Analysis of novel endosome-to-Golgi retrieval genes reveals a role for PLD3 in regulating endosomal protein sorting and amyloid precursor protein processing

The processing of amyloid precursor protein (APP) to the neurotoxic pro-aggregatory Aβ peptide is controlled by the mechanisms that govern the trafficking and localisation of APP. We hypothesised that genes involved in endosomal protein sorting could play an important role in regulating APP processing and, therefore, analysed ~ 40 novel endosome-to-Golgi retrieval genes previously identified in a genome-wide siRNA screen. We report that phospholipase D3 (PLD3), a type II membrane protein, functions in endosomal protein sorting and plays an important role in regulating APP processing. PLD3 co-localises with APP in endosomes and loss of PLD3 function results in reduced endosomal tubules, impaired trafficking of several membrane proteins and reduced association of sortilin-like 1 with APP. Electronic supplementary material The online version of this article (10.1007/s00018-018-2752-9) contains supplementary material, which is available to authorized users.


Introduction
The processing of amyloid precursor protein (APP) to form the neurotoxic pro-aggregatory Αβ peptide is believed to be a key initiating event in the pathogenesis of Alzheimer's disease (1). The trafficking and localisation of APP within the post-Golgi endocytic system plays an important role in regulating the exposure of APP to the secretases that mediate its cleavage to form Αβ peptides (2,3). Most evidence now supports a model whereby the cleavage of APP to generate Αβ occurs in an endocytic compartment where β-secretase (BACE) is predominately localised (4). Thus mechanisms that direct APP away from endosomes towards either the Golgi complex or the cell surface are considered to be protective of APP processing to Αβ (5,6).
The retromer complex is a key mediator of endosomal protein sorting and has been shown to operate in both endosome-to-Golgi and endosome-to-cell surface traffic, regulating the localisation of membrane proteins such as the cation independent mannose 6-phosphate receptor (CIMPR), sortilin and Glut-1 (7). The retromer complex comprises a stable trimeric protein complex containing Vps35, Vps26 and Vps29 that together select membrane proteins ('cargo') for packaging into tubular carriers that form through the action of the other functional unit of retromer, the sorting nexin (Snx) dimer. For endosome-to-Golgi traffic, the Snx dimer contains one copy of either Snx1 or Snx2 paired with either Snx5 or Snx6 (8,9).
Another notable cargo protein for retromer is the membrane protein SorL1 (also known as SorLA) that traffics from endosomes to the Golgi in a retromer-dependent manner. SorL1 can directly associate with APP and can bind to the retromer cargo-selective trimer through an interaction with Vps26. SorL1 can thereby direct APP into the retromer-mediated endosome-to-Golgi retrieval pathway thus protecting APP from cleavage by BACE (10-12).
Mutations in SorL1 can cause late-onset AD (LOAD) (13,14) and variants of genes that regulate recruitment of the retromer complex to endosomes have been shown to predispose to LOAD (15). Furthermore, a point mutation in the retromer protein, Vps35, that results in the protein becoming unstable, may be causal in early onset AD (16). Also, it has been shown that expression levels of retromer proteins are reduced in the brains of AD patients and that loss of retromer function results in increased processing of APP to Αβ (17,18).
Due to its key function in endosomal protein sorting and prominent role in regulating APP trafficking, there has been considerable interest in retromer as a potential engine of pathogenesis for AD (19,20), but retromer does not operate in isolation in endosomal protein sorting. We have recently reported the results of a genome-wide siRNA screen for novel endosome-to-Golgi retrieval genes that may function alongside retromer (21). Among the genes identified as new players in the endosome-to-Golgi pathway was a surprising number of multi-pass membrane spanning proteins including SFT2D2 and ZDHHC5. These two proteins are required for the endosome-to-Golgi retrieval of the CIMPR and both SFT2D2 and ZDHHC5 are localised to endosomes positive for retromer proteins. We hypothesised that any of the genes identified as novel endosome-to-Golgi retrieval genes may encode proteins that could function in endosomal protein sorting and may therefore regulate APP localisation and processing. We have undertaken an analysis of the endosometo-Golgi retrieval genes and identified those genes that, when silenced, result in increased processing of APP to Αβ. We report that among the hits, the PLD3 gene exerts a pronounced effect on Αβ secretion. Furthermore, we show that PLD3 localises to retromer-positive and APP-positive endosomes and regulates the localisation of SorL1 and its association with APP.

Cloning
The full-length PLD3 (WT) open reading frame (ORF) of 490 amino acids was amplified by polymerase chain reaction (PCR) using primers to introduce Bam HI and Sal I restriction enzyme sites to the 5' and 3' ends, respectively. All PCR products were first sub cloned using the PCR blunt vector (Invitrogen) and sequenced. The digested ORFs were then subcloned into the Bgl II and Sal I sites of the pEGFP-N3 vector (CLONTECH) for expression as a GFP-fusion protein in mammalian cells.

Western blotting
Cells were harvested with a sterile cell scraper and lysed in lysis buffer (20 mM HEPES-KOH, pH 7.2, 50 mM potassium acetate, 2 mM EDTA, 200 mM sorbitol, 1% Triton X-100, 0.1% SDS) containing Halt™ Protease Inhibitor Cocktail (Thermo Scientific). Cell debris was removed by centrifugation at 20000 x g, 4°C for 10 min. Supernatants were transferred to fresh microfuge tubes, and to an aliquot of the lysate appropriate volumes of 4x NuPAGE LDS sample buffer (Life Technologies) containing 50 mM DTT was added and heated to either 95°C for 5 minutes or 70°C for 10 minutes. Samples were resolved using NuPAGE Bis-Tris Novex 4-12% gels (Life Technologies) and electroblotted to a 0.2 µm PVDF membrane using the Transblot Turbo Transfer System (Bio-Rad). Membranes were blocked with 5% milk TBS-Tween 20 before incubation with primary antibodies overnight at 4°C.
Membranes were then probed with appropriate secondary antibodies conjugated with HRP for 1h. Membranes were washed repeatedly in TBS-0.1% Tween-20 after both primary and secondary antibody incubation. Blots were incubated with Pierce Super Signal or Millipore Immobilon enhanced chemiluminescence reagents for 5 min and exposed to X-ray film and developed or visualised using a ChemiDoc system (Bio-Rad).

Aβ Detection
For Aβ detection, appropriate volumes of 4x LDS sample buffer with 50mM DTT were added to conditioned cell culture media that had been spun for 2 minutes at 2000 rpm and then heated at 95°C for 5 minutes. Samples were then processed as described above. For detecting CTFβ and Aβ, the PVDF membrane was boiled post-transfer in pre-warmed PBS for 5 minutes prior to blocking. Membranes were processed as described above.  Anti-PACSIN2 (ab37615, Abcam)
Cells were counted and scored for the presence of tubules by visual inspection.

Structured Illumination Microscopy (SIM)
Cells grown on 18 mm square high-performance coverslip (no. 1.5, Zeiss) for 24 hours were washed with PBS before fixing in 4% formaldehyde in PBS at 37°C for 15 minutes.
Subsequent permeabilisation, blocking, staining and mounting steps were as for regular immunofluorescence (see above). SIM was performed on a Zeiss Elyra PS1 microscope at 23°C using a 63x 1.4 N.A. plan-apo objective lens (Zeiss) and Immersol 518F (Zeiss) immersion oil. Image stacks were acquired using the Zeiss ZEN Black 2012 software for five grating phases and five grating rotations at z positions spaced 110 nm apart. Channel alignment information was created using a 3D array of multispectral beads imaged with the same instrument settings. Structured illumination processing and channel alignment were performed using the ZEN Black ELYRA edition software. The presented data are a region of a single slice out of a z-stack.

High content imaging
Cells plated in 24-well plates were fixed and stained as for immunofluorescence. In a final staining step cells were labelled with a whole cell stain (Whole Cell Stain Blue, Thermo Fisher). Images were acquired on a CellInsight CX7 automated microscope and analysed using the HCS studio software and its spot detector bio-application. At least 400 cells ("objects" defined by the whole cell stain) were imaged per well. The whole cell stain, Alexa Fluor® 488, 555, or 647 images were acquired sequentially using a single multi-pass filter set. A 20x objective lens was used.

Results
To identify genes that have a role in APP processing we selected ~40 high confidence validated hits from the genome-wide siRNA screen [21] and tested each for a role in regulating APP processing to Αβ by silencing the expression of the gene-of-interest by siRNA. As positive controls we also silenced expression of Vps35 and Snx27 individually as both have been shown to cause increased Αβ production when silenced (18,22). The level of the Αβ peptide secreted into the media was assessed by Western blotting. Cell lysates were also generated and intracellular proteins (e.g. Snx27) detected by Western blotting. In increased Aβ similar to the SmartPool siRNA (see supplemental figure S1).
If PLD3 is influencing the processing of APP it would be predicted to localise to compartments that APP may traffic through, e.g. endosomes or the Golgi complex. In order to investigate the localisation of PLD3, a GFP tag was added to the C-terminus of PLD3 and HeLa cells were transfected with the PLD3-GFP construct. PLD3 is a type II membrane protein and therefore the GFP moiety will be on the lumenal side of the membrane. The PLD3-GFP construct was found to colocalise with a similar construct tagged with mcherry that has been reported previously by Sleat et al., (23). The PLD3-GFP construct also fractionated similarly to endogenous PLD3 when analysed by sucrose density gradient fractionation indicating that it traffics in a manner similar to the native PLD3 protein (see In previous studies we have observed that conditions that perturbed endosomal protein sorting often lead to changes in the numbers of Snx1-positive tubules. For example, knockdown of the WASH complex results in increased Snx1 tubules but loss of EHD1   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64 expression has the opposite effect. (24,25). Whilst we were examining PLD3 knockdown cells for changes in membrane protein localisation, we noticed that there appeared to be fewer tubular structures positive for Snx1. In figure 4a, control and PLD3 knockdown HeLa cells have been labelled with antibodies against Snx1. Tubules are evident in the control cells but generally absent in the PLD3 knockdown cells. We observed a similar loss of tubules positive for MICALL1, a protein that functions with EHD1, Pacsin2 and Snap29 in mediating traffic from recycling endosomes to the cell surface (26,27) (see figure 4b and supplemental figure S4 for images of Pacsin2-and Snap29-positive tubules). When tubule numbers were determined, the knockdown of PLD3 caused a ~40% reduction in Snx1 tubules whereas Snx1-tubules increased in cells stably expressing the PLD3-GFP construct.
The loss of MICALL1 tubules after PLD3 knockdown was more pronounced (figure 4d) but it should be noted that Western blotting of cell lysates revealed a marked loss of MICALL1 protein and a reduction in the levels of proteins associated with MICALL1 including EHD1, Pacsin2 and Snap29 (see figure 4e). We confirmed that the loss of MICALL1 tubules after knockdown of PLD3 was not an off-target effect by analysing MICALL1 tubules in cells where single siRNA oligos were used to silence PLD3 expression (see supplemental figure   S5).
The reduction in Snx1-postive tubules observed after loss of PLD3 expression, and the reduction in other endosomal trafficking machinery such as MICALL1 would be expected to affect many proteins that traffic through endosomes. Given that PLD3 knockdown elicited a marked increase in processing of APP to Αβ, we hypothesised that key proteins that govern APP localisation and/or processing would be similarly affected. The SorL1 protein associates with APP to regulate its localisation and processing (11-13). We therefore investigated whether SorL1 could associate with APP after PLD3 knockdown. In Figure 5a, cell lysates were treated with anti-APP antisera to recover APP and associated proteins. We observed a pronounced reduction in the amount of SorL1 associated with APP in PLD3 knockdown lysates. We next investigated the localisation of SorL1 but due to limitations of the anti-SorL1 antisera we could not determine SorL1 localisation by fluorescence microscopy. Therefore we examined the subcellular distribution of SorL1 by sucrose density gradient fractionation in control and PLD3 knockdown lysates. In figure 5b, after knockdown of PLD3, the SorL1 protein is shifted on sucrose density gradients being predominately present in lighter fractions (e.g. fractions 6 and 7) whereas SorL1 is generally detected in denser fractions in lysates from control cells. The distribution of the CIMPR is also altered but the transferrin receptor (TFRC) is not observably different.

Discussion
Here we report that PLD3, a type-II transmembrane protein, is an important regulator of endosomal protein sorting and loss of PLD3 function results in increased processing of amyloid precursor protein (APP) to Αβ -possibly as a consequence of the mistrafficking of SorL1. PLD3 is a member of the phospholipase D family and therefore predicted to function in the conversion of phosphatidyl choline (PC) to phosphatidic acid (PA) (28) but has yet to be formally shown to possess this activity.
Our studies of endosome-to-Golgi retrieval revealed PLD3 to be required for the efficient retrieval of a CD8-CIMPR reporter protein -PLD3 is one of ~40 high confidence hits from a genome-wide siRNA screen for novel endosome-to-Golgi retrieval genes (21). As endosomal protein sorting has been intimately linked with regulating APP localisation and processing, we hypothesised that any of the novel endosome-to-Golgi retrieval genes could be important regulators of APP processing. We therefore tested the ~40 high confidence hits for a role in controlling APP processing and found that loss of PLD3 markedly increased APP processing to Αβ, even more than the knockdown of VPS35 or SNX27, both of which have been shown to regulate APP processing (18,22). Interestingly, mutations in PLD3 have been linked to Alzheimer's disease (29,30) although this has become somewhat controversial with subsequent studies refuting the initial report (31)(32)(33)(34). The genome-wide screen we undertook for novel endosome-to-Golgi retrieval genes was an unbiased geneby-gene search for new players in endosomal protein sorting and revealed a role for PLD3 (21). The examination of the ~40 high confidence hits for a role in regulating APP processing that we report here is a similarly unbiased approach.
The localisation of PLD3 was determined to be endosomal and, at least partially, lysosomal.
The localisation of PLD3 to lysosomes is consistent with other reports describing the localisation of PLD3 (23). We however observed a significant pool of the PLD3 protein in structures that were positive for retromer proteins (i.e. Vps35), and significantly, also positive for APP. Thus it seems likely that PLD3 traffics through the post-Golgi endocytic system and may therefore have wide-ranging and pleiotropic effects on endosomal protein sorting. Indeed loss of PLD3 function did result in changes in levels of several membrane proteins that traffic through endosomes including the transferrin receptor, the CIMPR and Lamp1. The levels of other proteins that operate in endosomal protein sorting also appeared changed such as MICALL1. It has been reported that the coiled-coil domain of 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64 MICALL1 binds to phosphatidic acid (PA) (27) and in doing so provides a key interaction between recycling endosomes and the MICALL1 complex that includes EHD1 and Pacsin2 (also known as Syndapin2).
Thus, the loss of PLD3 may be affecting proteins such as MICALL1 through production of PA. It should be noted however that the predicted catalytic domain of PLD3 is on the lumenal side of the protein and thus if PLD3 is responsible for PA production on the cytoplasmic face of endosomes, a lipid flippase may be required to translocate the PA from the lumenal side to the cytoplasmic side. There have been reports recently from studies in yeast that the Neo1 lipid flippase localises to endosomes and is trafficked by Snx3, a retromer-associated protein (35). Thus it is plausible that PLD3 exerts it effects on endosomal protein sorting through its function as a phospholipase D enzyme. It does not however appear to be essential for the production of lysobisphosphatidic acid (LBPA). The LBPA lipid has been reported to be a marker of late-endosomes and lysosomes and has been linked with multivesicular body formation (36) but we did not detect any significant changes in LBPA levels in cells treated with siRNA to knockdown PLD3 (see supplemental figure S6).