The trans-Golgi network is a major site for α-secretase processing of amyloid precursor protein in primary neurons

Amyloid precursor protein (APP) is processed along the amyloidogenic pathway by the β-secretase, BACE1, generating β-amyloid (Aβ), or along the nonamyloidogenic pathway by α-secretase, precluding Aβ production. The plasma membrane is considered the major site for α-secretase–mediated APP cleavage, but other cellular locations have not been rigorously investigated. Here, we report that APP is processed by endogenous α-secretase at the trans-Golgi network (TGN) of both transfected HeLa cells and mouse primary neurons. We have previously shown the adaptor protein complex, AP-4, and small G protein ADP-ribosylation factor–like GTPase 5b (Arl5b) are required for efficient post-Golgi transport of APP to endosomes. We found here that AP-4 or Arl5b depletion results in Golgi accumulation of APP and increased secretion of the soluble α-secretase cleavage product sAPPα. Moreover, inhibition of γ-secretase following APP accumulation in the TGN increases the levels of the membrane-bound C-terminal fragments of APP from both α-secretase cleavage (α-CTF, named C83 according to its band size) and BACE1 cleavage (β-CTF/C99). The level of C83 was ∼4 times higher than that of C99, indicating that α-secretase processing is the major pathway and that BACE1 processing is the minor pathway in the TGN. AP-4 silencing in mouse primary neurons also resulted in the accumulation of endogenous APP in the TGN and enhanced α-secretase processing. These findings identify the TGN as a major site for α-secretase processing in HeLa cells and primary neurons and indicate that both APP processing pathways can occur within the TGN compartment along the secretory pathway.

Amyloid precursor protein (APP) 3 is a membrane protein that undergoes proteolytic processing to produce a variety of soluble secreted products, including the cytotoxic ␤-amyloid (A␤) peptide that promotes the pathology of Alzheimer's disease. There are two processing pathways of APP, initiated by ␣and ␤-secretases, resulting in the production of distinct products. Amyloidogenic processing of APP is initiated by the ␤-secretase, BACE1, to release the luminal or ectodomain of APP (sAPP␤) and is followed by ␥-secretase cleavage to release soluble A␤. There is substantial evidence that sAPP␤ and A␤ can be generated in both the endosomal system (1)(2)(3)(4)(5) and the secretory pathway (6 -8), in particular the TGN (9 -11). In the alternative nonamyloidogenic processing pathway, APP is initially cleaved by an ␣-secretase, which then precludes the formation of the A␤ peptide and amyloid (12). The nonamyloidogenic processing of APP is generally considered to occur predominantly on the plasma membrane due to abundance of ␣-secretase at this location (13), although the possibility of ␣-secretase cleavage of APP along the secretory pathway has not been excluded (14). Given that only minor levels of APP appear to reach the PM (11), it is likely that intracellular sites of ␣-secretase cleavage are important. Defining the sites of action of ␣-secretase is relevant for understanding the regulation of APP processing by the two pathways and for the potential to divert APP processing along the nonamyloidogenic protective pathway and reduce the level of A␤ produced.
Cleavage of APP by ␣-secretase occurs at the so-called ␣-site, a site located within the A␤ domain, in order to generate a membrane-bound C-terminal fragment (␣-CTF/C83) and a large secretory sAPP␣ luminal or ectodomain (12,15). The subsequent cleavage of C83 by ␥-secretase generates a truncated A␤ peptide (p3) and APP intracellular domain (16). A disintegrin and metalloprotease 10 (ADAM10) is considered the major ␣-secretase that regulates the constitutive cleavage of APP (15,17). Newly synthesized ADAM10 is transported along the secretory pathway to the PM (18). Like other metalloproteinases, ADAM10 is synthesized as an inactive preprotein and then is activated following cleavage by a convertase within the secretory pathway, most likely in the late Golgi and/or during transport from the Golgi to the cell surface (19,20).
It is not known whether ␣-secretase cleavage of APP can occur at the TGN or is restricted to post-Golgi compartments.
The conclusion that the cell surface is the major site for APP processing by ␣-secretase is based on early studies, which have limitations. Parvathy et al. (21) showed no detectable sAPP␣ when transport along the secretory pathway was blocked at 20°C, a nonpermissive temperature for the secretory pathway, or in the presence of a biotinylated zinc metalloproteinase inhibitor (AMG2380), considered to inhibit only ␣-secretase at the cell surface (22). However, there are potential complications associated with these studies. First, it remains unclear whether ␣-secretase activity had been compromised at 20°C, and second, as biotin has been shown to diffuse freely across the PM and into intracellular compartments (23,24), it is likely that biotinylated AMG2380 may also diffuse across the PM and inhibit intracellular ␣-secretase activity. In another study, enhanced secretion of APP species was detected when the endocytosis of APP was inhibited by mutating a cytosolic sorting signal (YENPTY) of APP (13). However, the sorting motif (YENPTY) of APP is also required for the regulation of intracellular trafficking of APP (25)(26)(27)(28)(29)(30)(31)(32) in addition to endocytosis (33,34), and mutation of this APP motif could alter the anterograde trafficking pathway of newly synthesized APP. Finally, recent studies have demonstrated that newly synthesized APP is diverted at the TGN and trafficked directly to early endosomes, with only low levels detected at the cell surface (11). Given these issues, the potential importance of intracellular sites of ␣-secretase cleavage of APP remains an open question and warrants further investigation.
Previously, we demonstrated that BACE1 cleavage of APP can occur in the TGN (11), and here we have assessed whether ␣-secretase processing of APP can also take place in the late Golgi. The TGN adaptor complex, AP-4, is required for exit of APP from the Golgi, and AP-4 silencing resulted in APP accumulating at the Golgi in cultured cells (6,11). Our previous studies also showed that the small G protein, Arl5b, is required for recruitment of AP-4 to the Golgi (11). There is a block in post-Golgi cargo transport of APP upon silencing of either AP-4 or Arl5b, which is selective as BACE1 is not affected by either AP-4 or Arl5b depletion (11). Here, we have shown that retarding the exit of APP from the Golgi resulted in enhanced processing of APP by ␣-secretase in both transfected HeLa cells and primary mouse cortical neurons demonstrating that both amyloidogenic and nonamyloidogenic APP-processing pathways can occur in the late Golgi.

Detection of secreted sAPP␣ ectodomain in APP stable HeLa cells
To assess the intracellular processing of APP by ␣-secretase, we used a HeLa cell line stably expressing human APP 695 (HeLa-APP 695WT ) (11). The APP 695 isoform was selected as this is the isoform predominantly expressed in the brain (35). The HeLa-APP 695WT cell line expresses APP at only modest levels, and moreover, the expression level of APP does not saturate the trafficking machinery along the secretory pathway. To detect APP and secreted sAPP␣ in conditioned medium from HeLa-APP 695WT , we used two commercial monoclonal antibodies to APP, namely W0-2 and Y188 (Fig. 1A). The monoclo-nal mouse anti-A␤ (W0-2), raised against residues 4 -10 of the A␤ domain, recognizes full-length APP and the processed products of APP, including sAPP␣, A␤ peptide, and ␤-CTF/ C99 (Fig. 1A and Table 1). However, the monoclonal rabbit anti-APP (Y188), raised against residues 682-687 (YENPTY motif) of APP, recognizes full-length APP and the C-terminal fragments of APP, ␤-CTF/C99, and ␣-CTF/C83 ( Fig. 1A and Table 1). When HeLa-APP 695WT cell extracts were probed with antibody W0-2, two bands of ϳ89 and 98 kDa were detected either in the presence or absence of the ␥-secretase inhibitor DAPT (Fig. 1B). The upper band detected in the cell extract likely corresponds in size to the mature glycosylated full-length APP (98 kDa) and the lower band (89 kDa) to either the immature (i.e. biosynthetic precursor) and/or processed APP (6). As two bands of similar size were also detected in the cell extracts with Y188, they probably represent the mature and immature (precursor) full-length APP. A single band of ϳ89 kDa was detected in the overnight conditioned medium with W0-2. The single band in the conditioned medium could represent either sAPP␣ or immature APP as the molecular sizes of sAPP␣, and the immature full-length APPs are similar (Table 1 and Fig. 1B); in contrast, secreted sAPP␤ would be not detected by antibody W0-2 as it lacks the epitope derived from the A␤ domain. No bands were detected in the conditioned medium with the Y188 antibody (Fig. 1B), suggesting that the ϳ89-kDa band detected by W0-2 in the conditioned medium represents sAPP␣. However, antibody W0-2 would also detect the secreted product of a minor and alternative BACE1 cleavage site of APP (␤Ј-site) at residue 10 of the A␤ domain to release sAPP␤Ј. To distinguish between these products, we inhibited ␣-secretase activity in HeLa-APP 695WT cells using the metalloprotease inhibitor, TAPI-1 (Fig. 1, C and D). There was a 78.0 Ϯ 7.5% reduction in the level of the 89-kDa APP product in the 6-h culture medium of TAPI-1-treated cells, confirming that the majority of the APP ectodomain is sAPP␣. As analyses in the subsequent experiments have analyzed the effect of intracellular processing in the presence of the ␥-secretase inhibitor, DAPT, we initially assessed whether DAPT affects the levels of secreted sAPP␣. There were no obvious changes in the levels of sAPP␣ in the presence of DAPT (Fig. 1B), suggesting that the treatment of cells with DAPT is unlikely to affect ␣-secretase activity.
To assess the intracellular levels of the membrane-bound product of ␣-secretase, namely ␣-CTF/C83, cell extracts were probed with Y188 antibody; Y188 will detect both ␣-CTF/C83 and the membrane-bound product of APP cleavage by BACE1, ␤-CTF/C99 (Table 1 and Fig. 1E). An ϳ9-kDa band, the size of C83, was detected in cell extracts without the addition of inhibitors, whereas the addition of the ␥-secretase inhibitor, DAPT, resulted in an increase in the level of the 9-kDa component as well as an 11-kDa component, the size of C99, at a considerably lower level. To confirm the identity of the 9-and 11-kDa components as ␣-CTF/C83 and ␤-CTF/C99, respectively, cells were treated with DAPT and the BACE1 inhibitor C3, or with DAPT, C3, and TAPI-1. In the presence of C3, there was a reduction in the 11-kDa component but not the 9-kDa component, whereas both components were dramatically reduced in the presence of both C3 and TAPI-1. Quantitation, by normalizing for tubulin ␣-Secretase processing of APP in the Golgi levels in the cell extracts, revealed that TAPI-1 reduced the level of the 9-kDa component by ϳ80% (Fig. 1F), a reduction similar to that of the secreted APP␣ product (Fig. 1D). These data con-firm that the larger 11-kDa component of the cell extracts is ␤-CTF/C99 and the smaller 9-kDa component is ␣-CTF/C83.

Production of sAPP␣ in the secretory pathway
We have previously demonstrated that depletion of either AP-4 or Arl5b accumulated APP in the TGN of HeLa-APP 695WT cells and resulted in the increased processing of APP by BACE1 (11). To determine whether APP could also be processed by endogenous ␣-secretase in the TGN, we depleted the ⑀ subunit of AP-4 in HeLa-APP 695WT cells using two independent AP-4⑀ siRNAs (Fig. 2). As expected, there was a dramatic increase in the level of intracellular APP that colocalized with the TGN marker, GCC88 ( Fig. 2A). Quantitation of total APP pixels that overlapped with GCC88 showed an increase in colocalization of APP with GCC88 from 8.55 Ϯ 2.27% in control Monoclonal mouse anti-A␤ (clone W0-2; red) was raised against residues 4 -10 of the A␤ domain. W0-2 recognizes full-length APP, sAPP␣, A␤ peptide, and ␤-CTF/C99 fragments of APP. Monoclonal rabbit anti-APP (clone Y188; green) was raised against residues 682-687 (YENPTY motif) of APP 695 . Y188 recognizes full-length APP, ␤-CTF/C99, and ␣-CTF/C83 fragments of APP. B, HeLa cells stably expressing APP 695WT (HeLa-APP 695WT ) were treated with either DMSO (carrier control) (Ϫ) or 250 nM ␥-secretase inhibitor, DAPT (ϩ), for 16 h. Conditioned medium (CM) and cell extracts (CE) were obtained. CM (10 l) and CE (20 g) were analyzed by immunoblotting with W0-2 or Y188 antibodies, using a chemiluminescence detection. The bands that correspond to sAPP␣ is marked (*). C, HeLa-APP 695WT cell monolayers were treated either DMSO (carrier) or with 50 M TAPI-1 (␣-secretase inhibitor) for 6 h. Conditioned medium was collected, and equivalent volumes of samples, based on protein content of extracts of cell monolayers, were analyzed by immunoblotting with W0-2 antibodies. Cell extracts were analyzed by immunoblotting with mouse anti-␣-tubulin antibodies. D, bar graph representing the percentage reduction of sAPP␣ levels in C. Data are represented as the mean Ϯ S.D. of three independent experiments and analyzed by paired, two-tailed Student's t test. ***, p Ͻ 0.001. E, HeLa-APP 695WT cell monolayers were treated either DMSO (carrier) or 250 nM DAPT in the presence or absence of 2 M C3 (BACE1 inhibitor) and/or 5 M TAPI-1 for 16 -20 h. CM was analyzed by immunoblotting with W0-2 antibodies. CE were immunoblotted with either mouse anti-␣-tubulin antibodies or rabbit anti-APP (Y188) antibodies. F, bar graph representing the percentage reduction of ␣-CTF/C83 levels in E. Data are represented as the mean Ϯ S.D. of three independent experiments and analyzed by paired, two-tailed Student's t test. ***, p Ͻ 0.001.  2B). Two independent AP-4⑀ siRNAs demonstrate that the accumulation of APP in the TGN was specifically associated with AP-4 depletion. Given the higher level of APP in the TGN, AP-4⑀ siRNA-2 was chosen for the subsequent experiments. AP-4⑀ siRNA-2 showed Ͼ90% reduction in levels of AP-4⑀, whereas AP-1, another adaptor of the TGN, was not affected in RNAi-treated cells (Fig. 3A). Analyses of the overnight conditioned medium for sAPP␣ by blotting with antibody W0-2 revealed a 1.37-fold increase in the level of secreted sAPP␣ in medium from cells treated with AP-4⑀ siRNA compared with the control siRNA-treated cells (Fig. 3, A and B). Arl5b, which is required for the TGN recruitment of AP-4 (11), was also silenced using conditions previously established to reduce Arl5b mRNA levels by 90% (36). Antibodies are not available to Arl5b, and we established in these experiments that Arl5b was effectively silenced in GFP-Arl5b stable HeLa cells (Fig. 3A). Treatment of HeLa-APP 695WT cells with Arl5b siRNA also resulted in an increase in the level of secreted sAPPs␣ by 1.65-fold when compared with control conditions (Fig. 3, A and B). These findings demonstrate that the accumulation of APP in the TGN enhances APP processing by ␣-secretase and secretion of sAPP␣.

Nonamyloidogenic and amyloidogenic processing of APP in the TGN
Next, we compared the nonamyloidogenic and amyloidogenic processing of APP in the TGN following the accumula- HeLa-APP 695WT cells were transfected with either control siRNA, AP-4⑀ siRNA-1, or AP-4⑀ siRNA-2, as indicated, for 72 h. Monolayers were fixed and stained with mouse anti-human APP antibodies (clone NAB228) (red), rabbit anti-GCC88 antibodies (green) as a TGN marker, and DAPI (blue). Higher magnification of the merged images of the cells marked * are also shown. Bars represent 10 m. B, percentage of APP at the TGN was calculated as a percentage of total APP pixels that overlapped with GCC88 using the OBCOL plugin on ImageJ. Data are represented as the mean Ϯ S.D. of three independent experiments (n ϭ 12) and analyzed by unpaired, two-tailed Student's t test. ***, p Ͻ 0.001.

␣-Secretase processing of APP in the Golgi
tion of APP in this compartment. Upon depletion of AP-4⑀ in HeLa-APP 695WT cells, there was a 1.40 -1.50-fold increase in the level of full-length APP compared with control conditions when immunoblots were probed with either antibody Y188 or W0-2 (Fig. 4, A and B). qPCR analyses revealed that APP mRNA was not increased in AP-4⑀ siRNA-treated cells, rather APP mRNA was reduced by 30% (Fig. 4C); hence, the accumulation of APP in AP-4⑀ siRNA-treated cells was not due to an increase in APP expression, rather it is likely that the increase is due to a reduced rate of turnover. However, there was no obvious alteration in the level of full-length APP in the presence of the ␥-secretase inhibitor (DAPT) or the BACE1 inhibitor, C3, in AP-4⑀ siRNA and control siRNA-treated samples (Fig. 4A).
To assess the intracellular levels of ␣-CTF/C83 and ␤-CTF/ C99, the membrane-bound products of ␣-secretase and BACE1, respectively, cells were treated with DAPT and cell extracts probed with Y188 and W0-2 antibodies. Y188 will detect both ␤-CTF/C99 and ␣-CTF/C83, and W0-2 will detect only ␤-CTF/C99 (Table 1 and Fig. 4A). In the absence of DAPT and C3 inhibitors, the 9-kDa ␣-CTF/C83 was detected with the Y188 antibody in both control siRNA-and AP-4⑀ siRNA-treated cells (Fig. 4A). A slightly higher level of the ␣-CTF/C83 component was detected in AP-4⑀-depleted cells compared with control conditions (Figs. 4A and 5A). In the presence of DAPT (for 16 h), which protects CTFs from further processing, a 2.05-fold increase in the level of ␣-CTF/C83 was detected in AP-4⑀depleted cells compared with control conditions (Figs. 4A and 5B). These results demonstrate that the C83 ␣-secretase product is rapidly cleaved by ␥-secretase.
In addition, a component of ϳ11 kDa was detected with both Y188 and W0-2 antibodies, consistent with the identity of ␤-CTF/C99. ␤-CTF/C99 was not detected in control siRNA-treated cells and only at very low levels in cells treated with AP-4⑀ siRNA in the absence of DAPT (Fig. 4A). In the

␣-Secretase processing of APP in the Golgi
presence of DAPT, both Y188 and W0-2 antibodies revealed a 2.60 -2.80-fold increase in the level of intracellular ␤-CTF/C99 in AP-4⑀depleted cells compared with control conditions (Fig. 4A). Notably, when ␥-secretase is inhibited by DAPT, ␣-CTF/C83 levels were considerably higher than ␤-CTF/C99, indicating that ␣-secretase processing of APP was the major processing pathway.
Cells were also treated with the BACE1 inhibitor C3 and/or DAPT. As expected, in the presence of C3, with or without DAPT, ␤-CTF/C99 was not detected by either Y188 or W0-2 antibody even in cells depleted of AP-4⑀ (Fig. 4A). This observation further verifies the identity of the 11-kDa component as ␤-CTF/C99, and the ϳ9-kDa component detected by the Y188 antibody, and not by W0-2, as ␣-CTF/C83. In the presence of C3 and DAPT, a 2.44-fold increase in the level of ␣-CTF/C83 was detected in AP-4⑀-depleted cells compared with control conditions (Figs. 4A and 5D), again demonstrating that APP is processed by ␣-secretase in the TGN. Depletion of Arl5b in HeLa-APP 695WT cells also showed similar increases in ␤-CTF/ C99 and ␣-CTF/C83 (Fig. S1), as for AP-4⑀-depleted cells.
We also analyzed the level of sAPP␣ in AP-4⑀-depleted cells treated with DAPT and C3. As expected from Fig. 3, increased levels (Ͼ2-fold) of 89-kDa sAPP were detected in the culture supernatant following AP-4 knockdown of HeLa-APP 695WT cells (Fig. 6A). The absence of detectable ␣-tubulin in the conditioned media indicates the 89-kDa protein is a secreted product and is not derived from cell lysis (Fig. 6, A and B). Similar levels of 89-kDa sAPP␣ were detected in the culture supernatant regardless of C3 treatment (Fig. 6A), indicating that the inhibition of BACE1 does not enhance ␣-secretase processing. In contrast, treatment of AP-4 knockdown cells with TAPI-1 dramatically reduced the level of the 89-kDa secreted product by 75-80% (Fig. 6B), confirming the identity of the 89-kDa sAPP from AP-4⑀depleted cells as sAPP␣.

Enhanced nonamyloidogenic processing of endogenous APP by AP-4 knockdown in neurons
To determine whether ␣-secretase processing occurs in the TGN of neurons, we initially assessed whether silencing AP-4 in primary mouse neurons resulted in retention of APP in the TGN. To silence AP-4 in primary mouse cortical neurons, neurons were transduced with AP-4⑀ shRNA recombinant lentivirus, and transduced neurons were detected by GFP expression. GFP-positive neurons had a dramatically reduced staining of AP-4⑀, as detected by confocal microscopy (Fig. 7A). In addition, staining for endogenous APP showed an increase in APP that colocalized with the TGN marker, p230/golgin-245 (Fig.  7B). Quantitative analyses, using Manders' coefficient, revealed a significant increase in APP colocalization with the TGN marker, p230/golgin245, in AP-4 -depleted neurons (Fig. 7C). Hence, depletion of AP-4 in primary neurons results in an accumulation of endogenous APP in the TGN.
To determine whether AP-4 increased the level of either ␣-secretase or BACE1 processing, we first assessed the level of 11-kDa ␤-CTF/C99 and 9-kDa ␣-CTF/C83 in neurons with and without DAPT. As for HeLa-APP 695WT cells, 11 kDa (marked *) and 9 kDa (marked **), components were detected in the extracts of primary mouse neurons with the Y188 antibody, both of which increased in the presence of the ␥-secretase inhibitor DAPT (Fig. 8A), demonstrating that the C-terminal fragments of APP are rapidly processed by ␥-secretase in primary neurons as for HeLa cells. In the presence of the C3 inhibitor, there was a reduction in the 11-kDa component, whereas both components were dramatically reduced in the presence of both C3 and TAPI-1. These data confirm that the larger 11-kDa component of neuron cell extracts recognized by the Y188 antibody is ␤-CTF/C99, and the smaller 9-kDa component is ␣-CTF/C83.
Next, we determined whether depletion of AP-4 increased the level of either ␣-secretase or BACE1 processing. We used two different shRNA targets to AP-4 and assessed the reduction of AP-4 in the population of recombinant lentivirustransduced neurons by immunoblotting. AP-4⑀ was depleted by 37-41% in AP-4⑀ shRNA1 and AP-4⑀ shRNA2 lentivirustransduced neuron population compared with nontransduced neurons (Fig. 8B). The transduction efficiency of the transduced neuronal populations, based on GFP expression, was 30 -40%, indicating that the depletion of AP-4 in the transduced cells was Ͼ90%. In the presence of DAPT, there was an increase in the level of both ␤-CTF/C99 and ␣-CTF/C83 in AP-4⑀-shRNA lentivirus-transduced neurons compared with ␣-Secretase processing of APP in the Golgi nontransduced neurons (Fig. 8, B-D). ␣-CTF/C83 was increased by 1.93-and 2.17-fold, and ␤-CTF/C99 was increased by 1.65-and 1.61-fold in AP-4⑀ shRNA1 and AP-4⑀ shRNA2 lentivirus-transduced neuron populations, respectively, compared with nontransduced neurons (Fig. 8, B-D). In addition, we also determined the increase in total full-length APP in AP-4 -depleted neurons (Fig. 8, B, and E). There is a trend for an increase in the total level of APP (122.3 Ϯ 10.0% for shRNA1 and 132.7 Ϯ 17.6% for shRNA2), although not statistically significant from the control sample based on a paired t test (p ϭ 0.061 and p ϭ 0.094, respectively). Nonetheless, given that the transduction efficiency was 30 -40% in these neuronal popula-tions, the data are consistent with the findings in HeLa-APP 695WT cells depleted of AP-4. Taken together, AP-4 depletion retains APP in the TGN and results in an increased level of both ␤-CTF/C99 and ␣-CTF/C83, demonstrating that AP-4 depletion enhances intracellular APP processing by both BACE1 and ␣-secretase, respectively.

Discussion
Under physiological conditions, the nonamyloidogenic and amyloidogenic processing pathways of APP are considered to represent the major and minor pathway in cells, respectively (6,37). The cleavage and shedding of the luminal, or ecto, domain of APP by the metalloprotease ␣-secretase is a key processing event, as it is considered to minimize the generation of the pathogenic A␤ peptides associated with Alzheimer's disease (38,39). In addition, soluble sAPP␣ is known to have neurotrophic (40) and neuroprotective (41)(42)(43) properties. APP cleavage by membrane-bound ␣-secretase can occur at the plasma membrane (13,44); however, other cellular location(s) of ␣-secretase cleavage remain poorly defined. An important unresolved question is whether ␣-secretase can cleave APP along the anterograde transport pathway. Our findings demonstrate that ␣-secretase cleavage of APP occurs in the TGN; moreover, the TGN may represent a major site for nonamyloidogenic processing. This conclusion is based on the following observations presented in the paper. 1) Depletion of AP-4 or Arl5b retarded the exit of APP from the TGN, resulting in APP accumulation in the TGN, and dramatically enhanced the cleavage of APP by endogenous ␣-secretase, as well as BACE1, in stable HeLa cells expressing APP 695WT and primary neurons. 2) Depletion of AP-4 or Arl5b resulted in an increase in the secretion of soluble luminal domain of APP␣ in stable HeLa cells. 3) Depletion of AP-4 also resulted in the accumulation of endogenous APP in the TGN of primary mouse cortical neurons and enhanced ␣-secretase processing of APP in neurons. 4) Overall, a 1.5-3.0-fold increase in APP at the TGN resulted in an increase in the ␣-secretase APP product, ␣-CTF/C83, in the cell by Ͼ200% in both stable HeLa cells and primary neurons, indicating that the TGN represents a major site for ␣-secretase processing.
The use of the two anti-APP antibodies, Y188 and W0-2, together with inhibitors of the three secretases allowed the identification of sAPP␣ in the conditioned medium, and ␣-CTF/C83 and ␤-CTF/C99 in cell lysates, based on immunoblotting analyses. sAPP␣ in the conditioned medium is recognized by the W0-2 antibody, whereas sAPP␤ is not detected by antibody W0-2 as it lacks the epitope. However, an alternative minor BACE1 cleavage site of APP, the ␤Ј-site at residue 10 within the A␤ domain (45), would result in the release of the product sAPP␤Ј and would be recognized by the W0-2 antibody. We demonstrated that most if not all the secreted 98-kDa ectodomain of APP in the culture medium represents sAPP␣, as the BACE1 inhibitor C3 did not affect the level of sAPP; moreover, the ␣-secretase inhibitor, TAPI, reduced the levels of sAPP by 78 -85%. Using the ␥-secretase inhibitor (DAPT) and/or BACE1 inhibitor (C3), we have also verified the identity of the C-terminal fragments of secretase cleavage of APP as ␣-CTF/C83 and ␤-CTF/C99 by immunoblotting analyses with

␣-Secretase processing of APP in the Golgi
the two antibodies, Y188 and W0-2. The C83 product was unaffected by treatment with the BACE1 inhibitor, whereas the C99 product was sensitive.
Treatment with TAPI inhibited the majority of the ␣-secretase cleavage of APP. There are three members of the ADAM family, namely ADAM9 (46), ADAM10 (20), and ADAM17 (TACE) (47), that have been reported to exhibit ␣-secretase activity. Moreover, ADAM10, but not ADAM9 and ADAM17, has been identified as the physiologically essential ␣-secretase that regulates the constitutive cleavage of APP in primary mouse neurons (15,17).
In contrast to a number of earlier studies, which have proposed that APP is processed by ␣-secretase mainly at the cell surface (13), we have shown that APP can also be processed by ␣-secretase in the TGN. Depletion of AP-4 and Arl5b allows the selective block of cargo transport from the TGN to endosomes. We and others have shown that post-Golgi transport of APP in HeLa cells depends on AP-4 and Arl5b (6,11). The block in post-Golgi cargo transport of APP is selective as BACE1 is not affected by either AP-4 or Arl5b depletion (11). Upon depletion of AP-4 in HeLa cells, there was increased processing of APP by ␣-secretase as demonstrated by an increased level of secreted sAPP␣ and, in the presence of DAPT, a 2-fold increase in intracellular ␣-CTF/C83, compared with control conditions. In neurons there was a similar increase, given the transduction efficiency of 30 -40%. These findings demonstrate that the cleavage of APP by endogenous ␣-secretase is enhanced when the Golgi export of APP was impaired. Depletion of AP-4 also resulted in increased cleavage of APP by BACE1; however, the cleavage of APP by ␣-secretase in the TGN was four times higher than BACE1. These results indicate that the nonamyloidogenic processing of APP represents the major pathway and the amyloidogenic processing the minor pathway in this compartment. In addition, as only low levels of CTFs were detected in either untreated cells or in AP-4 -depleted cells in the absence of DAPT, these findings indicate that ␥-secretase rapidly processes ␣-CTF/C83 and ␤-CTF/C99. As the CTFs contain the cytoplasmic tail of APP with the motifs for binding AP-4, the post-Golgi transport of these C-terminal membrane fragments are also likely to be retarded in AP-4 -depleted cells; hence, the rapid processing by ␥-secretase is also probably occurring in the TGN. Rapid processing of C83 by ␥-secretase has been reported previously (15) and is consistent with the known intracellular location of ␥-secretases (7,48,49). Given that the increase in APP levels in the Golgi parallels the increase in ␣-secretase processing detected in the entire cell, and that the ratio of ␣-secretase processing and BACE1 processing remains unchanged following AP-4/Arl5b depletion, our findings indicate that the Golgi is likely to represent a major site under physiological conditions.
A key finding in our study is that the TGN adaptor, AP-4, plays an important role in the post-Golgi export of APP in primary neurons. To date, the impact of AP-4 on post-Golgi trafficking of APP has only been assessed in nonpolarized HeLa cells (6,11). Our finding that AP-4 is also relevant to post-Golgi trafficking of APP in primary neurons is an important verification of the previous work in non-neuronal cells and has broader

␣-Secretase processing of APP in the Golgi
impact to the relevance of AP-4 mutations in neurological diseases and disorders in mice and humans (50 -52).
Significantly, primary cortical neurons showed enhanced levels of ␣-secretase processing after AP-4 depletion, with results very similar to the HeLa-APP 695WT stable cell line. The finding that ␣-secretase processing of endogenous APP in mouse neurons also occurs in the TGN, with a similar ratio of ␣-secretase processing and BACE1 processing as for HeLa cells, supports the relevance of our findings using HeLa-APP 695WTtransfected cells and also indicates shared characteristics of the TGN between primary neurons and cultured HeLa cells.
A previous study also indicated that ␣-secretase can cleave APP in the TGN by using an APP construct with the cytoplasmic domain of furin replacing the APP tail (53). This APP/furin construct is concentrated in the Golgi under steady-state conditions (53) and resulted in enhanced ␣-secretase processing compared with WT APP. However, a complication of this earlier study, which was not addressed, is that the sorting motifs in the cytoplasmic domain of furin directs recycling between the plasma membrane and the TGN via the late endosomes (54); therefore, the APP/furin hybrid construct will be recycling between the Golgi and the plasma membrane with a different intracellular itinerary compared with WT APP. Hence, it was not clear in this earlier study (53) whether the increased ␣-secretase processing occurred in the Golgi or elsewhere in the intracellular recycling pathway. Our approach overcomes the limitation of the previous study by modulating the trafficking machinery and by characterizing the processing of endogenous APP in primary cells.
The majority of APP is processed by the ␣-secretase pathway (6,37). It is of interest that the balance of ␣-secretase and BACE1 cleavage of APP is similar in untreated and AP-4depleted cells, indicating that the ratio of ␣-secretase and BACE1 processing remains unchanged. This finding suggests that there is a tight regulation in APP processing along the two pathways, even within the one compartment. Of relevance to this observation is that the TGN is known to be organized into distinct membrane domains (55,56). Perhaps the different secretases and APP are segregated into distinct domains within the TGN, which in turn would dictate the level of physical proximity between the enzymes and substrate and thereby regulate the cleavage events. A consequence of this scenario is that any perturbation of the partitioning of the individual membrane proteins, or of the subdomain organization of the TGN, would then influence the relative levels of processing by the ␣and ␤-secretases.
A potential caveat to the interpretation of the data presented in this study is that depletion of the TGN-localized AP-4 may have disrupted the distribution of ␣-secretase, ADAM10, thereby enhancing the cleavage of APP. However, we consider this unlikely as AP-4 is an adaptor that selectively transports cargo to the early endosomes and is not required for transport of cargo directly to the cell surface (6,57,58). Of relevance is that ␣-secretases are most likely transported directly from the Golgi to the cell surface (39).
In conclusion, our findings demonstrate that ␣-secretase can process APP in the TGN, and the accumulation of APP in this Golgi compartment enhances processing of APP along both the amyloidogenic and nonamyloidogenic pathways.

Primary mouse cortical neuronal cultures
All experiments carried out on animals were approved by the Animal Ethics Committee, University of Melbourne (approval number 1212502.1), in accordance with animal ethics guidelines. Pregnant mice (C57BL/6) at gestational days 15-16 were euthanized by CO 2 asphyxiation, and the embryos collected by cesarean section. Primary cortical neurons were prepared from the collected embryos, as described previously (63), and resuspended in Neurobasal medium supplemented with 2.5% B-27, 0.25% GlutaMAX, and 100 units/l penicillin and 0.1% streptomycin (complete NBM) (Life Technologies, Inc.). Cells were plated at a density of 0.6 ϫ 10 5 and 5 ϫ 10 5 cells/well in 12-and 6-well plates, respectively. After 24 h, the medium was replaced with fresh complete NBM. Cells were grown for 3 days (DIV 3) before transduction. All experiments conducted on animals were approved by the Animal Ethics Committee, University of Melbourne (Ethics ID: 1613960).

Generation of lentivirus
Lentivirus was generated as described previously (64). HEK293T packaging cells were seeded at a density of 2 ϫ 10 6 cells (DMEM, 10% v/v FCS, 1% v/v GlutaMAX TM ) in 10 cm by 10-cm tissue culture dishes. Prior to transfection, the existing medium was replaced with 6 ml of fresh medium (DMEM, 10% v/v FCS, 1% v/v GlutaMAX TM ). For calcium phosphate transfection of each 10 ϫ 10-cm dish, 7.5 g of pCMV-VSV-G, 3 g of psPAX2, 10 g of pGIPZ-shRNA, 250 l of UltraPure TM H 2 O (ThermoFisher Scientific), and 250 l of 0.5 M CaCl 2 were prepared in this order (DNA mix). The DNA mix was added dropwise to 500 l of 2ϫ HEPES-buffered saline, pH 7.0 (VWR, Avantor). The transfection mix was vortexed for 30 s and incubated for 10 min at room temperature. The transfection mix was added to cells and incubated overnight in a humidified 10% CO 2 incubator at 37°C. The medium was then replaced with 6 ml of fresh medium (DMEM, 10% v/v FCS, 1% v/v Gluta-MAX TM ). Cells were incubated for another 48 h in the 37°C incubator to allow the production of viral particles. Virus-containing medium was harvested and centrifuged at 3000 ϫ g for 30 min at 4°C to pellet cell debris, and the viral supernatant was filtered using Steriflip-HV sterile centrifuge tube top filter unit (Merck). PEG-it Virus Precipitation Solution (Integrated Sciences, Australia) was added to the filtered viral medium and incubated overnight at 4°C. Viral particles were pelleted by centrifugation at 1500 ϫ g for 30 min at 4°C, and the viral pellet was resuspended in DMEM/NBM supplemented with 10 mM HEPES (Gibco ThermoFisher Scientific) at 1:100 of its original harvested volume. Concentrated virus was allocated into cryovials, snap-frozen on dry ice, and stored at Ϫ80°C. All lentiviral waste was discarded in sodium hypochlorite.

Transduction of primary mouse cortical neuronal cultures
Primary mouse cortical neurons were seeded and cultured as monolayers in 12-well plates with or without coverslips for lentivirus transduction. Concentrated lentivirus was thawed at room temperature before use. Neurons were transduced at DIV 3 with 10 -15 l/well of AP-4⑀ shRNA or AP-1␥ shRNA lentivirus diluted with fresh C-NBM to a final volume of 200 l/well. Transduced neurons were incubated for 24 h (DIV 4) in a 37°C incubator with 10% CO 2 , followed by replacement of lentiviral medium with conditioned C-NBM (half-fresh C-NBM and half-C-NBM recovered from neuronal culture) and incubated for a further 72 h (DIV 7).

Quantitative real-time PCR
Total RNA was purified using Qiagen RNeasy mini kit (Qiagen, Germany) from ϳ2 ϫ 10 6 cells treated with either control siRNA or AP-4⑀ siRNA-1 for 72 h. RNA preparations were treated with RNase-free DNase set (Qiagen) during the purification process as per manufacturer's protocol. RNA samples were stored at Ϫ80°C. The RNA integrity number (RIN) of the RNA samples was determined (65). Briefly, RNA samples were diluted to 250 ng/l and prepared in RNA ScreenTape sample buffer (Agilent Technologies). The RNA sample mix was loaded on the RNA analysis ScreenTape (Agilent Technologies) and analyzed by the 4200 TapeStation instrument (Agilent Technologies) located at the Department of Anatomy and Neuroscience, University of Melbourne. Total RNA (1 g) was then converted to cDNA using a SuperScript III First-Strand Synthesis kit (Life Technologies, Inc., and Thermo-Fisher Scientific) according to manufacturer's protocol. To set up singleplex reactions (20 l/reaction) in quadruplicate for APP (Hs00169098_m1) and internal reference gene GAPDH (Hs02758991_g1), the cDNA sample (1:20 dilution) was prepared with TaqMan Fast Advanced Master mix (Applied Biosystems and ThermoFisher Scientific) and human-specific TaqMan gene expression assays (Applied Biosystems and ThermoFisher Scientific). qRT-PCR was conducted using QuantStudio 6 qPCR instrument (Thermo-Fisher Scientific) and analyzed using the comparative ⌬⌬CT method. Control siRNA cells were used as normalizer, and GAPDH was used as the internal reference gene for the relative quantification of mRNA levels.

Indirect immunofluorescence
Monolayers of cultured mammalian cells and primary mouse cortical neurons were fixed in 4% paraformaldehyde (PFA) (Wako Pure Chemical Industries, Japan) for 15 min at room temperature, quenched in 50 mM NH 4 Cl/PBS for 10 min at room temperature, permeabilized with 0.1% v/v Triton X-100/ PBS for 4 min at room temperature, and blocked in blocking solution (5% v/v FCS and 0.02% v/v sodium azide, in PBS) for 30 min to reduce nonspecific binding. For AP-4⑀ staining, when required cells were fixed and permeabilized with chilled 100% methanol at Ϫ20°C for 5 min, washed four times in PBS, and blocked in blocking solution for 30 min at room temperature. Cultured cells were stained with primary antibodies diluted in blocking solution for 1 h at room temperature and washed six times with PBS. Neurons were stained overnight at 4°C with primary antibodies and washed six times with PBS. Cultured cells and neurons were then stained with fluorophore-conjugated secondary antibodies diluted in blocking solution for 30 min at room temperature and washed six times with PBS. Coverslips were washed with distilled water and mounted in Mowiol (10% w/v Hopval 5-88, 25% w/v glycerol, 0.1 M Tris in milli-Q water) on microscope glass slide. Images were acquired sequentially for multicolor imaging on a confocal laser-scanning microscope (Leica TCS SP8 confocal imaging system) using a ϫ63 1.4 NA HCX PL APO CS oil immersion objective and a Leica HyD photodetector. GFP and Alexa Fluor 488 were excited using the 488-nm line source of an argon laser. Alexa Fluor 568 and Alexa Fluor 647 were excited with the 543-and 633-nm helium-neon (HeNe) laser, respectively. DAPI was excited with a 405-nm UV laser.

Cell extracts
HeLa cells and primary mouse cortical neurons were extracted in radioimmunoprecipitation (RIPA) lysis buffer (50 mM Tris-HCl, pH 7.3, 150 mM NaCl, 0.1 mM EDTA, 1% w/v sodium deoxycholate, 1% v/v Triton X-100, 0.2% w/v NaF, and 100 M Na 3 VO 4 ) supplemented with 1ϫ cOmpleteTM Mini Protease Inhibitor Mixture (Roche Applied Science, Sigma, and Merck). Cell lysates were incubated on ice for 10 min followed by centrifugation at 16,000 ϫ g at 40°C for 10 min. The supernatant was collected and protein concentration determined by the Bradford protein assay (Bio-Rad) using bovine serum albumin (BSA) protein standards (Pierce BSA standard, Thermo-Fisher Scientific).

␤-CTF/C99 and ␣-CTF/C83 blots
To detect levels of ␤-CTF/C99 and ␣-CTF/C83 in HeLa-APP 695WT cells, 15 g of cell extracts were prepared in 2ϫ reducing sample buffer containing 10% ␤-mercaptoethanol and boiled at 100°C for 7 min. Samples were resolved on 12% NuPAGE TM BisTris SDS-polyacrylamide gel (Invitrogen and ThermoFisher Scientific) at 125 V for ϳ2.5 h. Transfer of proteins onto 0.2-m nitrocellulose membrane (Bio-Rad) was conducted at 400 mA for 1 h on ice. The membrane was incubated in warm PBS for 5 min and blocked with 10% w/v skim milk/ PBS Tween 20 for 1 h at room temperature to reduce nonspecific binding. To probe for ␤-CTF/C99, the membrane was incubated overnight at 4°C with W0-2 antibodies diluted 1:6000 in PBS Tween 20. To probe for ␤-CTF/C99 and ␣-CTF/ C83, the membrane was incubated overnight at 4°C with Y188 antibodies diluted in PBS Tween 20 (1:6000). The membrane was washed three times with PBS Tween 20 at 10 min/wash and incubated with 1:500 diluted secondary antibodies conjugated with HRP for 1 h and washed as above. An ECL Western blotting chemiluminescence detection system (GE Healthcare) was used to detect bound antibodies, and the protein band signals were imaged using the Gel-Pro TM Analyzer version 4.5 software (Media Cybernetics, Bethesda, MD). The Gel-Pro TM Analyzer software was used to measure the densitometry of the protein bands.
To detect levels of ␤-CTF/C99 and ␣-CTF/C83 in primary mouse cortical neurons, 10 g of cell extracts were prepared as above. Samples were resolved on 12% NuPAGE TM BisTris SDSpolyacrylamide gel and proteins transferred onto 0.2-m nitrocellulose membrane as described above. The membrane was incubated overnight at 4°C with Y188 antibodies diluted in PBS Tween 20 (1:6000) and washed three times with PBS Tween 20 at 10 min/wash. The membrane was then incubated with 1:500 diluted secondary antibodies conjugated with horseradish peroxidase (HRP) for 1 h and washed as above. Bound antibodies were detected and imaged as described.

sAPP␣ blots
To measure levels of secreted sAPP␣, conditioned medium from HeLa-APP 695WT cells was collected and mixed with 2ϫ reducing sample buffer containing 10% ␤-mercaptoethanol. Samples were boiled at 100°C for 7 min, and 15 l of samples were resolved on 4 -12% NuPAGE TM BisTris SDS-polyacrylamide gel (Invitrogen and ThermoFisher Scientific) at 125 V for ϳ2 h. Proteins were transferred onto a 0.2-m nitrocellulose membrane (Bio-Rad) at 400 mA for 1 h on ice. The membrane was then incubated on warm PBS for 5 min and blocked with 10% w/v skim milk/PBS Tween 20 for 1 h at room temperature. The membrane was incubated overnight at 4°C with W0-2 antibodies diluted in PBS Tween 20 (1:6000) and washed three times with PBS Tween 20 at 10 min/wash. The membrane was then incubated with 1:500 diluted secondary antibodies conjugated with HRP for 1 h and washed as above. Bound antibodies were detected and imaged as described.

Quantitation of colocalization
Quantitation of the colocalization between APP 695WT and endogenous organelle markers in HeLa cells was conducted using an organelle-based colocalization (OBCOL) plugin (66) on FIJI/ImageJ (National Institutes of Health public domain software). Images were thresholded to remove background fluorescence signals, and organelles were segmented into individual structures. The percentage of colocalization of cargo with various organelles was calculated by dividing the total cargo pixels that overlapped with each organelle marker with total cargo pixels within each cell. Volocity TM imaging software (Perkin-Elmer Life Sciences, UK) was used to calculate Manders' coefficient M1 values (67) to determine the percentage colocaliza-␣-Secretase processing of APP in the Golgi tion of the endogenous APP with endogenous organelle markers in primary mouse cortical neurons. All analysis was carried out on the indicated number of cells over three independent experiments.
The generation of graphs and statistical analyses were conducted using Prism (GraphPad Software). Data for fold change of ␣-CTF/C83 levels and sAPP␣ levels were plotted as bar graphs and analyzed by paired, two-tailed, Student's t test. Data from quantitation of colocalization were plotted as dot-plots and analyzed by unpaired, two-tailed, Student's t test. A p value of Ͻ0.05 (*) was considered significant; a p value of Ͻ0.01 (**) was considered highly significant; and a p value of Ͻ0.001 (***) was considered very highly significant.