP ubiquitination. Watanabe et al. (36) showed that FBL2 induces APP ubiquitination at Lys-726, leading to a reduction in A␤ generation, attributed to increased APP proteasomal degradation. El Ayadi et al. (41) showed that ubiquilin-1 induces APP ubiquitination at Lys-763 to sequester APP in the Golgi and prevent maturation. Bustamante et al. (42) showed that expression of C99, with all five C-terminal lysines substituted to arginine, is not degraded as efficiently as wild-type C99 and accumulates in Golgi-like structures, potentially due to a deficiency in ERAD. In a screen of interactors for the APP cytodomain, Del Prete et al. (37) identified a number of components of ubiquitin-processing machinery, including several E3 ubiquitin ligases and deubiquitinases, and showed that all five C-terminal lysines were ubiquitinated in mouse brain).
Thus, it appears that all five C-terminal lysines may be ubiquitinated physiologically.
We have previously identified a string of lysine residues in the juxtamembrane region of APP, mutation of which led to a redistribution of APP from ILVs to the limiting membrane and a subsequent alteration of APP processing (7). However, the terminal two APP lysines in the cytodomain were still capable of ubiquitination in that analysis; thus, it did not fully represent a model of lack of APP ubiquitination. Here, we show that loss of APP ubiquitination achieved by systematic substitution of all the cytodomain lysines residues has a profound impact on APP endosomal trafficking and metabolism. Specifically, we found that decreased ubiquitination causes defects in APP sorting into ILVs and exosomal release of APP-CTFs as well as an increase in A␤ production by PSEN2-containing ␥-secretase complex, which is localized to the endosomal limiting membrane (43). The increase is specific to A␤40, thus decreasing the A␤42/40 ratio in a manner known to be protective against A␤ aggregation and toxicity (44,45). Our data thus underscore the importance of ubiquitination as a signal for endosomal sorting of APP and contribute to growing recognition of ␥-secretase composition as a significant factor in differential APP processing.

APP lysine mutations lead to a deficiency in ubiquitination
To assess whether the five lysines present in the APP C-terminal domain (Fig. 1A) are sites of ubiquitination, we generated a panel of GFP-tagged APP mutants with each lysine mutated to arginine individually and in several combinations. We have previously shown that mutation of three consecutive lysines in the APP ubiquitin deficiency in APP trafficking and processing juxtamembrane region, Lys-724 -Lys-726, to arginine (termed APP 3R ) leads to a deficiency of ubiquitination of exogenously expressed APP by endogenous ubiquitin in HeLa cells (7). APP constructs with individual mutations K724R, K725R, and K726R allow us to identify whether all three of these residues are necessary for APP ubiquitination. Two downstream residues, Lys-751 and Lys-763, which have been identified as sites of ubiquitination in several screens (37,39,40), suggest that the APP 3R mutant may have residual ubiquitination at these two lysines. We thus generated APP mutant constructs with several combinations of lysine-to-arginine mutations, including five individual lysine mutations, the previously studied APP 3R mutant (K724R/K725R/K726R) as well as APP K751R ϩ K763R, APP K724R/K725R/K726R ϩ K751R, APP K724R/ K725R/K726R ϩ K763R, and mutation of all five lysines, termed APP 5R .
Polyubiquitination occurs when ubiquitin chains assemble at any of seven lysine residues of ubiquitin and can signal different fates for the targeted protein. The UbK48 chain directs protein cargo to the proteasome for degradation (46). To hinder this pathway, we co-expressed APP GFP constructs in cells with Ub K48R HA , a ubiquitin mutant which cannot form UbK48 polyubiquitin chains. Indeed, we are specifically interested in ubiquitination of APP as a signal for membrane trafficking, and A, hAPP C-terminal domain contains 5 lysine residues that are potential sites of ubiquitination. A␤40 and A␤42 arrows indicate sites of ␥-secretase cleavage. B, representative Western blot of APP ubiquitin immunoprecipitation (IP) in HEK-293T cells with co-expression of Ub HA and APP WT GFP and untransfected (indicated by Ϫ), untreated, or after 4 h of treatment with 20 M MG-132. C, representative Western blot of APP ubiquitin immunoprecipitation in HEK-293T cells with co-expression of Ub K48R HA and APP GFP lysine-toarginine mutant. Numbered lanes indicate APP lysine residue(s) mutated to arginine (WT ϭ wild type; 3R ϭ APP K724R/K725R/K726R; 5R ϭ all C-terminal lysines mutated). D, quantification of immunoprecipitated ubiquitin levels, normalized to immunoprecipitated APP GFP and expressed in arbitrary units (A.U.) relative to APP WT GFP . Values denote mean Ϯ S.E., n ϭ 4 biological replicates. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001 as measured by one-tailed, one sample Student's t test. The one-tailed test was used based on the prediction that mutating lysines would cause decreased ubiquitination. E, Western blot of APP ubiquitin (Ub) IP in N2a cells co-expressing APP WT GFP or APP 5R GFP and Ub K48R HA or Ub L73P mCherry , immunoprecipitated by anti-GFP. Lanes indicate both ubiquitin mutation (Ub) and APP GFP mutation. Right panel shows higher exposure of same anti-ubiquitin Western blotting. F, quantification of immunoprecipitated ubiquitin levels, with immunoprecipitated APP 5R GFP normalized to immunoprecipitated APP WT GFP in each case and expressed in arbitrary units (A.U.) relative to APP WT GFP . Values denote mean Ϯ S.E., n ϭ 3 biological replicates. *, p Ͻ 0.05 as measured by Student's t test.

APP ubiquitin deficiency in APP trafficking and processing
this allows us to study those polyubiquitin signals more likely to be involved in trafficking, such as UbK63 polyubiquitination or monoubiquitination (35). As a control, we show that the ubiquitin signal in the anti-GFP immunoprecipitate of HEK-293T cells co-transfected with APP GFP and Ub HA is a smear above ϳ130 kDa, suggesting multi-or polyubiquitination of APP WT GFP (Fig. 1B). Pre-treatment with 20 M MG-132, a proteasome inhibitor, did not alter the levels of full-length APP (FL-APP) nor did it increase the polyubiquitin smear in the immunoprecipitate, indicating that ubiquitinated APP GFP is not degraded by the proteasome. To measure the level of ubiquitination of the APP GFP lysine-to-arginine mutants, we co-expressed each in HEK-293T cells with Ub K48R HA , then immunoprecipitated the exogenous APP GFP with an anti-GFP antibody, and quantified the level of ubiquitin via Western blotting ( Fig.  1, C and D). Several APP mutants showed a decrease in ubiquitination as compared with APP WT GFP , indicating that most APP C-terminal lysine residues can be ubiquitinated or that there is a level of redundancy between the residues. Of note, K751R ϩ K763R and APP 3R ϩ K751R had consistently lower levels of ubiquitination, indicating that the Lys-751 residue may be an important residue. Although the APP 5R GFP mutant lacks all C-terminal lysines and thus should not be ubiquitinated, there was some residual ubiquitin signal in the APP GFP immunoprecipitation, which may be attributed to aberrant ubiquitination of N-terminal lysines or ubiquitination of APP GFP -interacting partners that co-immunoprecipitate. Untagged APP 5R immunoprecipitated with an anti-APP antibody still exhibited residual ubiquitin signal, suggesting that it is not due to ubiquitination of the GFP tag (data not shown). We had previously shown a deficiency in ubiquitination of APP 3R GFP (7) in studying endogenous ubiquitination of APP in HeLa cells. However, we did not see a similar deficiency in HEK-293T cells co-expressing APP 3R GFP and Ub K48R HA (Fig. 1D), possibly indicating differences in cell types or in the ubiquitin chain linkages modifying the two downstream APP lysines. This discrepancy underlines the importance of studying the APP 5R GFP mutant to understand the effect of complete lack of APP cytodomain ubiquitination.
For the remainder of the experiments, N2a cells were used instead of HEK-293T cells because 1) they are of neural origin, 2) they release detectable amounts of A␤42, and 3) their mouse origin allows us to distinguish exogenous human APP GFP fragments from endogenous murine APP. In N2a cells, the ubiquitination of APP GFP was more difficult to detect, even with cotransfection of Ub K48R HA , rendering it more difficult to quantify ubiquitination of all APP mutants in this cell line. We reasoned that this could be due to a shorter half-life of ubiquitinated species of APP GFP in this cell line, and we chose to confirm the decrease in ubiquitination of the APP 5R GFP mutant in N2a cells in a model of stabilized ubiquitination. By co-expressing APP GFP in N2a cells with Ub L73P mCherry , a ubiquitin mutant that is resistant to deubiquitinases (47), we were able to detect an accumulation of ubiquitinated APP WT GFP , as compared with cells expressing Ub K48R HA . With expression of Ub L73P mCherry , we could detect a significant reduction in levels of ubiquitination of APP 5R GFP compared with APP WT GFP ; however, there was only a trend toward significance in the N2a cells co-transfected with Ub K48R HA perhaps in part due to the lower signal-to-noise ratio (Fig. 1, E and F). This indicates that APP 5R GFP undergoes less ubiquitination than APP WT GFP , as predicted. Accumulation of ubiquitinated APP with co-expression of Ub L73P mCherry is consistent with the existence of a pool of transiently ubiquitinated APP that leads to a short-lived ubiquitination signal. Although the transient nature of this ubiquitin signal makes it difficult to detect (as also suggested by our failed attempts to profile the ubiquitination sites of APP by mass spectrometry), it may represent an important regulator of APP trafficking and processing.

Lysine-to-arginine APP mutants are highly metabolized
We next investigated the effect of ubiquitin deficiency on the metabolism of APP in cultured cells. We expressed APP GFP lysine-to-arginine mutants in N2a cells, and we measured the fragments that are produced by cleavage of FL-APP GFP by Western blotting or ELISA ( Fig. 2A). When expressed in N2a cells, all APP GFP lysine-to-arginine mutants exhibited decreased amount of FL-APP GFP (Fig. 2B). Consistent with increased processing of APP, we detected higher levels of the sAPP␣ and sAPP␤ secreted into the culture media. There was a trend toward an increase in processing with the number of APP lysines mutated, such that the APP 5R GFP mutant, with no remaining lysines in the C-terminal domain, underwent the most extensive processing. Compared with APP WT GFP , the APP 5R GFP mutant had an approximate reduction of FL-APP GFP of 50% and a corresponding 3-fold increase in cleaved soluble fragment sAPP␣ and 2-fold increase in cleaved soluble fragment sAPP␤, normalized to the expression levels of FL-APP GFP . In line with increased processing, APP GFP lysine-to-arginine mutants led to a selective increase in secreted A␤40, with no change in A␤42, thus causing a significant decrease in A␤42/40 ratios in some APP mutants (Fig. 2B). Of note, there was high variability in the levels of A␤42, likely because levels were closer to background, and the decrease in A␤42/40 ratio was only evident in APP mutants with several lysines mutated.
We chose to focus further analysis on APP 5R , the mutant with the strongest phenotype in terms of APP processing, and because the APP 3R mutant has residual ubiquitination at the two C-terminal lysines. To confirm that the increase in A␤40 generation of APP lysine-to-arginine mutants is due to deficiency of ubiquitination, we sought to determine whether artificial ubiquitination of the APP 5R mutant could reverse the phenotype. We generated constructs to express dimerizable APP and ubiquitin with DmrA and DmrC tags, respectively, based on a commercially available rapalog-induced dimerization system (48,49). By expressing APP DmrA and Ub DmrC in N2a cells and adding A/C heterodimerizer, we can induce physical proximity between APP DmrA and Ub DmrC (Fig. 3A). Indeed, artificial ubiquitination of APP 5R DmrA reduced secretion of A␤40 after treatment of heterodimerizer compared with control solution (ethanol) (Fig. 3C). A␤42 levels were also decreased for both heterodimerizer-treated APP WT DmrA and APP 5R DmrA . The reduction in A␤ is not likely due to an increase in APP DmrA degradation, because levels of FL-APP DmrA are higher in heterodimerizer-treated cells as compared with control-treated cells (Fig. 3D). Of note, the cells expressing APP 5R DmrA alone

APP ubiquitin deficiency in APP trafficking and processing
did not show a significant increase in A␤40 compared with APP WT DmrA , which we speculate may be due to a shorter collection period of culture media (12 h compared with 24 h in all other experiments). This was necessary to preserve cellular viability in conditions containing the A/C heterodimerizer. We also note that the ethanol-treated cells had lower levels of FL-APP DmrA compared with untreated and A/C heterodimerizertreated conditions, which we do not have a clear explanation for but may be due to potential effects of ethanol on APP transcription, mRNA stability, or processing. In all, these results indicate that the chemically induced addition of ubiquitin to the C-terminal domain of APP mutants that are deficient in ubiquitination is sufficient to decrease levels of A␤40, lending further support to the notion that the APP 5R -mediated increase in A␤40 is due to a ubiquitin deficiency.
Importantly, we confirmed that the APP 5R GFP mutant is similarly processed in primary neuronal cultures, using lentivirusmediated expression of APP GFP , except levels of FL-APP GFP were unchanged and levels of A␤42 secreted by primary neurons were significantly decreased (Fig. 4, A and B). APP-CTF GFP was not detectable in neuronal cultures under basal conditions; however, a band appears upon treatment with ␥-secretase inhibitor, likely representing both CTF␣ and CTF␤. APP-CTF 5R GFP accumulates ϳ75% more than APP-CTF WT GFP after ␥-secretase inhibition, suggesting that ␥-secretase cleaves more APP-CTF 5R GFP than APP-CTF WT GFP (Fig. 4, C and D). Because ubiquitin can be a signal for endocytosis, the increase in sAPP␣ from both N2a cells and primary neuronal cultures expressing APP 5R GFP led us to hypothesize that APP 5R GFP was accumulating on the cell surface, where it is cleaved by ␣-secretase. However, in a biotinylation assay to quantify cell-surface levels of APP GFP , we did not note any significant difference between surface levels of APP WT GFP and APP 5R GFP normalized to total APP GFP (Fig. 5A). We speculate that the increase in sAPP␣ may be due to an increase in cleavage of FL-APP GFP that occurs on the endosomal membrane, because this is the predominant site of APP 5R accumulation, or that ␣-secretase processing may occur in the secretory pathway before APP reaches the cell surface.
Because the APP lysine-to-arginine mutations include residues close to the ␥-secretase cleavage site, we checked whether the increase in processing by ␥-secretase could be due to a change in the interaction between ␥-secretase and the APP mutant independently of potential differences in intracellular

APP ubiquitin deficiency in APP trafficking and processing
protein sorting. To this end, we generated and purified C99 WT and C99 5R and subjected them to cleavage by purified ␥-secretase membranes in a cell-free assay, and we measured the levels of A␤40 and A␤42 produced (Fig. 5B). Using purified C99 5R as a substrate did not result in any change in A␤ levels compared with C99 WT (Fig. 5B). This suggests that the APP 5R -mediated increase in A␤40 in cell culture is not simply due to an intrinsic change in the interaction of APP 5R with the ␥-secretase complex, but it is more likely due to a change in APP trafficking, as hypothesized.

APP 5R is redistributed to endosomal limiting membrane
As membrane-bound cargo traffics through the endolysosomal system, it undergoes sorting at the endosomal mem-

APP ubiquitin deficiency in APP trafficking and processing
brane. Ubiquitin is a known signal for ESCRT, which is responsible for sorting cargo into ILVs of MVEs (35). As shown previously, APP is a cargo of ESCRT, and APP 3R accumulates on the endosomal limiting membrane instead of being sorted onto ILVs (7). To determine endosomal localization of APP 5R GFP , we co-expressed APP GFP with Rab5 Q79L mCherry , a constitutively active form of Rab5 that has been used as a tool to distinguish endosome-limiting membrane and endosome interior (i.e. ILVs) (7,33,43,50). Immunogold electron microscopy labeling in cells that do not express Rab5 Q79L and immunofluorescence of naturally occurring large endosomes previously confirmed that endogenous APP is indeed sorted into ILVs of MVEs (7). Here, we show that APP 5R GFP is partially redistributed from the endosome ILVs to the endosome-limiting membrane in N2a cells (Fig. 6, A and B). As amyloidogenic cleavage of APP is thought to occur primarily on the endosomal membrane, prevention of APP sorting into ILVs could increase interaction between APP, BACE1, and ␥-secretase, facilitating processing of APP and generation of A␤. Airyscan confocal microscopy uses a specialized detector array that allows inclusion of the fluorescent signal that would otherwise be rejected by the single confocal pinhole, and it enhances the resolution by a factor of 1.7. Airyscan images of N2a cells expressing APP GFP and Rab5 Q79L mCherry show multiple examples of round APP WT GFP vesicles inside enlarged Rab5 Q79L mCherry endosomes, which are absent in APP 5R GFP -expressing cells and thought to be APP GFP located on ILV membranes (Fig. 6C).
Protein cargo sorted onto ILVs can undergo several fates. Two downstream pathways of ILVs include degradation of ILVs and cargo via the lysosome or extracellular release of ILVs as exosomes. The missorting of APP 5R led us to hypothesize that one or both of these downstream pathways could be affected. To determine whether lack of APP ILV sorting would affect degradation of FL-APP or APP-CTFs, we inhibited protein synthesis with cycloheximide (CHX) and chased the levels of each over time (Fig. 7, A and B). Because the APP-CTFs are rapidly degraded by ␥-secretase, we applied a ␥-secretase inhibitor to identify differences in APP-CTF degradation that are not due to ␥-secretase cleavage. We did not detect any differences in FL-APP or APP-CTF degradation between APP WT and APP 5R . By adding endolysosomal proton pump inhibitor bafilomycin A1 to specifically examine changes in lysosomal degradation, we again noted no difference in degradation of APP WT and APP 5R (Fig. 7, C and D). Thus, the increase in APP

APP ubiquitin deficiency in APP trafficking and processing
processing of APP 5R is not largely due to a disruption of APP lysosomal degradation.
The disrupted ILV sorting of ubiquitin-deficient APP mutants led us to suspect that the APP content on exosomes may also be diminished. Exosomes are small extracellular vesicles that are secreted from cells upon fusion of MVEs with the plasma membrane and release of ILVs contained within. Protein cargo sorted into ILVs can be released on exosomes, including APP and APP-CTFs (51)(52)(53). To determine whether the disrupted ILV sorting of APP 5R GFP leads to a deficiency of APP on exosomes, we expressed the APP GFP mutants in N2a cells and then collected and purified exosomes from the cell culture media (Fig. 8A and as characterized in Ref. 54). Compared with APP-CTF WT GFP , APP-CTF 5R GFP is significantly reduced in exosomes (Fig. 8B), consistent with a reduced ability of the mutant to be sorted into ILVs. Levels of A␤ associated with exosomes, as measured by ELISA, reveal a trend for an increase in the levels of A␤40 associated with exosomes from APP 5R GFP -expressing cells, but not A␤42. There was a significant decrease in the A␤42/40 ratio associated with exosomes for APP 5R GFP (Fig. 8C), similar to the profile of secreted A␤ in the media. We note that the ratio of CTF GFP /AICD GFP is different for APP WT compared with APP 5R , which may indicate that mutation of AICD lysines changes its stability, because it is already known that the GFP tag artificially stabilizes AICD (55). However, establishing the effects of lysine mutation on AICD will require further investigation.

PSEN2 cleaves missorted APP, leading to A␤40 increase
The ␥-secretase complex is composed of four proteins that are necessary to form a functional enzyme complex: nicastrin, anterior pharynx defective 1 (APH1) A or B/C, presenilin enhancer 2 (PEN-2), and the catalytic subunit presenilin (PSEN) 1 or 2 (56,57). A recent paper by Sannerud et al. (43) suggested that the subcellular localization of PSEN1 and PSEN2 is vastly different. Specifically, they localized PSEN2 to the limiting membrane of the endosome, reminiscent of APP 5R GFP endosomal localization. We hypothesized that the increase in A␤40 generated by APP 5R may be due to increased cleavage by PSEN2-containing ␥-secretase complexes. To test this hypothesis, we successfully knocked down PSEN2 in N2a cells via siRNA (Fig. 9, A and B), and we observed that the APP 5R GFPmediated increase in A␤40 is abolished (Fig. 9, C and D). Sannerud et al. (43) also show that cleavage by PSEN2-containing complexes leads to an increase in intracellular A␤. Indeed, we observed that intracellular A␤40 was increased in cells expressing the APP 5R mutant (Fig. 9E). Furthermore, PSEN2 depletion in APP 5R GFP -expressing cells abrogated this increase. No significant changes were observed for A␤42. Collectively, these results corroborate our hypothesis that the higher A␤40 levels measured in APP 5R -expressing cells are due to increased cleavage by PSEN2-containing ␥-secretase complexes. Although Sannerud et al. (43) showed that PSEN1 contributes more to extracellular A␤ secretion than PSEN2, our data suggest that PSEN2-containing ␥-secretase complexes on the endosomal limiting membrane cleave an additional pool of APP-CTF GFP that accumulates when APP 5R GFP fails to be sequestered in ILVs. Unexpectedly, the reduction in the APP 5R GFP A␤42/ A␤40 ratio was unchanged upon PSEN2 depletion, although there was a greater and more significant reduction in the APP WT GFP condition. To confirm the specific role of PSEN2 in the increased secretion of APP 5R GFP A␤40, we successfully knocked down PSEN1 in N2a cells using siRNA (supplemental Fig. S1, A and B). However, PSEN1 depletion led to a compensatory increase in PSEN2 that was significant for APP 5R GFP only, which complicated interpretation of any changes in levels of A␤. Nonetheless, the depletion of PSEN1 did not lead to a significant reduction in A␤40 in APP 5R GFP -expressing cells (37.3 Ϯ 11.3% increase in A␤40 in control-treated cells compared with a 14.1 Ϯ 9.0% increase in PSEN1-depleted cells; p ϭ 0.004 and 0.372, respectively), as it did upon depletion of PSEN2. However, PSEN1 depletion did abolish the difference in A␤40 between APP WT GFP -and APP 5R GFP -expressing cells, which may be due to an overall decrease in ␥-secretase activity and reflect the large contribution of PSEN1 to cleavage of APP. Consistent with our hypothesis, the increase in intracellular

APP ubiquitin deficiency in APP trafficking and processing
A␤40 in APP 5R GFP -expressing cells is unchanged upon PSEN1 depletion (63.2 Ϯ 17.5% increase in A␤40 in control-treated cells compared with a 77.7 Ϯ 15.0% increase in PSEN1-depleted cells; p ϭ 0.002 and p Ͻ 0.0001 respectively), suggesting that PSEN2 is also more important for the intracellular accumulation of A␤40 in this mutant. Thus, our data suggest that the specific increase in A␤40 exhibited by APP 5R GFP is primarily due to cleavage of APP 5R GFP by PSEN2-containing ␥-secretase complexes subsequent to endosomal redistribution of the ubiquitin-deficient APP mutant.

Discussion
Endosomes have been subject to increasingly intense focus in understanding AD pathophysiology because subtle alterations in endosomal sorting can have important implications for the intracellular itinerary of APP and thus its amyloidogenic cleavage (9,18). We initially implicated APP as a cargo of the ESCRT pathway after an investigation into lipid changes in the AD brain and the discovery of a specific reduction in the endosomal bioactive lipid phosphatidylinositol 3-phosphate (PI3P) that was common to both FAD mouse models and human AD brains in areas associated with AD pathology (7,58). PI3P is a master regulator of endosomal signaling and binds to hepatocyte growth factor-regulated tyrosine kinase substrate, a component of the ESCRT-0 complex, via its FYVE (Fab-1, YGL023, Vps27, and EEA1) domain. Others have recognized a connection between APP and ESCRT (32,33), and mutation of an ESCRT-III component, charged multivesicular body protein 2B (CHMP2B), is associated with frontotemporal dementia (59). However, it remains unclear whether ESCRT disruption increases or decreases A␤ generation, and the study of intracellular levels of A␤ has not been consistently performed, although it might represent an important aspect of AD pathogenesis (32,33). As opposed to altering levels of ESCRT components, which are responsible for normal endosomal sorting of numerous transmembrane cargos, we instead investigated the effects of APP lysine-to-arginine mutants that cannot undergo sorting into ILVs via the ESCRT pathway.
Previous investigators have shown that APP can be ubiquitinated at several residues in the C-terminal domain, as discussed above (7, 36, 37, 39 -41). However, the role of APP ubiquitination in the secretory pathway remains unclear. We previously identified a string of three juxtamembrane lysines in the APP C-terminal domain, mutation of which leads to a deficiency of APP ubiquitination in HeLa cells with subsequent redistribution of APP to the limiting membrane of the endosome and selective increase in A␤40 in primary neuronal cultures (7). BACE1 and components of the ␥-secretase complex can also be ubiquitinated. ␥-Secretase ubiquitination has been studied in the context of the ubiquitin-proteasome system, as opposed to cellular trafficking, although some evidence suggests that ubiquilin and the proteasome are responsible for presenilin endoproteolysis (60,61). BACE1 is ubiquitinated at a single lysine in the C terminus, leading to accumulation of BACE1 in endosomes and an increase in APP processing (62). A␤ levels can be reduced by depletion of the deubiquitinase ubiquitinspecific protease 8 (USP8), which increases BACE1 ubiquitination and leads to a redistribution of BACE1 away from Rab11positive recycling endosomes (63). These studies underline the effect that ubiquitin regulators can have on A␤ production and the need for identification of ubiquitin ligases and deubiquitinases affecting APP itself. This study is an important development beyond our previous work, thanks to growing evidence that the C-terminal two lysine residues of APP can also be ubiquitinated, as discussed in the Introduction. Although our

APP ubiquitin deficiency in APP trafficking and processing
results here are predominantly consistent with those previously published in Morel et al. (7), we show here that N2a cells expressing APP 3R GFP exhibit an increase in levels of sAPP␣, whereas our prior results do not show such an increase in primary neurons expressing APP 3R GFP . Although we have no clear explanation for this discrepancy between cell types, we speculate that expression levels may be a key factor. Expression in primary neurons with lentivirus is typically lower (and closer to endogenous levels) than what is achieved with transient transfections in cell lines. Importantly, the key results are congruent between our previous results on APP 3R and our current data on APP 5R , namely the selective increase in A␤40 for both.
The increase in ubiquitinated APP immunoprecipitated from N2a cells expressing APP WT GFP with co-expression of Ub K48R HA versus deubiquitinase-resistant Ub L73P mCherry suggests that the pool of ubiquitinated APP is small or rapidly turned over. This is consistent with our inability to capture sufficient amounts of ubiquitinated APP to assess the ubiquitin chain types decorating the APP cytodomain. Recent work on misfolding-associated protein secretion showed that cells make a concerted effort to remove ubiquitin from quality control substrates prior to secretion and that similar systems could make it difficult to see accumulation of modified APP (64). The APP 5R GFP mutant shows a significant reduction of ubiquitina-

APP ubiquitin deficiency in APP trafficking and processing
tion in Ub L73P mCherry -expressing cells, and there is a clear effect of lysine mutation. Nonetheless, we cannot rule out at this point a contribution from other post-translational modifications that occur at lysines such as SUMOylation, neddylation, or acetylation (65). Importantly, chemically inducing a physical interaction between ubiquitin and the APP cytodomain was sufficient to decrease the levels of secreted A␤, even though the ubiquitin modification was at the distal end of APP, suggesting that the specific location of ubiquitination sites may not be as critical as the extent of ubiquitination present on the cytodomain.
Deficiency of APP ubiquitination also impacts the exosomal content of APP and its metabolites, with implications for AD pathology. The role of exosomes in neurodegenerative disease is unclear, but some suggest that they may play a role in the spread of pathology, by carrying misfolded proteins from one cell to another (66). A␤ can be found associated with neuronal exosomes, and along with APP-CTFs, they are increased in exosomes from AD brains (52,67,68). Exosomes may also repre-sent a protective element, whereby neuronal exosomes with abundant glycophospholipids could capture extracellular A␤ and prevent subsequent synaptic or neuronal damage (69). The APP 5R GFP mutant is not efficiently sorted to ILVs, and thus we see a reduction of APP-CTF on exosomes. Conceivably, this could have implications for the spread of APP fragments, including a reduction of APP-CTF and A␤ spread. However, the role of exosomes in neurodegenerative disease is a growing field, and it is difficult to make predictions at this early stage.
Although its significance is uncertain, intraneuronal accumulation of A␤ has long been observed to precede AD pathology. Recently, Sannerud et al. (43) have provided a link between intracellular A␤ and cleavage by PSEN2-containing ␥-secretase complexes that are specifically localized to the late endosomal/ lysosome compartment. The study showed that PSEN2 is enriched on the limiting membrane of enlarged endosomes, reminiscent of the localization of APP 5R GFP . Moreover, many more genetic FAD mutations have been found in PSEN1 as

APP ubiquitin deficiency in APP trafficking and processing
opposed to PSEN2. For these reasons, the role of PSEN2 in disease has been largely neglected until recent years. We hypothesized that the increase in APP 5R GFP -derived A␤40 is caused by aberrant cleavage of APP-CTF 5R GFP that accumulates on the endosomal membrane in proximity to PSEN2. Indeed, depletion of PSEN2 significantly rescues the increase in A␤40 seen in APP 5R GFP , both extracellularly and intracellularly to levels of APP WT GFP . Although Sannerud et al. (43) report an increase in the ratio of A␤42/40 intracellularly due to PSEN2, we saw the most significant change in levels of A␤40. This could be due to a difference in cell types or in the particular isoform of APH1 (A or B/C, as well as differently spliced isoforms) in the complex that has been shown to also affect the species of A␤ produced (70 -72).
Another recent paper identified a role for CD2AP, a genetic risk factor for LOAD, as a regulator in the sorting of APP from the endosomal limiting membrane to ILVs. Silencing of CD2AP led to an increase in the ratio of A␤42/40 extracellularly and to an increase in levels of intracellular A␤42 in dendrites, as measured by immunofluorescence, as well as a deficiency in APP degradation (34). The difference in APP metabolism observed between CD2AP silencing and removal of ubiquitination sites in APP suggests that CD2AP and ubiquitination regulate distinct pathways of ILV sorting, likely with different downstream fates (i.e. lysosomal degradation versus exosomal sorting, respectively).
The most striking phenotype of the APP 5R mutant is the selective increase in A␤40 and the resulting decrease in the A␤42/40 ratio. Most FAD mutations lead to an increase in total A␤ production or skew cleavage toward longer, more aggregate-prone species of A␤ thereby increasing the A␤42/40 ratio, possibly through a partial loss of function in the case of presenilin mutations (14,73). Selectively increasing A␤40 is protective against plaque formation by interfering with A␤42 aggre- Figure 9. APP 5R -induced increase in A␤40 is prevented by presenilin-2 depletion. A, representative Western blot of N2a cells treated with non-targeting (CTRL) or PSEN2 siRNA and expressing APP GFP . B, quantification of Western blot for PSEN1 and PSEN2; C, full-length APP (FL-APP), normalized to actin. D, secreted A␤ measured by ELISA from culture media, normalized to levels of FL-APP. E, intracellular A␤ measured by ELISA of cell lysate, adjusted to protein concentration, and normalized to FL-APP. Values are expressed relative to APP WT GFP control condition, in arbitrary units (A.U.). Values denote mean Ϯ S.E., n ϭ 10 -13 total samples from four biological replicates. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001 as measured by two way ANOVA with Bonferroni correction.
gation (74 -77) and may inhibit A␤42-induced pathology (45). Thus, reducing APP ubiquitination, which leads specifically to an increase in A␤40, may prevent pathology. However, we must emphasize caution in extending this interpretation in vivo, because the main component of amyloid deposits in cerebral amyloid angiopathy appears to be A␤40, and it is thus unlikely that an excess of A␤40 produced by decreased APP ubiquitination would be beneficial in this setting (78). These APP lysineto-arginine mutants are thus perfectly suited to studying the effect of A␤40 increase in vivo. This also has important implications in understanding mechanisms of action of ␥-secretase modulators that aim to shift the C-terminal cleavage of A␤ to reduce production of A␤42, sometimes at the expense of an increase in the relative amount of shorter A␤ species (79,80).
AD afflicts a growing number of people every year, but clinical trials thus far have been disappointing. Recently, aducanumab has shown some promise in amyloid plaque reduction and slowing cognitive decline, although we await phase 3 clinical trials (81). Elucidating the endosomal trafficking itinerary of APP is crucial for understanding the pathophysiology of AD and why promising new drugs have been ineffective. Furthermore, the recognition of the endosome as a crucial decision point in the amyloidogenic pathway opens the door for discovery of much needed novel therapeutic targets, such as ubiquitin ligases or deubiquitinases.
Human APP plasmids were generated by excising GFP from APP GFP pEGFP-N3 with corresponding APP mutations using BsrGI/SalI (New England Biolabs), followed by DNA end blunting with T4 DNA polymerase (New England Biolabs) and blunt end ligation with T4 DNA ligase (New England Biolabs).

APP ubiquitin deficiency in APP trafficking and processing
J-051123-10 and J-051123-12). All siRNAs were applied to N2a cells in 6-well plates via Lipofectamine 3000 following the manufacturer's directions, at a concentration of 25 nM. Cells and culture media were harvested after 3 days.

Cell transfection and lentivirus production
HEK-293T and N2a cells were grown at 37°C in a 5% humidified CO 2 incubator in DMEM (Invitrogen) and 10% FBS (Life Technologies, Inc.). After 24 h of plating, cells were transfected with DNA using Lipofectamine 2000 or 3000 (Invitrogen) or JetPEI (VWR) for 24 -48 h via the manufacturer's directions. Cortical neurons were obtained from P0 mice brains, as described previously (7). Briefly, cortices were separated from brain and dissociated in 2.5% trypsin (Life Technologies, Inc.). Neurons were plated on polyornithine (Sigma)-coated six-well plates and were incubated with minimum Eagle's medium (Invitrogen) with 10% FBS. After 5 h, neurons were transferred into serum-free Neurobasal medium (Life Technologies, Inc.) supplemented with B27 (Invitrogen) ϩ Glutamax (Invitrogen) and cultured for 7 days in vitro (DIV). Neurons were infected with lentivirus after 7 days in culture and were cultured up to 14 days. Where indicated, ␥-secretase inhibitor was applied 24 h before cell harvest. Lentiviruses were generated by transfecting lentiviral vectors (APP WT GFP , APP 3R GFP , and APP 5R GFP ) into HEK-293T cells using Lipofectamine LTX (Invitrogen). A pPACKH1 packaging mix (System Biosciences) was added to the transfection reagents according to the manufacturer's instructions. The medium was collected 48 and 72 h after transfection, passed through a 45-nm filter, and applied at 1:4 ratio to media.

Fluorescence microscopy
For immunofluorescence experiments, cells grown on glass coverslips were washed once with Hanks' balanced salt solution (Gibco) and fixed with 4% paraformaldehyde (Electron Microscopy Sciences) for 20 min at room temperature. Cells were then washed twice in PBS (Boston Bioproducts) and permeabilized with a buffer containing 0.05% saponin (Acros), 5% BSA in PBS for 45 min at 37°C. They were then incubated with primary antibodies diluted in the buffer for 1 h at room temperature, washed in buffer three times, incubated with fluorescent secondary antibodies (Invitrogen) diluted in buffer for 1 h at room temperature, and washed again three times in buffer. Cells were finally washed once with PBS, and coverslips were mounted in Vectashield mounting medium (Vector Laboratories). Images were acquired by confocal laser-scanning microscopy (Zeiss LSM 700 and 800) and analyzed with Zeiss Zen and ImageJ software.

Protein biochemistry and immunoblotting
To detect ubiquitinated APP, HEK-293T cells were transiently transfected with human APP GFP and Ub K48R HA and then lysed for 30 min at 4°C in IP buffer (0.5% Nonidet P-40, 500 mM Tris-HCl, pH 7.4, 20 mM EDTA, 10 mM NaF, and a mixture of protease and phosphatase inhibitors (Roche Diagnostics)), and centrifuged for 15 min at 13,000 rpm. Supernatants were diluted to equal concentration and pre-cleared with protein G-Sepharose beads (GE Healthcare) before an overnight incu-bation at 4°C with 4 g of anti-GFP mA␤ and a subsequent 2-h incubation at 4°C with protein G-Sepharose beads. After extensively washing the beads with IP buffer, proteins were eluted, separated on a 4 -12% BisTris gel (Invitrogen), and transferred by iBlot (Invitrogen) on a nitrocellulose membrane. Separate membranes were probed with anti-ubiquitin or anti-GFP antibodies. Images were acquired via the LAS4000 imager (GE Healthcare), and quantification was done using ImageJ. For all other immunoblots, cells were lysed for 30 min at 4°C in Thermo Fisher Scientific RIPA buffer or Pierce IP buffer with protease and phosphatase inhibitor mixtures and centrifuged for 15 min at 13,000 rpm, and proteins in the supernatant were processed for SDS-PAGE and immunoblotting. Tricine 10 -20% gels were used for experiments measuring APP-CTFs, and 4 -12% BisTris gels were used for all other experiments (Invitrogen). For analysis of exosomes and lysates by Western blotting, exosomes secreted by N2a cells (two dishes of 70 cm 2 , ϳ28 g of protein) were loaded in parallel to 70 g of protein lysates, i.e. 2% of a total dish. 8% acrylamide glycine gels and electrophoresis in glycine buffer were performed to detect APP GFP , CTF GFP , and Alix. Semi-dry transfer of proteins was done on 0.45-m PVDF membranes (Millipore) for 1 h. Membranes were blocked in 3% skimmed milk in Tris-buffered saline with 0.01% Tween (TBS/Tween) and incubated with primary antibodies in the blocking solution for 1 h at room temperature or overnight at 4°C. HRP-bound secondary antibodies were diluted 1:15,000 in TBS/Tween and incubated 1 h on membranes at room temperature. Detection reagent was homemade using a freshly prepared mix of 50% 250 mM luminol and 50% of 90 mM coumaric acid and 0.015% H 2 O 2 .

Purification of exosomes from N2a cells
Exosomes were isolated as described previously (54) in Mov cells. Exosome-free medium was made by ultracentrifuging complete culture medium at 150,000 ϫ g for 18 h followed by sterilization through 0.2-m filters. 24 h after plating, N2a cells were transfected, and 24 h later complete medium was replaced by exosome-free medium. Cell supernatant was collected after 20 h of secretion, and exosomes were purified by differential ultracentrifugation protocol: 2000 ϫ g, 10 min at 4°C followed by 20,000 ϫ g, 30 min at 4°C. Supernatants were filtered through 0.22-m Millex GV PVDF filter (Millipore) and finally centrifuged at 110,000 ϫ g, 1 h and 30 min at 4°C in SW41 or SW32 rotors. 110,000 ϫ g pellets were resuspended either in sample buffer (2% SDS, 2.5% ␤-mercaptoethanol, 125 mM Tris-HCl, pH 6.8, 10% glycerol, 2.5% bromphenol blue) for Western blot analysis or in lysis buffer (20 mM Tris, 150 mM NaCl, 1 mM EDTA, 1% Nonidet P-40, anti-protease mix from Roche Diagnostics) to measure protein concentration before anti-A␤ ELISA. When cell lysate was required, N2a cells were lysed in 1 ml of lysis buffer for 30 min on ice and centrifuged at 20,000 ϫ g, 10 min at 4°C before use.

Cell surface biotinylation assay
Components of the Pierce cell-surface protein isolation kit (Thermo Fisher Scientific 89881) were used with modifications to the kit protocol. N2a cells were grown in 6-well plates, transfected with APP GFP , and harvested after 24 h. Plates were APP ubiquitin deficiency in APP trafficking and processing washed with PBS on ice and incubated with biotin for 30 min at 4°C and then with quenching solution for 5 min at 4°C. Cells were then washed with TBS, scraped into 100 l of Pierce IP lysis buffer, sonicated five times for 1 s on a low setting, and centrifuged at 10,000 ϫ g for 2 min at 4°C. A sample of the total lysate was saved for Western blot analysis, and the remaining was incubated with bead slurry (50% washed beads, 50% lysis buffer) for 1 h at 4°C. Beads were collected by centrifugation, washed three times in wash buffer, and diluted in sample buffer and reducing agent for further SDS-PAGE and immunoblotting analysis.

Chemically induced ubiquitination assay
DNA constructs and heterodimerizer agent were used from the commercially available iDimerize Inducible Heterodimerizer System (Clontech). N2a cells at 60% confluence in 10-cm plates were transfected with APP WT DmrA-RFP or APP 5R DmrA-RFP alone or co-transfected with Ub DmrC . Culture media were changed 24 h after transfection, and ethanol or 2.5 M A/C heterodimerizer dissolved in ethanol were added as indicated. Culture media and cells were harvested as described above, 12 h after media change, to measure A␤40 and A␤42 levels via ELISA.

Enzyme-linked immunosorbent assays
To measure human A␤40 and A␤42, or human sAPP␤ from cell culture of N2a cells transfected with APP GFP , media were changed 24 h before cell harvest. For cortical neuronal cultures expressing APP GFP , media were collected at DIV 14, 7 days after application of lentivirus. Media were collected at 4°C and supplemented with 0.25 mg/ml Pefabloc SC 4-(2-aminoethyl)benzenesulfonyl fluoride hydrochloride (Fuka Analytical) and centrifuged at low speed to remove cell debris for 5 min. Intracellular A␤40 and A␤42 was measured from cells lysed for 30 min in 20 mM Tris, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% Nonidet P-40, and protease and phosphatase inhibitor mixtures at 4°C and then centrifuged at 13,000 rpm for 15 min. Thermo Fisher Scientific (Invitrogen) ELISA kits were used for A␤40 and A␤42 measurements from cell culture media and in vitro ␥-secretase assay. Mesoplex V-PLEX A␤ Peptide Panel 1 (4G8) kits (Meso Scale Discovery) were used for A␤40 and A␤42 measurements from cell lysate and exosomes, where an equal concentration of protein, measured by BCA assay (Pierce), was loaded for each sample. Covance kits were used for measurements of sAPP␤ from cell culture media. All kits were used according to the manufacturer's instructions.

Statistics
Statistical analysis was performed using GraphPad Prism software (version 6). All mutant samples were compared with WT using the two-tailed Student's t test (for the comparison of two averages) and the one-sample t test (for the comparison of one average to a normalized value where variability is lost). A one-tailed t test was exceptionally used in Fig. 1 based on similar experiments in previous studies (7). Two-way ANOVA was used for APP degradation experiment and PSEN2 knockdown experiment.