APP interacts with the Vac14 subunit of the PIKfyve complex
Until recently the study of interaction partners of the intracellular domain of transmembrane receptors, particularly their association with coats and other trafficking regulators has been exceedingly difficult. The major contributing factor to this difficulty has been that coats and their regulators rely on membrane attachment as well as the binding to cytoplasmic domains of receptors . In the absence of a membrane context, it can be extremely difficult to detect the binding of coats to receptors.
This limitation has been overcome with the creation of the proteo-liposome system, in which the cytoplasmic domains of transmembrane proteins is covalently coupled to preformed liposomes in a sterically defined manner [35, 37, 42]. This mimics the situation encountered in cells where the cytoplasmic receptor domain is presented with a membrane environment, allowing the efficient recruitment of coats and their regulators [35, 42]. In our recent study we utilised this powerful system for identifying novel interaction partners of the APP intracellular domain (known as AICD) using mass spectrometry (Balklava et al., in press).
Here, we utilised this system to study the interaction between APP and the PIKfyve complex in detail. First we created three C-terminal truncations and one N-terminal deletion of the 47 amino acid long AICD and tested their potential for binding the PIKfyve complex using proteo-liposomes. Full length AICD and an N-terminal deletion removing 10 amino acids (called AICD-Tr.4) are both capable of binding both Vac14 and PIKfyve from pig brain cytosol (Fig. 1a–d). In contrast, any C-terminal truncation removing 37, 26 or 7 amino acids from AICD (AICD-Tr.1, 2 and 3, respectively), abolished binding of both Vac14 and PIKfyve. To explore the nature of the binding site in more detail, we turned our attention to the highly conserved YENPTY motif next to the deletion introduced in AICD-Tr.3. This motif is conserved in evolution from humans down to very simple metazoans such as the cnidarian Nematostella, arguing for an important and conserved function of this motif. We created three ‘double point’ mutations, exchanging two amino acids in full length AICD for two alanines, yielding the mutants: AICD-AANPTY, AICD-YEAATY, and AICD-YENPAA. We tested their binding capacity and found that any of the three mutations either abolished or strongly reduced binding of Vac14 and PIKfyve in proteo-liposome recruitments, demonstrating that this motif in addition to sequence elements identified using AICD-Tr.3 are necessary for APP binding of the PIKfyve complex (Fig. 1e–g). These data suggested that the PIKfyve complex binding site is located in close proximity to the C-terminus of AICD and spans the conserved YENPTY motif.
Next we tested the binding capability of bacterially expressed and purified Vac14 and found that it bound AICD on proteo-liposomes (Fig. 1h, i). We also tested the binding of purified Vac14 using classical pull-downs. We found that both AICD and Tr.4 were able to bring down Vac14, while Tr. 1–3 and the negative control MBP failed to bind Vac14, fully recapitulating the results obtained using proteo-liposome recruitments (Fig. 1j). These data suggest that APP forms a direct protein–protein interaction with the Vac14 subunit of the PIKfyve complex.
As the YENPTY sequence motif is also found in the APP related genes, APLP1 and APLP2 (alignment shown in Fig. 1k) we tested their ability to bind Vac14. We found that both APLP1 and APLP2 were able to bind Vac14 and PIKfyve (Fig. 1l–n), suggesting that all members of the mammalian APP gene family are likely to be PIKfyve complex interactors.
APP or AICD overexpression stimulates the formation of ML1Nx2 positive vesicles
What is the functional significance of this interaction? Most previous work has focused on the roles that APP or the APP gene family may play in neurons. However, APP and APLP2 are expressed ubiquitously, while APLP1 expression appears to be restricted to neurons . This clearly suggests that the APP gene family is likely to have functions not limited to the brain. The PIKfyve complex is also widely expressed . To test for a generic APP function we choose the epithelium derived HeLa cell line, in which the function of PIKfyve has been thoroughly characterised [16, 20].
First we asked whether overexpression of APP or APP’s intracellular domain may alter production of PI(3,5)P2. To test this, we utilised the recently established PI(3,5)P2 specific probe ML1Nx2 . Fusion of a tandem repeat of this lipid binding domain of the PI(3,5)P2 binder TRPML-1 to a fluorescent protein allows the detection of PI(3,5)P2 in a spatially and temporally defined manner which had not been previously possible . This probe has also been used successfully in a study that established that PIKfyve function is required for AMPA receptor trafficking and synaptic depression, further validating it as a tool for analysing PI(3,5)P2 dynamics in vivo . mCherry-ML1Nx2 was transfected into cells and analysed by confocal microscopy. Structures labelled by ML1Nx2 were analysed using the MosaicSuite segmentation tool of ImageJ, allowing unbiased, automated detection of ML1Nx2 positive structures and their intensity.
Overexpression of APP-GFP led to a strong increase in the number of ML1Nx2 positive vesicles per cell compared to the expression of GFP as a negative control (Fig. 2a, c). By contrast, overexpression of APP lacking the intracellular domain (APPΔAICD) did not significantly alter the number of ML1Nx2 positive vesicles, suggesting that the intracellular domain of APP is required for stimulating the formation of ML1Nx2 positive vesicles (Fig. 2a, c). To confirm that the increase of ML1Nx2 positive vesicles upon overexpression of APP truly depends on PIKfyve activity we combined APP overexpression with pharmacological PIKfyve inhibition using YM201636 . Upon PIKfyve inhibition the number of ML1Nx2 vesicles was drastically reduced, demonstrating that the APP-induced increase of ML1Nx2 positive vesicles is indeed dependent on PIKfyve activity (Fig. 2b, c). This is fully in line with the recent report of the Weisman lab that PIKfyve-dependent phosphorylation of PI(3)P is the only source for the production of PI(3,5)P2 in mammals . Next we asked whether expression of the intracellular domain of APP, AICD, was also capable of stimulating the formation of ML1Nx2 positive vesicles. AICD, when released from APP by gamma-secretase cleavage, is a soluble, cytosolic molecule that loses its membrane attachment. We found that AICD-GFP expression strongly stimulated the formation of ML1Nx2 positive vesicles, mimicking the effect observed upon APP overexpression (Fig. 2a, c). When testing the truncation mutants we found that AICD-Tr.4 was able to stimulate the formation of ML1Nx2 positive vesicles to a similar extent as full length AICD (Fig. 2a, c), while AICD-Tr.2 and AICD-Tr.3 failed to do so (Online Resource 1). It is interesting to note that both AICD and AICD-Tr.4 are still able to stimulate formation of ML1Nx2 positive vesicles, despite neither of them being membrane attached by a transmembrane domain. Binding of the PIKfyve complex (as established in Fig. 1a–d) seems to be sufficient for the ability to provoke an increase in ML1Nx2 positive vesicles.
In summary, in all AICD mutants characterised so far, the ability to bind Vac14 perfectly correlated with their ability to stimulate the formation of ML1Nx2 positive vesicles upon overexpression. These data suggest that APP requires Vac14 binding to stimulate PIKfyve-dependent formation of ML1Nx2 positive vesicles.
To explore the nature of the ML1Nx2 vesicles induced by overexpression of APP we performed triple labelling using APP-GFP, mCherry-ML1Nx2 and staining for EEA1 or LampI. The large majority of APP/ML1Nx2 positive structures were also positive for EEA1, suggesting that these vesicles are early endosomal in nature (Online Resource 2).
We also wanted to test whether APP or AICD overexpression has the same effect in a neuronal cell line. APP-GFP or AICD-GFP were co-expressed together with mCherry-ML1Nx2 in the SH-SY5Y neuroblastoma line and compared to a GFP control (Online Resource 3). As in HeLa cells both APP and AICD expression increased the number of ML1Nx2 positive structures. Interestingly, in SH-SY5Y cells APP seemed more effective than AICD at increasing the number of ML1Nx2 vesicles.
These data show that in both neuronal and non-neuronal cell lines overexpression of APP or AICD increased the number of ML1Nx2 positive vesicles.
Suppression of APP and/or APLP2 compromises PIKfyve-dependent processes
Are APP gene family members required for PIKfyve activity? We tested this question using RNAi mediated suppression of APP and the paralogue APLP2. Effective suppression of APP, APLP2 or double suppression was achieved with two different siRNA duplexes per gene (Fig. 3a, b). Automated detection of ML1Nx2 positive structures was used to test whether ML1Nx2 positive vesicles are affected by APP gene family suppression. Both single and double suppressions of APP and APLP2 led to a reduction of the number of ML1Nx2 positive vesicles (Fig. 3c). This showed that APP and APLP2 are required for PI(3,5)P2 vesicle formation, not entirely surprising given the high similarity of their intracellular domains and their potential to interact with the PIKfyve complex.
It is well established that loss of PIKfyve activity induced by expression of a kinase-dead version of PIKfyve, RNAi suppression, mutation of PIKfyve complex genes or pharmacological PIKfyve inhibition leads to a dramatic and consistent accumulation of vacuoles in cells, aberrant structures that were previously shown to be derived from the endosomal system [16, 17, 20, 46]. We analysed whether RNAi suppression of APP and/or APLP2 led to such vacuoles. Suppression of APP and/or APLP2 led to an increase of vacuole incidence compared to a control siRNA (Fig. 3a). It is worthwhile noting that occurrence of vacuolation is less pronounced with APP gene family knock-down than with PIKfyve RNAi , suggesting that APP family genes play an auxiliary rather than an essential role in preventing vacuolation.
This idea would suggest that suppression of APP and APLP2 will sensitise cells for PIKfyve inhibition. We tested this by inhibiting PIKfyve for a short period of time (45 min). In control RNAi treated cells this brief PIKfyve inhibition led to the formation of a number of small vacuoles. However, single suppression of APLP2 using RNAi duplex I or double suppression of APP and APLP2 significantly increased the incidence of vacuoles when PIKfyve was briefly inhibited (Fig. 3d, e). APP or APLP2 RNAi duplex II had no significant effect on vacuolation. These data show that double suppression of APP and APLP2 sensitises cells for reduced PIKfyve activity, leading to increased vacuolation. In this assay single suppressions had no statistically significant effect (APP duplexes) or variable effects in the case of APLP2.
APP requires PIKfyve activity for its trafficking
What is the purpose of APP binding to PIKfyve and stimulating the formation of ML1Nx2 positive vesicles? We noticed that upon PIKfyve inhibition using YM201636 a marked redistribution of APP-GFP occurred; instead of localising to small vesicles throughout the cell, APP became trapped in large, vacuolar structures (Fig. 2b). We analysed APP-GFP localisation in more detail by inhibiting PIKfyve using YM201636 and the recently established Apilimod  in HeLa cells. APP-GFP accumulation in vacuoles was dose dependent for both inhibitors with mild effects observed with as little as 3 nM Apilimod (Online Resource 4) and 100 nM YM201636 (Online Resource 5) for 4 h. APP-GFP accumulation in vacuoles became progressively worse over a 4 h time course with both inhibitors (1 µM YM201636 or 30 nM Apilimod) (Online Resources 6 and 7). Next we analysed in what organelles APP-GFP accumulated. We analysed the early endosomal marker EEA1, late endosomal and lysosomal marker LampI as well as the Golgi marker GM130. In control cells APP could be detected in EEA1, LampI and GM130 compartments, consistent with its established trafficking pattern [29, 48]. However, when PIKfyve was inhibited using Apilimod we found a marked redistribution, with APP accumulating in EEA1 positive structures and the pools of APP in LampI positive late endosomes/lysosomes and the Golgi apparatus diminished (Fig. 4a, b).
These data suggested that APP became trapped in early endosomes or early endosome-derived vesicles upon PIKfyve inhibition with markedly reduced transport to both late endosomes/lysosomes and the Golgi apparatus.
We also studied the protein levels of endogenous APP and APLP2 to test whether they are affected by PIKfyve inhibition. While APLP2 appeared largely unchanged, the overall levels of APP increased strongly and its band pattern in Western blot was altered (Fig. 4c, d). Particularly the 135 kDa form strongly increased in quantity while the approx. 110 kDa form remained largely unchanged. It has previously been shown that human APP expressed from cDNA results in a 135 and 110 kDa form, most likely differing in their degree of glycosylation . Taken together these data show that APP trafficking and APP levels depend on PIKfyve activity.
PIKfyve has two well-established roles in endosomal function: Mediating endosome-to-TGN transport and facilitating endosome/lysosome fusion [16, 19]. APP traffics between the Golgi, plasma membrane and endosomes. APP has previously been shown to undergo retromer-mediated endosome-to-TGN transport . Our data show that APP requires PIKfyve to avoid getting ‘stuck’ in early endosomal-derived vacuoles which is fully compatible with the important role that PIKfyve plays for the sorting of receptors in endosomes .
Is PIKfyve function also required in neuronal cells for APP trafficking? We tested this question by studying APP trafficking in SH-SY5Y cells. As in HeLa cells, inhibition of PIKfyve using Apilimod led to strong vacuolation and APP-GFP trapping in vacuoles (Online Resource 8), suggesting that PIKfyve is also required in neuronal cells for APP trafficking.
It is conceivable that the endosomal population of APP, by binding to and stimulating PIKfyve can drive local production of PI(3,5)P2 and formation of carriers (as suggested by the ML1Nx2 probe in Fig. 2) that may allow APP sorting and escape from endosomes. If this is the case we would expect a fraction of the APP label to co-localise and co-migrate with ML1Nx2. We tested this by co-expressing APP-GFP and mCherry-ML1Nx2 and analysed their behaviour in live cell imaging (Fig. 5a, b). APP-GFP, when expressed at low levels, displayed extensive co-localisation and co-migration with mCherry-ML1Nx2 (Online Resource 9), consistent with the idea that PI(3,5)P2 production is required for APP trafficking as suggested by our PIKfyve inhibition experiments.
The analysis of the interaction interface between APP and Vac14 highlighted the significance of the YENPTY motif and the C-terminal, adjacent sequence (eliminated in AICD-Tr.3). By consequence deletion of this sequence is expected to disrupt PIKfyve-dependent APP sorting. However, an important caveat is that the YENPTY motif contains a classical NPxY endocytosis motif. Additionally the C-terminal tyrosine of the YENPTY motif is part of the YKFFE AP-4 binding motif required for APP exit from the Golgi . We created the APP-Tr.3-GFP deletion construct in which the FFEQMQN motif was deleted and analysed its trafficking. Entirely consistent with work by  APP-Tr.3-GFP accumulated in the Golgi (Online Resource 10), virtually eliminating endosomal localisation, precluding a closer analysis of PIKfyve-dependent, endosomal sorting.