Phosphatidylinositol (3,5)-bisphosphate machinery regulates neurite thickness through neuron-specific endosomal protein NSG1/NEEP21

Phosphatidylinositol (3,5)-bisphosphate [PtdIns(3,5)P2] is a critical signaling phospholipid involved in endolysosome homeostasis. It is synthesized by a protein complex composed of PIKfyve, Vac14, and Fig4. Defects in PtdIns(3,5)P2 synthesis underlie a number of human neurological disorders, including Charcot-Marie-Tooth disease, child onset progressive dystonia, and others. However, neuron-specific functions of PtdIns(3,5)P2 remain less understood. Here, we show that PtdIns(3,5)P2 pathway is required to maintain neurite thickness. Suppression of PIKfyve activities using either pharmacological inhibitors or RNA silencing resulted in decreased neurite thickness. We further find that the regulation of neurite thickness by PtdIns(3,5)P2 is mediated by NSG1/NEEP21, a neuron-specific endosomal protein. Knockdown of NSG1 expression also led to thinner neurites. mCherry-tagged NSG1 colocalized and interacted with proteins in the PtdIns(3,5)P2 machinery. Perturbation of PtdIns(3,5)P2 dynamics by overexpressing Fig4 or a PtdIns(3,5)P2-binding domain resulted in mislocalization of NSG1 to nonendosomal locations, and suppressing PtdIns(3,5)P2 synthesis resulted in an accumulation of NSG1 in EEA1-positive early endosomes. Importantly, overexpression of NSG1 rescued neurite thinning in PtdIns(3,5)P2-deficient CAD neurons and primary cortical neurons. Our study uncovered the role of PtdIns(3,5)P2 in the morphogenesis of neurons, which revealed a novel aspect of the pathogenesis of PtdIns(3,5)P2-related neuropathies. We also identified NSG1 as an important downstream protein of PtdIns(3,5)P2, which may provide a novel therapeutic target in neurological diseases.

Neurite thickness is an essential determinant of neuronal function. It has been well established that conduction velocity in axons is proportional to fiber diameter in myelinated axons and the square root of diameter in unmyelinated axons (41,42). In recent years, it was found that dendrite diameter also plays an important role in excitability (43,44), action potential amplitude (45), and voltage propagation (46)(47)(48). In the physiological setting, dendrite diameters of central auditory neurons are correlated with their frequency response (49). In the pathological setting, exposure of hippocampal neurons to amyloid Aβ results in a decrease in dendrite diameters (50), which could contribute to defective information processing in Alzheimer's disease. However, the regulation of neurite thickness remains little understood.
In this study, we found that PtdIns(3,5)P 2 regulates neurite thickness through NSG1/NEEP21. Suppressing either PtdIns(3,5)P 2 pathway or NSG1 expression resulted in thinner neurites. Proteins in the PtdIns(3,5)P 2 machinery colocalized and physically interacted with NSG1. Perturbation of PtdIns(3,5)P 2 dynamics led to the mislocalization of NSG1. Importantly, overexpression of NSG1 rescued the neurite A B C D E F Figure 1. Inhibition of PtdIns(3,5)P 2 biosynthesis leads to neurite thinning in primary neurons. A, diagram of PtdIns(3,5)P 2 biosynthesis and turnover pathways. B and C, cultured mouse hippocampal neurons were treated with DMSO (Control) or PIKfyve inhibitor YM201636 at 0.8 μM or 1.6 μM for 48 h from 1 div to 3 div after plating (B) or at 1.6 μM from 7 div to 9 div (C). Top: Phase contrast microscopy. Bottom: Immunofluorescence with dendritic marker anti-Map2 (green) and axon marker anti-Tau (red thickness defect in neurons defective of PtdIns(3,5)P 2 synthesis. This study revealed a novel role of PtdIns(3,5)P 2 in neuron morphology and provided a novel target pathway for treating PtdIns(3,5)P 2 -related neuropathies.

Results
Defective PtdIns(3,5)P 2 biosynthesis results in decreased neurite diameters In previous studies, we noticed that neurites of PtdIns(3,5) P 2 -impaired neurons such as Vac14 −/− neurons were often thinner than WT controls (28), suggesting that PtdIns(3,5)P 2 is possibly involved in regulating neurite thickness. To examine this possibility, we investigated the role of PtdIns(3,5)P 2 in neuron development by inhibiting PIKfyve. PIKfyve is the only lipid kinase that synthesizes PtdIns(3,5)P 2 , so PIKfyve inhibitors are known to abolish PtdIns(3,5)P 2 almost completely (29). Mouse hippocampal neurons were cultured in vitro and treated with a selective PIKfyve inhibitor YM201636 (51). Consistent with previous observations (30), swollen vacuoles were observed in both cell bodies (soma) and neurites in YM201636-treated neurons (Fig. 1, B and C). Such vacuolation is a hallmark phenotype of PtdIns(3,5)P 2 -deficient cells (52). When neurons were exposed to YM201636 during early development, from 0 days in vitro (div) to 3 div after plating, neurite outgrowth was impaired, forming shorter and thinner neuronal processes (Fig. 1, B and D). Meanwhile, axon specification, an event during early development characterized by the selection of only one of the neurites to become the axon (53), was not affected by YM201636. The axon marker, Tau, was still restricted to only one neurite in YM201636-treated neurons at 3 div, indicating that PtdIns(3,5)P 2 was not involved in axon/dendrite polarization (Fig. 1, B and E). The effect of PtdIns(3,5)P 2 abolishment was also examined during late neuron development (7 div to 9 div), when neurites were already formed. In YM201636-treated older cultures, dendrites labeled by Map2 were noticeably thinner, and axon networks marked by Tau were much sparser ( Fig. 1, C and F). Taken together, these results suggest that PtdIns(3,5)P 2 is involved in regulating and maintaining neurite morphology, particularly neurite thickness, but not axon/dendrite polarization.
It was reported that inhibition of PIKfyve could result in cell death in primary neuron cultures (54). To exclude the effect of neuronal viability on neurite morphology, we made use of Cath.a-differentiated cells (CAD) cells, a neuron cell line derived from catecholaminergic neurons in the central nervous system (55). Neurite growth in CAD cells can be conveniently initiated by serum withdrawal from culturing media, which provides a unique system of "on-demand" neurite growth. A fluorescent protein, Citrine, was transfected into CAD cells to serve as a volume marker. When treated by YM201636, or another structurally distinct PIKfyve inhibitor apilimod (56), CAD neurites were significantly thinner than the controls (Fig. 2, A and B), consistent with the results in hippocampal neurons (Fig. 1, D and F). Moreover, to exclude possible side effects of chemical inhibitors, PIKfyve expression was specifically stably knocked down with lentivirus-mediated shRNA expression (shPIKfyve) in CAD cells (Fig. 2C). After differentiation, stable shPIKfyve cells actively developed much thinner neurites than control cells with empty vector infection (shNC) (Fig. 2, D and E). Note that neurite growth occurred after PIKfyve knockdown, excluding cell viability as the explanation for the thinner neurites. Collectively, these results demonstrate that PIKfyve inhibition resulted in thinner neurites in both primary hippocampal neurons and CAD neurons, suggesting that regulation of neurite thickness by PtdIns(3,5)P 2 is a common mechanism in multiple neuronal types.
NSG1 colocalizes and physically interacts with PtdIns(3,5)P 2 machinery NSG1/NEEP21 is a neuron-specific membrane protein involved in the trafficking of a number of physiologically critical cargoes. We noticed that there is a significant overlap in phenotypes regulated by PtdIns(3,5)P 2 or NSG1, including AMPA receptor-mediated synaptic plasticity (37) and APP processing (38). Thus, we set out to determine the relationship between NSG1 and PtdIns(3,5)P 2 in regulating neurite thickness. Firstly, NSG1 expression was knocked down in CAD cells by three independent shRNA constructs. Neurite thinning was observed in all of them (Fig. 3, A and B), similar to shPIKfyve cells. This phenotype could be rescued when an mCherrytagged NSG1 (mCherry-NSG1) was reintroduced to the NSG1 knockdown cells (Fig. S1, A-C). This result suggests that both PtdIns(3,5)P 2 and NSG1 regulate neurite thickness.
The localization of NSG1 was determined with mCherry-NSG1 in CAD cells. Consistent with previous reports (35), mCherry-NSG1 localized to punctate endosomal structures (Video S1 and Fig. 3C). Notably, mCherry-NSG1 puncta exhibited active long-range and bidirectional movements along the neurites, as well as between neurites and soma (Video S1). At the steady state, the distribution of mCherry-NSG1 between soma and neurites varied among individual cells, ranging from mainly soma-localized (Fig. 3, C1) to almost exclusively neurite-localized (Fig. 3, C4), with many intermediate states in between. Similar patterns were also observed in PC12 cells, another commonly used neuronal cell line (Fig. S1D). These results suggest an active transport route for NSG1-positive vesicles between soma and neurites.
To determine the spatial relationship between NSG1 and PtdIns(3,5)P 2 machinery, mCherry-NSG1 was coexpressed with Citrine-tagged Vac14 (Cit-Vac14). Vac14 is an essential component in the PtdIns(3,5)P 2 machinery that forms a starshaped pentameric scaffold and nucleates PIKfyve and Fig4 in the same assembly (4). A significant fraction of punctate Vac14 (57.8 ± 0.07%, n = 9) was found colocalizing with mCherry-NSG1 (Fig. 3D), indicating that both proteins were on the same endosomes. Next, we determined if they physically interact. HEK293T cells were cotransfected with 3×FLAG-tagged NSG1 and Cit-Vac14 or Citrine control. NSG1 was pulled down with anti-FLAG resin. Cit-Vac14, but not Citrine, was found in the precipitates (Fig. 3E), suggesting that NSG1 and Vac14 were in the same physical complex.
It has been reported that NSG1 is a type II transmembrane protein that travels rapidly from Golgi to the plasma membrane, followed by internalization into the endocytosis pathway (58). We assessed whether mCherry-NSG1 functions with PtdIns(3,5)P 2 at specific endocytic stages. Three endocytic markers, EEA1 (early endosomes), CD63 (late endosomes), and LAMP1 (late endosomes/lysosomes) were used to mark various endosomes (Fig. S2). Consistent with the previous report (58), mCherry-NSG1 could be found throughout the endocytic pathway, with the highest colocalization with late endosomes (Fig. S2, B and D). Furthermore, mCherry-NSG1 endosomes that colocalized with Cit-Vac14 or TRPML1 had a similar distribution of endocytic markers as those of overall mCherry-NSG1 puncta (Fig. S2, B and D), suggesting that the colocalization was not restricted to a particular endocytic stage.

Perturbation of PtdIns(3,5)P 2 dynamics mislocalizes NSG1
We initially hypothesized that Fig4, the phosphatase within the PtdIns(3,5)P 2 biosynthesis complex, also colocalized with NSG1. Unexpectedly, the punctate localization of mCherry-NSG1 proteins was largely lost in Cit-Fig4-expressing cells (Fig. 4, A and E). When Cit-Fig4 was overexpressed, mCherry-NSG1 proteins were either concentrated in the perinuclear region, on the cell surface, or appeared diffused (Fig. 4, A and E), suggesting that mCherry-NSG1 proteins were trapped in the endoplasmic reticulum/Golgi step or at the plasma membrane before internalization into endosomes. In Cit-Fig4overexpressed cells, neurites were often ill-formed and developed curly spiny branches instead of a stable shaft ( (Compare arrowheads with arrows in Fig. S3A). We also repeated this experiment in PC12 cells and observed similar mislocalization patterns of mCherry-NSG1 upon Cit-Fig4 overexpression (Fig. S3B). Fig4 is a unique member in the PIKfyve/Vac14/Fig4 complex since it harbors phosphatase activity and could dephosphorylate PtdIns(3,5)P 2 (59,60). In addition, Fig4 also acts as a protein phosphatase and stimulates the kinase activity of PIKfyve (Fig. 1A) (4). Therefore, Fig4 functions in both biosynthesis and turnover of PtdIns(3,5)P 2 (4,61,62). It was possible that overexpression of Fig4 led to reduced PtdIns(3,5) P 2 levels and hence NSG1 mislocalization. To explore this possibility, we employed a phosphatase-dead Fig4 C486S mutant (62), where the catalytic cysteine was mutated to serine. Nonetheless, when the Cit-Fig4 C486S mutant was coexpressed with mCherry-NSG1, mislocalization of mCherry-NSG1 was still observed in both CAD cells (Fig. 4, B and E) and PC12 cells (Fig. S3C), albeit with somewhat lower efficiency, suggesting that the phosphatase activity of Fig4 was not required for NSG1 mislocalization. Rather, overexpression of Fig4 probably perturbed the assembly stoichiometry of PtdIns(3,5)P 2 machinery and impaired the dynamics of PtdIns(3,5)P 2 , which in turn caused NSG1 mislocalization.

Inhibition of PtdIns(3,5)P 2 biosynthesis traps NSG1 in EEA1positive endosomes
The results above demonstrated that mCherry-NSG1 is sensitive to perturbations in the PtdIns(3,5)P 2 pathway. To further investigate the effect of PtdIns(3,5)P 2 levels on NSG1 localization, we directly inhibited PtdIns(3,5)P 2 synthesis with either YM201636 or apilimod in CAD cells transfected with mCherry-NSG1. NSG1-positive endosomes were found enlarged (Fig. 5, A and B), suggesting defective endosome fission or increased endosome fusion. The enlargement occurred around 30 min to 1 hour after the application of PIKfyve inhibitors (Videos S3 and S4). Similarly, stable shPIKfyve CAD cells transfected with Cherry-NSG1 also showed enlargement of NSG1-positive endosomes (Fig. S6, A-B), although the defects in shPIKfyve neurons seemed milder: mCherry-NSG1 endosomes were mostly absent from YM201636 or apilimod-treated neurites (Fig. 5A) but still present in shPIKfyve neurites (Fig. S6A). This was probably due to residual PIKfyve activity in shPIKfyve cells. These results suggest that PtdIns(3,5)P 2 is necessary for NSG1 trafficking in endosomes.
To further elucidate the trafficking defects of NSG1 in PtdIns(3,5)P 2 -deficient neurons, we colabeled mCherry-NSG1 with an array of endocytic markers: EEA1 (early endosomes), Rab4 (recycling endosomes), Rab5 (early endosomes), Rab9 (transport between endosomes and trans-golgi network) (Fig. 5, C and D). Consistent with previous reports (58), mCherry-NSG1 had limited colocalization with EEA1, Rab4, or Rab5 under normal conditions (Fig. 5, C and D). Interestingly, the colocalization between NSG1 and Rab9 was very significant (Fig. 5D), which was consistent with reports that Rab9 and its effector p40 physically interact with the PIKfyve/ Vac14 complex (64,65). After PIKfyve inhibitor treatment, mCherry-NSG1 endosomes had significantly increased colocalization with EEA1 (Manders' Coefficients from 0.03 to 0.12, also see Videos S5-S7), while the colocalization with Rab5 did not change much (Fig. 5, C and D). A slight increase was observed in the colocalization of NSG1 and Rab4 after PIKfyve inhibitor exposure; however, it was only statistically significant for apilimod but not YM201636 (Fig. 5D). The overlap between NSG1 and Rab9 trended downwards after inhibitor treatment, although this decrease was only significant in YM201636 but not apilimod (Fig. 5D). Collectively, these results suggested that NSG1 transport is delayed in EEA1positive endosomes when PtdIns(3,5)P 2 biosynthesis is impaired. Significance was determined by Kruskal-Wallis test followed by Dunn's multiple comparison test. C, CAD cells transfected with mCherry-NSG1 were treated as in (A). Endocytic vesicles were labeled by cotransfection with GFP-EEA1, GFP-Rab4, GFP-Rab9, or immunostained with anti-Rab5. Scale bars represent 10 μm for the main panels and 2 μm for the zoomed insets. D, quantitation of colocalization between mCherry-NSG1 and endocytic markers in (C). N = 10 to 25 cells for EEA1, 9 to 11 cells for Rab4, 12 to 17 cells for Rab5, and 13 to 17 cells for Rab9, from three independent experiments. Significance was determined by Kruskal-Wallis test followed by Dunn's multiple comparison test. Error bars, mean ± SD. All images were single z-plane. CAD, Cath.a-differentiated cells.
It was intriguing that direct inhibition of PtdIns(3,5)P 2 biosynthesis led to less severe NSG1 mislocalization than overexpression of Fig4 or ML1N×2 (Fig. 4), which implies that the dynamics of PtdIns(3,5)P 2 could be more important than absolute levels of PtdIns(3,5)P 2 per se, at least for NSG1 trafficking. We did notice that, in selected cells with shPIKfyve knockdown (Fig. S6C) or PIKfyve inhibitor treatment (Fig. S7A), mCherry-NSG1 was mislocalized to the perinuclear region or the surface, similar to cells overexpressing of Fig4 or ML1N×2 (Fig. 4). The perinuclear localization of mCherry-NSG1 proteins overlapped with a Golgi marker, GFP-Rab6, consistent with a blockage at the Golgi step (Fig. S7B).

NSG1 rescues neurite thickness phenotype in PtdIns(3,5) P 2 -impaired neurons
In experiments above, we noted that CAD cells overexpressing mCherry-NSG1 generally had thicker neurites than normal CAD cells when exposed to PIKfyve inhibitors (Fig. 2), suggesting that, even though some of the mCherry-NSG1 proteins were trapped in early endosomes in PtdIns(3,5)P 2impaired neurons, additional expression of NSG1 molecules could functionally overcome this traffic delay. To test this notion, we transfected shNC and shPIKfyve CAD cells with mCherry or mCherry-NSG1, together with Citrine as a volume marker. In shPIKfyve cells transfected with mCherry control, neurite diameters were 2.1 μm on average (Fig. 6, A and B). However, with mCherry-NSG1 expression, the average neurite diameter was increased to 3.3 μm (Fig. 6, A and B). In addition, we transfected normal CAD cells with Citrine and mCherry or mCherry-NSG1, followed by treatment with PIKfyve inhibitors. Similarly, mCherry-NSG1 overexpression increased neurite diameters to 126% and 140% in YM201636or apilimod-treated cells, respectively (Fig. 6, C and D). These results demonstrated that increased NSG1 expression could alleviate neurite thickness defects in both PIKfyve shRNA knockdown and PIKfyve inhibitor-treated CAD cells.
Next, the functional link between NSG1 and PtdIns(3,5)P 2 was examined in primary neurons. We utilized mouse neurons from the layer V neocortex, one of the brain regions with the highest NSG1 expression during development (66). Neurites were delineated by antibody to Tuj1 (neuronal-specific class III beta-tubulin). Similar to hippocampal neurons and CAD cells, treatment with PIKfyve inhibitors significantly reduced neurite thickness in cortical neurons from 2.1 μm down to 1.0 μm (Fig. 7, A and B). In addition, endogenous NSG1 vesicles were enlarged by both PIKfyve inhibitors (Fig. 7, C and D). Most importantly, overexpression of mCherry-NSG1 increased the neurite diameter by 30% and 74% in YM201636-or apilimodtreated neurons, respectively (Fig. 7, E and F), in agreement with results in CAD cells. Collectively, these results suggest that NSG1 is a downstream target of PtdIns(3,5)P 2 signaling for maintaining the homeostasis of neurite thickness.

Discussion
Neurons are highly susceptible to disruption in the endosome/lysosome system (67,68). Genetic variants of endosomal proteins such as BIN1 (69), SORL1 (70), CD2AP (71) have been implicated in Alzheimer's disease. Similarly, mutations in Vps35 (72), LRRK2 (73), and RME8 (74) are associated with Parkinson's disease. In recent years, an increasing number of reports have shown that dysregulation of signaling lipid PtdIns(3,5)P 2 underlie a range of human neurological disorders, such as Charcot-Marie-Tooth disease, Yunis-Varón syndrome, familial epilepsy, progressive dystonia, and parkinsonism. In addition, one computational analysis associated a Vac14 variant with Alzheimer's disease and bipolar disorder (75). These findings point to the importance of PtdIns(3,5)P 2 in multiple human neuropathies. Hence, the study of PtdIns(3,5)P 2 in neurons is of great importance to understand the etiology of various neuropathies and identify new therapeutic targets.
In this study, we described the relationship between PtdIns(3,5)P 2 machinery and a neuron-specific endosomal protein NSG1. We found that neurite thickness is regulated and maintained by PtdIns(3,5)P 2 and NSG1. NSG1 colocalized and physically interacted with Vac14, the scaffolding protein for PtdIns(3,5)P 2 biosynthesis. The transport of NSG1 within neurons required proper PtdIns(3,5)P 2 dynamics. Importantly, overexpression of NSG1 rescued neurite thinning in PtdIns(3,5)P 2 -deficient neurons, suggesting that NSG1 is downstream of PtdIns(3,5)P 2 function and that activation of NSG1 function could serve as a new target for PtdIns(3,5)P 2related neuropathy. It remains to be investigated whether NSG1 could also rescue other PtdIns(3,5)P 2 -related neuronal defects such as AMPA-type glutamate receptor trafficking.
The effect of PtdIns(3,5)P 2 on neurite thickness was observed in primary hippocampal neurons, cortical neurons, and CAD cells, suggesting that PtdIns(3,5)P 2 could be a general mechanism for regulating neurite thickness. Furthermore, in addition to PIKfyve knockdown, both knockdown of NSG1 and overexpression of GFP-ML1N×2 also resulted in thinner neurites in CAD cells. The presence of the same phenotype in multiple systems after distinct molecular interventions supports the notion that neurite thickness is physiologically regulated. This neurite thinning phenotype was unlikely due to impaired neuron viability (54). Stable shPIKfyve CAD cells could be passaged and revived from frozen stocks similar to parental cells, suggesting that their viability was unaffected. Exact molecular details for the regulation of neurite thickness by PtdIns(3,5)P 2 are still under investigation. It is likely that cytoskeleton and adhesion proteins are modulated by PtdIns(3,5)P 2 and NSG1. Consistent with this notion, PtdIns(3,5)P 2 was shown to be involved in the recycling of tight junction proteins claudin-1/2 in MDCK cells (76), while NSG1 is important for the trafficking of NgCAM/L1 adhesion molecules (35,39). Further study on this topic will significantly deepen our understanding of neuron morphogenesis.
To date, how NSG1 exerts its function on endosomes still remains unclear, but it is known to be a very dynamic protein.
It has been reported that NSG1 is not a resident protein on endosomes but travels rapidly along the endocytic pathway (58). We showed that treatment with PIKfyve inhibitors delayed NSG1 trafficking in EEA1-positive endosomes. This likely reduced the motility of NSG1. In a previous study, an estimate of 63% of NSG1+/EEA1-vesicles were motile, while only 8% of NSG1+/EEA1+ endosomes were motile (77). This notion is consistent with another study showing that PIKfyve is important for the motility of late endosomes/lysosomes in neurites (78). Interestingly, neurites were often ill-developed in cells with defective NSG1 localization (Fig. 4), suggesting that the transport of NSG1 is linked with neurite development. Three types of perturbation to PtdIns(3,5)P 2 were used in this study (in the order of phenotype severity): overexpression of Fig4 or ML1N×2, pharmacological inhibition of PIKfyve activity, and shRNA knockdown of PIKfyve expression. A serendipitous discovery was that overexpressing Fig4 or ML1N×2 had a more significant influence than PIKfyve inhibitors on mCherry-NSG1 localization. While PIKfyve inhibitors mainly caused enlarged mCherry-NSG1 vesicles, overexpressing Fig4 or ML1N×2 led to mislocalization of mCherry-NSG1 to nonendosomal locations. Accordingly, additional expression of mCherry-NSG1 rescued neurite phenotype in inhibitor-treated CAD cells (Fig. 6), but apparently not in ML1N×2/Fig4-overexpression cells (Fig. 4). The inter-conversion between phosphoinositide species is well known to coordinate membrane trafficking events both spatially and temporally (79). It is conceivable that NSG1 functions in a sequence of events that rely on the correct timing of PtdIns(3,5)P 2 synthesis and turnover. The molecular identity of such events awaits further research. Another example supporting the importance of PtdIns(3,5)P 2 dynamics comes from MTMR2 and Fig4, causal genes for Charcot-Marie-Tooth disease type 4B and 4J, respectively. Loss of either MTMR2 or Fig4 results in neuronal demyelination, while PtdIns(3,5)P 2 levels were increased in the former but decreased in the latter (14). This example emphasizes that correct PtdIns(3,5)P 2 dynamics could be as critical as its absolute levels.
It was also intriguing that overexpression of Fig4, but not Vac14, resulted in NSG1 mislocalization to nonendosomal structures. In the PtdIns(3,5)P 2 biosynthesis complex, PIKfyve: Fig4: Vac14 is present at a 1:1:5 ratio (4). Hence, Fig4 function might be more sensitive to changes in expression than Vac14. In addition, Fig4 is involved in both the biosynthesis and turnover of PtdIns(3,5)P 2 (4,61,62), which likely requires more delicate control of expression levels. Along the same line, it was reported that overexpression of Fig4 potentiated the aggregation of Lewy body-associated Synphilin-1 protein, while overexpression of Vac14 alleviated the aggregation (34).
CAD cells provide a unique "on-demand" model system for studying neurite growth. Neurite growth can be initiated after a specific molecular intervention such as PIKfyve knockdown and overexpression of ML1N×2, which is not possible in primary neurons and helps exclude the influence of cell viability. CAD cells also provide a simplified system to support in vivo findings. Consistent with our conclusion, it has been reported that axon calibers of optic nerves in Fig4 −/− mice were shifted to the smaller end (80). Note that this in vivo finding alone could be explained by either neurite growth defect or myelination defect in Fig4 −/− animals. Axon caliber and myelination regulation are intertwined, as axon caliber is increased by myelination (81), while the degree of myelination is correlated with axon caliber (80). In this case, CAD cells help delineate the cause and effect between PtdIns(3,5)P 2 and neuron morphology. The limitation of CAD cells is that it cannot be used to study electrophysiological responses due to the lack of axon/dendrite differentiation. It remains to be shown whether enhancing the function of the NSG1 pathway could rescue PtdIns(3,5)P 2 -related defects in animals and humans.
A possible limitation of this study is that the mCherry tag and/or expression levels of mCherry-NSG1 could alter the function or dynamics of NSG1 trafficking. In this regard, the fact that mCherry-NSG1 rescued neurite thickness defects in PtdIns(3,5)P 2 -impaired neurons suggests that this protein was indeed functional. In addition, the correlation between mCherry-NSG1 mislocalization and deformed neurite morphology in Fig4 and ML1N×2 overexpression cells indicates that the transport of this tagged protein is tightly linked to neurite growth.
In summary, we identified neurite thickness as a novel physiological parameter regulated by PtdIns(3,5)P 2 . We further found that NSG1, a neuronal-specific endosomal protein, functions downstream of PtdIns(3,5)P 2 and supports neurite morphogenesis. Our results indicate that the transport of NSG1 contributes to the sensitivity of neurons to PtdIns(3,5)P 2 and this pathway warrants further studies as a target to alleviate PtdIns(3,5)P 2 -related neuropathies.

Cell culture
All animal protocols were approved by the Animal Care and Use Committee at Soochow University. Mouse hippocampal neurons were cultured as described in (30). CAD cells were a gift from Dr Kristen Verhey at the University of Michigan and cultured in DMEM/F-12 (Life Technologies, C11330500BT)based medium supplemented with 10% fetal bovine serum (FBS, Biological Industries, 04-001-1a-us), 1% GlutaMAX (Life Technologies, 35050061), and 1% Penicillin-Streptomycin (Life Technologies, 15140122). HeLa and HEK293T cells were gifts from Dr Lois Weisman at the University of Michigan and cultured in Dulbecco's modified Eagle's medium (Invitrogen C11965500BT)-based medium supplemented as above. PC12 cells were obtained from the Institute of Neuroscience at Soochow University and cultured in RPMI (Life Technologies, C22400500BT)-based medium supplemented as above.
Transfection CAD cells were seeded at 4 × 10 5 cells per 35 mm dish in culturing medium without antibiotics. After 24 h, cells were transfected in the absence of FBS with 1 mg/ml PEI MAX (Polysciences Inc, 24765) at the ratio of 2:8 (μg plasmids: μl PEI MAX) diluted in Opti-MEM. After incubation for 4 to 6 h, the transfection medium was replaced with fresh media for another 24 h before being changed into the differentiation medium (full medium without FBS). CAD cells were differentiated for 36 to 48 h, followed by fixation for microscopy.
Primary cortical neurons were transfected with lipofectamine 2000 (Invitrogen, 11668027) according to the manufacturer's instruction at 2 div and used for downstream experiments at 24 h after transfection.
For live cell imaging, CAD cells were grown on glass-bottom culture dishes and transfected with fluorescent proteins for 24 h, followed by differentiation for another 24 h. Cells were changed into Hepes-buffered phenol-red free medium, placed in a heated humidity chamber (Tokai Hit), and imaged as above. For live imaging experiments with PIKfyve inhibitors, the imaging was started at 30 min after inhibitor addition to minimize the effect of phototoxicity.
For immunofluorescence, fixed cells were permeabilized with 0.1% Triton X-100 for 5 min, washed with PBS, and then incubated with blocking solution (2 % normal goat serum in PBS) for 1 h. After blocking, cells were incubated with primary antibody diluted in the blocking solution (anti-Rab5, 1:200; 1:100 for all other antibodies in this study) for overnight at 4 C, washed with PBS, followed by incubation with secondary antibodies (Alexa 488 goat anti-mouse for Rab5 and Tuj1; Alexa 488 goat anti-rabbit for Map2; Alexa 555 goat antimouse for Tau and NSG1, Alexa 647 goat anti-mouse for EEA1 and CD63; Alexa 647 goat anti-rat for LAMP1, 1:200 in blocking solution). Slides were mounted and imaged as above.

RNA interference
Lentiviral plasmids (pLKO.1-puro based, encoding puromycin resistance and no fluorescent protein) containing shRNA against mouse PIKfyve gene and nontargeting control shRNA were described in (29). Lentivirus particles were packaged in HEK293T cells by cotransfection with pMD2G and psPAX2 plasmids at the following ratio: 1 μg viral vector, 0.5 μg pMD2G, and 0.75 μg psPAX2. Supernatant media were collected at 36 to 72 h and filtered with 0.45 μm filters (Merck, SLHV033RB). For infection, CAD cells were incubated with virus-containing media mixed with fresh medium at 1:1 ratio, together with 4 μg/ml polybrene for 24 h. After growth for another 24 h, infected cells were selected with 2 μg/ml puromycin for 2 to 3 days until> 90% of control cells were killed. Stably infected cells were maintained in 2 μg/ml puromycincontaining medium during growth and differentiation and used within 1 month.
Plasmids encoding shRNA against mouse NSG1 were designed and cloned by GenePharma Inc. The plasmid backbone used was pGPU6/GFP/Neo, which also encoded GFP to help identify transfected cells. The following targeting sequences were used: shNSG1-800: GCT TCG ACA CCA TTC CTT TGA; shNSG1-1073: GCC CTG ATG GGT TTG TCT TGA; shNSG1-1125: GAG CTA CTA CAC GGA GCA AGA; nontargeting control: GTT CTC CGA ACG TGT CAC GT. CAD cells were transiently transfected with shRNA-expressing plasmids, differentiated for 24 h, and fixed for imaging.

Coimmunoprecipitation
HEK293T cells were transfected with 3xFLAG-NSG1 and Cit-Vac14 or Citrine control plasmids. After 24 h, cells were scraped and homogenized by forcing through a 22-gauge needle several times in lysis buffer (10 mM Hepes, 100 mM NaCl, 1 mM EDTA, 1× Proteinase inhibitor cocktail). Crude extracts were spun down at 14, 000 rpm for 20 min at 4 C. The supernatant was incubated with washed Anti-FLAG M2 affinity gel (Sigma-Aldrich, A2220) overnight at 4 C. The resin was washed three times by spinning down at 2000 g for 3 min and incubation with 1:1 mixed lysis buffer and PBS. Immunoprecipitated proteins were eluted by heating in Laemmli sample buffer at 95 C for 5 min and spun down at 2000 g. Proteins were separated by SDS-PAGE electrophoresis and transferred to PVDF membrane. The membrane was blocked with 5% nonfat milk and incubated with primary antibodies (rabbit anti-FLAG or rabbit anti-GFP, 1: 1000 in the blocking buffer) overnight at 4 C, followed by washing and incubation with HRP-conjugated goat anti-rabbit secondary antibodies (1:1000). The membrane was developed by Immobilon Western Chemiluminescent HRP Substrate (Millipore, WBKLS0500).

Neurite thickness and colocalization analysis
Primary hippocampal neurons were immunostained with anti-Map2 and anti-Tau to label dendrites and axons, respectively. CAD neurons were transfected with Citrine as a volume marker to delineate neurites. Primary cortical neurons were labeled with anti-Tuj1 to identify neurites. At least five random fields were imaged for each coverslip. Neurite diameter was measured on individual neurons pooled from two to three independent experiments. As neurites taper gradually from the base toward the tip, the diameter was measured at the proximal segment of the thickest neurite with ImageJ's line tool manually. Neurite diameters in different conditions were compared using nonparametric tests, which do not require the assumption of normality. For two-group comparisons, Mann-Whitney U test was performed. For three-group comparisons, Kruskal-Wallis test was performed, followed by post hoc Dunn's multiple comparisons test.
Colocalization was measured on images of single z-planes using the JACoP plugin of ImageJ. Images were autothresholded. Manders' coefficient was used, which indicates the fraction of red pixels that overlaps with the green channel or the fraction of green pixels that overlaps with the red channel. Normality of the colocalization coefficient was not assumed, and nonparametric tests were used as above for comparison among different groups.

Statistics
Statistical analysis and graphing were carried out with Prism 9 (GraphPad). Statistical tests used were indicated in the figure legends. Student's t test and one-way ANOVA were used for parametric testing. Kruskal-Wallis test and Mann-Whitney U test were used for nonparametric testing. Significance levels were set at p < 0.05. Data were shown as mean ± (SD).

Data availability
All data are contained within this article and in the supporting information.
Supporting information-This article contains supporting information.