Comprehensive analysis of autophagic functions of WIPI family proteins and their implications for the pathogenesis of β-propeller associated neurodegeneration

Abstract β-propellers that bind polyphosphoinositides (PROPPINs) are an autophagy-related protein family conserved throughout eukaryotes. The PROPPIN family includes Atg18, Atg21 and Hsv2 in yeast and WD-repeat protein interacting with phosphoinositides (WIPI)1–4 in mammals. Mutations in the WIPI genes are associated with human neuronal diseases, including β-propeller associated neurodegeneration (BPAN) caused by mutations in WDR45 (encoding WIPI4). In contrast to yeast PROPPINs, the functions of mammalian WIPI1–WIPI4 have not been systematically investigated. Although the involvement of WIPI2 in autophagy has been clearly shown, the functions of WIPI1, WIPI3 and WIPI4 in autophagy remain poorly understood. In this study, we comprehensively analyzed the roles of WIPI proteins by using WIPI-knockout (single, double and quadruple knockout) HEK293T cells and recently developed HaloTag-based reporters, which enable us to monitor autophagic flux sensitively and quantitatively. We found that WIPI2 was nearly essential for autophagy. Autophagic flux was unaffected or only slightly reduced by single deletion of WIPI3 (encoded by WDR45B) or WIPI4 but was profoundly reduced by double deletion of WIPI3 and WIPI4. Furthermore, we revealed variable effects of BPAN-related missense mutations on the autophagic activity of WIPI4. BPAN is characterized by neurodevelopmental and neurodegenerative abnormalities, and we found a possible association between the magnitude of the defect of the autophagic activity of WIPI4 mutants and the severity of neurodevelopmental symptoms. However, some of the BPAN-related missense mutations, which produce neurodegenerative signs, showed almost normal autophagic activity, suggesting that non-autophagic functions of WIPI4 may be related to neurodegeneration in BPAN.

The PROPPIN family includes Atg18, Atg21 and Hsv2 in yeast, ATG-18 and EPG-6 in Caenorhabditis elegans, and WD-repeat protein interacting with phosphoinositides (WIPI)1-4 in mammals. PROPPINs are phylogenetically divided into two paralog groups (10,11): one group includes Atg18, Atg21, ATG-18 and WIPI1/2, and the other includes Hsv2, EPG-6 and WIPI3/4 (10,11). The functions and hierarchical relationships of the three yeast PROPPINs have been well documented. Among them, Atg18 is essential for autophagy (12,13); Atg18 forms a complex with Atg2, and the formation of this complex is necessary for the recruitment of both proteins to phagophores (14,15). Atg21 is only partially required for autophagy; Atg21 interacts with Atg16 and recruits the Atg12-Atg5-Atg16 complex to induce Atg8-PE conjugation (16)(17)(18). Although Hsv2 is dispensable for autophagy, Hsv2 also interacts with Atg2 (19). In addition to autophagic functions, yeast PROPPINs have non-autophagic functions. For example, Atg18 is involved in the regulation of vacuolar size (20), vacuolar membrane fission (21) and transport from endo-lysosomal compartments to the Golgi apparatus (20,22,23). This function of Atg18 requires interaction with the retromer proteins Vps26 and Vps35, but this Atg18 complex is distinct from the retromer (22,23). It has also been reported that Hsv2 is localized to endosomes under basal conditions, although its physiological role is unknown (19). In C. elegans, EPG-6 associates with ATG-2 and regulates late steps of autophagosome formation, whereas ATG-18 functions in earlier steps of autophagosome formation (24). Thus, the functions of the two paralog groups in autophagy, especially in the recruitment of Atg2 to phagophores, are complicated and even differ among species.
In contrast to yeast PROPPINs, the function of each WIPI protein and the functional relationships among WIPI1-4 have not been fully elucidated. Although WIPI2 has been shown to be involved in autophagy through recruiting ATG16L1 to phagophores (25), the roles of WIPI1, WIPI3 and WIPI4 in autophagy remain poorly understood, and it is still unclear whether WIPI proteins contribute to the recruitment of ATG2A/B to phagophores in mammalian cells. One study on the functions of WIPI1-4 showed that the stable knockdown of any WIPI impaired autophagosome formation in the human melanoma cell line G361 and suggested that WIPI3 and WIPI4 function downstream of WIPI1 and WIPI2 (26). Biochemical and structural studies revealed that WIPI4 strongly interacts with ATG2A/B (19,26,27), and other studies showed that ATG2A/B also interacts with WIPI3 to a lesser extent in vivo (28) and that a short peptide derived from ATG2A/B interacts with WIPI3 in vitro (29), suggesting possible contributions of WIPI3 and WIPI4 to the recruitment of ATG2A/B. Although WIPI3, WIPI4 and ATG2A/B are expressed ubiquitously, simultaneous deletion of WIPI3 and WIPI4 causes an autophagy defect observed only in cultured neuronal cells and mouse brains, but not in cultured non-neuronal cells, suggesting that WIPI3 and WIPI4 are important only in neuronal tissues (30,31). Moreover, it was shown that an ATG2A mutant (residues Y1395, F1396 and S1397 all mutated into alanines), which is unable to interact with WIPI4, supported the degradation of p62 and LC3B during starvation (32), suggesting the interaction between ATG2A/B and WIPI4 (and likely WIPI3) (28,29) is dispensable for autophagic f lux. It was also shown that ATG2A/B proteins are recruited to phagophores by the interaction between ATG2A/B and ATG8 family proteins such as GABARAP, GABARAPL1 and LC3A (32). Therefore, the functional relationships among WIPI proteins seem to be complicated and merit further clarification.
Mutations in WIPI genes are known to cause human neuronal diseases. For example, mutations in WDR45, which encodes WIPI4, cause β-propeller protein-associated neurodegeneration (BPAN, formally called static encephalopathy of childhood with neurodegeneration in adulthood [SENDA], OMIM#300894), which is typically characterized by static neurodevelopmental symptoms in the early stage of the disease and neurodegeneration in the late stage of the disease (33)(34)(35). Mutations in WDR45B, which encodes WIPI3, also cause a complex neurodevelopmental disorder called El-Hattab-Alkuraya syndrome (OMIM#617977) (36)(37)(38), while mutations in WIPI2 cause a neurodevelopmental disorder accompanied by skeletal and cardiac abnormalities (OMIM#618453) (39,40). To understand the pathogenesis of these WIPI-related diseases, it would be helpful to reveal the function of each WIPI protein more precisely.
Because the contributions of each WIPI protein and of the interaction of ATG2 with WIPIs and ATG8s to autophagy may be only partial, more sensitive, quantitative methods for monitoring autophagic f lux would be necessary for a systematic analysis of these proteins. Recently, we developed a highly sensitive and quantitative method, the HaloTag (Halo)-based reporter processing assay (41). Using this method, we comprehensively analyzed the function of individual WIPI proteins using single or multiple WIPI-knockout HEK293T cells and revealed the necessity of WIPI2 in autophagy and the rigorous contribution of WIPI4 to the recruitment of ATG2 to phagophores. Finally, we found a possible association between the magnitude of autophagy defect and the severity of neurodevelopmental symptoms by assessing the autophagic activities of disease-related mutants of WIPI3 and WIPI4. In contrast, no clear correlation was found between the magnitude of autophagy defects and the age of the onset of neurodegeneration, suggesting that neurodegeneration in BPAN may be caused by a defect in a non-autophagic function of WIPI4.

WIPI2 is nearly essential for mammalian autophagy
To investigate the importance of individual WIPI proteins in mammalian autophagy, we generated WIPI1-4 single knockout (KO) and WIPI1-4 quadruple KO (QKO) HEK293T cells (Supplementary Material, Fig. S1). In addition, because of the close phylogenetic relationships between WIPI1 and WIPI2 and between WIPI3 and WIPI4 (10,11), we generated WIPI1/2 double KO (DKO) and WIPI3/4 DKO HEK293T cells (Supplementary Material, Fig. S1). As ATG16L1 and ATG2A/B function immediately downstream of WIPIs (1,2), we also generated ATG16L1 KO and ATG2A/B DKO HEK293T cells (Supplementary Material, Fig. S1). To measure the autophagic activity of these cells using the Halo-based reporter processing assay, we stably introduced the Halo-monomeric GFP (Halo-GFP) and Halo-LC3B (simply referred to as Halo-LC3) reporters (41). When these Halo-based reporters are delivered to lysosomes via autophagy, they are efficiently degraded. However, Halo becomes resistant to lysosomal degradation when it is conjugated with Halo ligands, producing a free Halo fragment in an autophagydependent manner (Fig. 1A).
First, we used the Halo-GFP reporter, which is incorporated into autophagosomes non-selectively. When wild-type HEK293T cells stably expressing Halo-GFP were incubated with a tetramethylrhodamine (TMR)-conjugated ligand for 30 min and further incubated in a ligand-free amino-acid starvation medium for 4 h, the band corresponding to free Halo was observed by ingel f luorescence detection (Fig. 1B). In ATG16L1 KO cells, Halo-GFP processing was markedly impaired, but a faint Halo fragment was still detected, which is consistent with previous findings that a low level of autophagic f lux remains in cells lacking the ATG8 conjugation machinery (42)(43)(44). Halo-GFP processing was completely blocked in ATG2A/B DKO cells.
Among WIPI single KO cells, WIPI2 KO cells showed an almost complete absence of Halo-GFP processing, suggesting that WIPI2 is nearly essential for autophagy ( Fig. 1B and C). WIPI1 KO and WIPI3 KO had no significant impact on Halo-GFP processing, whereas WIPI4 KO caused a slight reduction. Halo-GFP processing was further impaired in WIPI3/4 DKO cells. Considering that WIPI3 and WIPI4 are closely related based on their protein sequences (10,11,29), these findings suggest the potential functional redundancy between WIPI3 and WIPI4, as previously suggested in neuronal cells (30). Next, we conducted the processing assay using the Halo-LC3 reporter, which is selectively sequestered into autophagosomes. Again, the processing of Halo-LC3 was completely blocked in both ATG2A/B DKO and WIPI1-4 QKO cells ( Fig. 1D and E). WIPI2 KO cells showed an almost complete absence of Halo-LC3 processing, and the residual Halo-LC3 processing was further impaired by the simultaneous knockout of WIPI1 and WIPI2 (in WIPI1/2 DKO cells), suggesting that, although WIPI2 plays a dominant role, WIPI1 may have a weak function similar to that of WIPI2 or a distinct function ( Fig. 1D and E). WIPI3/4 DKO cells showed a partial but still significant impairment of Halo-LC3 processing. Unlike the Halo-GFP processing assay, the Halo-LC3 processing assay showed no significant defect in WIPI4 KO cells. This may be caused by a difference in the dynamic range or sensitivity between the Halo-GFP and Halo-LC3 processing assays. Because Halo-LC3 is selectively and efficiently incorporated into autophagosomes, the Halo-LC3 processing assay has higher sensitivity to detect low autophagic activities (e.g. in WIPI2 KO cells). However, when autophagic activities are relatively high (e.g. in WIPI4 KO cells), Halo-LC3 processing can become easily saturated and accordingly unable to reveal a subtle decrease in autophagic f lux. Thus, the Halo-GFP processing assay would be more suitable for detecting a slight defect.
In contrast to our Halo-based processing assay showing a clear decrease in autophagic f lux in WIPI3/4 DKO HEK293T cells (Fig. 1), a recent study suggested that WIPI3 and WIPI4 are dispensable for autophagy in non-neuronal cells, as p62 (also known as SQSTM1) and LC3 do not accumulate in WIPI3/4 DKO COS7 cells and mouse embryonic fibroblasts (30). Consistently, we confirmed no accumulation of lipidated LC3 (LC3-II) and p62 as well as normal degradation of LC3-II during starvation in WIPI3/4 DKO HEK293T cells (Fig. 1B, Supplementary Material, Fig. S2A-C). We also assessed autophagic f lux in WIPI3/4 DKO HEK293T cells using another reporter, GFP-LC3-RFP (Supplementary Material, Fig. S2D) (42). GFP-LC3-RFP is cleaved by endogenous ATG4 proteins to produce an equimolar amount of GFP-LC3 and RFP. GFP-LC3 is conjugated to autophagosomal membranes and then degraded or quenched, while RFP remains in the cytosol. Therefore, the reduction in the GFP:RFP f luorescence ratio represents autophagic f lux. We observed that the GFP:RFP f luorescence ratio was similar in wild-type and WIPI3/4 DKO HEK293T cells before and after starvation (Supplementary Material, Fig. S2E). Thus, the conventional methods using LC3 and p62 were unable to detect the slight autophagy defect in WIPI3/4 DKO HEK293T cells. This is probably due to the lower sensitivity of these degradation-based assays, which measure the reduction of autophagy substrates, compared with the Halo-based processing assay, which measures the production of the Halo fragment (i.e. positive signal).
In summary, in HEK293T cells, WIPI2 is nearly essential for autophagy, whereas the role of WIPI1 is limited. WIPI3 is not required and WIPI4 is only partially required for autophagy on their own, but they likely function redundantly. Considering the non-selective incorporation of Halo-GFP into autophagosomes, the Halo-GFP processing assay should ref lect the true activity of bulk autophagy. Therefore, we used it in the following experiments.

WIPI2 and WIPI4 regulate autophagy at different steps
Next, we introduced individual WIPIs or combinations of WIPIs into WIPI QKO cells. Among two WIPI1 isoforms (A and B) and six WIPI2 isoforms (A, B, C, D, E and delta), WIPI1A, WIPI2B and WIPI2D have been reported to be recruited to phagophores (10,11,25,26,43). In particular, WIPI2B has been shown to play an important role in autophagy in vivo (25). Therefore, we used WIPI1A and WIPI2B for this experiment. When expressed alone, only WIPI2B restored Halo-GFP processing in WIPI QKO cells ( Fig. 2A and B). The p62 level was also reduced. The restoration of Halo-GFP processing by WIPI2 was not changed by the further addition of WIPI1A or WIPI3, but was significantly enhanced by the addition of WIPI4 ( Fig. 2A and B). Almost full recovery was observed when both WIPI2B and WIPI4 were expressed. WIPI3 and WIPI4 could not restore Halo-GFP processing, even when both proteins were expressed ( Fig. 2A and B). These results suggest that WIPI2 and WIPI4 support autophagy at different steps and that WIPI2 is essential for the function of WIPI4 in autophagy.

WIPI4 regulates autophagy through its interaction with PI3P and ATG2A/B and maintains the size of autophagosomes
By using Halo-based reporters, we were able to detect a clear autophagy defect in WIPI3/4 DKO cells, which was difficult to detect by using conventional methods (Fig. 1, Supplementary Material, Fig. S2), enabling us to further examine the roles of WIPI3 and WIPI4 in autophagy. To investigate the significance of PI3P binding of WIPI3 and WIPI4, we generated WIPI3 R225T/R226T (WIPI3 LTTG) and WIPI4 R232T/R233T (WIPI4 LTTG) mutants, which completely lost their binding to PI3P (26,44,45). In WIPI3/4 DKO cells, the expression of either WIPI3 or WIPI4 significantly restored Halo-GFP processing, confirming that WIPI3 and WIPI4 are indeed functionally redundant ( Fig. 3A and B). In contrast, the WIPI3 LTTG and WIPI4 LTTG mutants failed to restore Halo-GFP processing, suggesting that WIPI3 and WIPI4 function through binding to PI3P ( Fig. 3A and B).
WIPI4 is known to interact with ATG2A/B (26,28,29). We confirmed that ATG2A/B interacts with WIPI4 almost exclusively among WIPI1-4 ( Fig. 3C). WIPI2B appeared to faintly interact with ATG2A/B. WIPI1A and, in contrast to the results of previous studies (28,29), WIPI3 did not interact with ATG2A/B in this experimental condition. Therefore, we investigated whether the WIPI4-ATG2A/B interaction is necessary for the function of WIPI4. For this purpose, the WIPI4 N15A/D17A mutant, which is unable to interact with ATG2A/B, was prepared (26) (Supplementary Material, Fig. S3). Halo-GFP processing was significantly impaired in WIPI4 N15A/D17A-expressing WIPI3/4 DKO cells in contrast to almost normal processing in wild-type WIPI4-expressing cells ( Fig. 3D and E). In this experiment, the expression level of WIPI4 N15A/D17A was remarkably decreased compared with wild-type WIPI4, but the level of WIPI4 N15A/D17A was still higher than that of endogenous WIPI4 (Fig. 3D). These results suggest that the interaction between WIPI4 and ATG2 is important for autophagy (28,29).
It was previously reported that the size of autophagosomes is smaller in mouse WIPI3/4 DKO Neuro-2a cells compared with parental cells (30). To confirm this pattern in HEK293T cells, we observed autophagosomes using 3D correlative light electron microscopy using GFP-LC3 as a marker (46). After a 1-h starvation period, GFP-LC3-positive closed autophagosomes were abundant in both wild-type and WIPI3/4 DKO cells (Fig. 3F, Supplementary Material, Fig. S4A). Furthermore, we confirmed that the number of GFP-LC3-positive structures was not reduced in WIPI3/4 DKO cells (Supplementary Material, Fig. S4B). However, the size of autophagosomes was strikingly reduced in WIPI3/4 DKO cells ( Fig. 3F and G). Collectively, these results suggest that WIPI4, and possibly WIPI3 also, support autophagic f lux through interactions with PI3P and ATG2A/B and maintain the size, rather than the number, of autophagosomes even in non-neuronal cells.

Recruitment of ATG2A to phagophores is mediated by its interaction with WIPI4 and ATG8s
Regarding the recruitment of ATG2A/B to phagophores, a previous study showed the significance of their interaction with ATG8s and suggested that the interaction between WIPI3/4 and ATG2A/B is dispensable for autophagy (32). However, using the Halo-GFP reporter, we instead found that the interaction between WIPI4 and ATG2A/B is crucial ( Fig. 3D and E). Therefore, we next investigated the relative contributions of these two interactions using Halo-based reporters. For this purpose, we prepared several mutants for ATG2A only, as ATG2A and ATG2B function redundantly (47) (Fig. 4A). ATG2A interacts with WIPI3/4 through its WIPI-interacting region (WIR, residues 1374-1404) (29), while ATG2A interacts with ATG8s through its LC3-interacting region (LIR, residues 1362-1365) (32). We mutated these regions to render ATG2A unable to interact with either ATG8s or WIPI3/4 or both (Fig. 4A). The WIR in ATG2A has a loop-like structure and is not involved in the formation of any specific domain of ATG2A (28,29). The LIR is located close to the WIR (32). Therefore, these mutations would not affect the overall structure and functions of ATG2A. We confirmed that LIR mutation reduced the interaction with GABARAP but not WIPI4, whereas WIR deletion almost completely blocked the interaction with WIPI4 but not GABARAP (Supplementary Material, Fig. S5A and B).
In ATG2A/B DKO cells, wild-type ATG2A fully rescued Halo-GFP processing ( Fig. 4B and C). In contrast, the LIR mutant (ATG2A mLIR) or the WIR deletion mutant (ATG2A WIR) showed partially impaired Halo-GFP processing ( Fig. 4B and C). ATG2A with both LIR mutation and WIR deletion (ATG2A mLIR WIR) showed a further decrease in Halo-GFP processing, suggesting that LIR and WIR collaboratively function ( Fig. 4B and C). These ATG2A mutants still interacted with WIPI2B to similar extents as wildtype ATG2A (Supplementary Material, Fig. S5C), suggesting that the interaction between ATG2A and WIPI2B is important at a different step than the interaction between ATG2A and WIPI4 in autophagy or is not important for autophagy.
Next, we investigated whether the reduced autophagic activity of the ATG2A mutants was attributable to defects in their recruitment to phagophores. ATG2A colocalized with WIPI2B, a phagophore marker and LipidTOX, a lipid droplet marker, as previously reported (47,48) (Fig. 5A and B). In contrast, the colocalization rates between ATG2A mutants (mLIR, WIR and mLIR WIR) and WIPI2B were significantly reduced ( Fig. 5A and B). Instead, most puncta of the ATG2A mutants colocalized with LipidTOX. These results suggest that the interactions of ATG2A with WIPI4 and ATG8s contribute to the recruitment of ATG2A to phagophores.
To further confirm that the defects of the ATG2A mutants were indeed attributable to impaired recruitment, we assessed whether the forced recruitment of ATG2A mLIR WIR to phagophores could recover the function of ATG2A. We fused the PI3Pbinding 2 × FYVE domain (49) or WIPI4 to ATG2A mLIR WIR (Fig. 5C). These fusion proteins significantly enhanced Halo-GFP processing in ATG2A/B DKO cells compared with ATG2A mLIR WIR, confirming that proper recruitment of ATG2A is important ( Fig. 5D and E).
Furthermore, we investigated the relationship between the magnitude of autophagy defect and the clinical phenotype of each mutation in WIPI4. Patients with these mutations vary in age, and some of the patients may have not yet developed symptoms indicative of neurodegeneration, such as parkinsonism and dementia. In addition, clinical phenotypes were not described in detail for all the patients. Therefore, we first focused on the severity of intellectual disability, which is relatively detailed in most of the reports, as an indicator of the severity of neurodevelopmental symptoms. We examined 2 male patients and 14 female patients (34,(50)(51)(52)(53)(54)(55)(56)(57)(58)(59)(60). In 5 female patients with mutations D84G, F100S, R134P, G168E and N202K, which were determined to have no impact or only a mild impact on autophagic activity, intellectual disability tended to be mild (52)(53)(54)56). Patients with mutations D84G and F100S could speak in sentences and began to walk at a typical age (52,53). Patients with mutations R134P and G168E, who were 2-year-7-month-old and 5-year-old patients, respectively, could each speak a couple of words (54). The patient with mutation N202K could not communicate verbally but could communicate partially in a non-verbal way (56). In contrast, the intellectual disabilities of patients with mutations L160R, A209D and S210P, which caused a profound defect in autophagic activity, seemed to be more severe, as they did not acquire any language abilities (55,58,59). Because WDR45, the gene encoding WIPI4, is on the X chromosome, the expression level of WIPI4 depends on the extent of X chromosome inactivation in each female case. In addition, the stabilities of mutant proteins may affect the expression level of each mutant WIPI4. Nevertheless, our data suggest a possible association between autophagic activity and the severity of intellectual disability.
Seven female patients developed symptoms indicative of neurodegeneration, such as dementia and parkinsonism  (35,56,(58)(59)(60). However, there was no clear correlation between the autophagic activity and the age of onset of these symptoms (Fig. 7). For example, although the N202K mutation had almost no impact on the autophagic activity of WIPI4, the patient with the N202K mutation developed parkinsonism at the age of 27 (56). Additionally, a patient with the D84G mutation, which also had no impact on the autophagic activity of WIPI4, showed iron accumulation in both the globus pallidus and substantia nigra, as detected by brain magnetic resonance imaging, indicating that the patient would likely develop parkinsonism in the future. Neurodevelopmental symptoms were very mild in this patient, probably ref lecting almost normal autophagic f lux. These findings suggest that WIPI4 has non-autophagic functions, which may be important for the prevention of neurodegeneration and iron accumulation.

Discussion
In this study, we conducted a systematic analysis of WIPI proteins using highly sensitive autophagic f lux reporters and multiple  Figure 1B. (C) Data from (B) were quantified, and the results are shown as described in Figure 1C. Solid bars indicate the means, and dots indicate the data from three independent experiments. Data were statistically analyzed by one-way ANOVA with Dunnett's test. knockout cell lines. We confirmed the nearly essential function of WIPI2 in autophagy (Fig. 1), which is consistent with a previous report showing that the depletion of WIPI2 caused the most drastic inhibition of autophagosome formation among the depletions of each of the WIPI proteins (26). It was also previously shown that the depletion of WIPI1 moderately inhibits autophagosome formation in G361 cells (26). However, our quantitative study in HEK293T cells showed, although WIPI1 may be redundant with WIPI2 possibly through its interaction with ATG16L1 (25,26), the contribution of WIPI1 was very limited (Fig. 8). This result may correspond to the finding that the interaction between WIPI1 and ATG16L1 is far weaker than that between WIPI2 and ATG16L1 (25). Instead, WIPI1 is endowed with a non-autophagic function related to endosomal membrane fission and endosomal transport (61), which may correspond to the non-autophagic function of yeast Atg18 in retrograde transport from endosomes to the Golgi apparatus (20,22,23) (Fig. 8).
The roles of WIPI3 and WIPI4 in autophagy have been controversial. LC3 accumulates upon double knockdown of WIPI3 and WIPI4 in G361 cells (26). Although p62 accumulates in the brain (but not in the liver or kidney) of Wipi3/4 DKO mice, it is virtually absent from the brain of Wipi3 KO and Wipi4 KO mice (31). The degradation of LC3 is reduced in fibroblasts derived from BPAN patients (62). On the other hand, it was reported that the levels of LC3 and p62 were normal in WIPI3/4 DKO COS7 cells and Wipi3/4 DKO mouse embryonic fibroblasts (30). In the present study, our Halo-based processing assay clearly revealed a decrease in autophagic f lux in WIPI3/4 DKO HEK293T cells, while canonical methods, such as LC3 and p62 degradation and GFP-LC3-RFP reporter assays, failed to detect the autophagy defect in WIPI3/4 DKO HEK293T cells. We assume that these previous inconsistencies might be a consequence of the relatively low sensitivities of LC3-based autophagic f lux assays to detect subtle autophagy defects. Given the similarity of the protein sequences of WIPI3 and WIPI4 (10,11,29), our results using appropriate methods suggest that WIPI3 and WIPI4 are redundantly important for autophagy, even in non-neuronal cells. Our data also suggest that, at least for autophagic f lux in HEK293T cells, WIPI4 appears to be more important than WIPI3, because the deletion of WIPI4 showed a larger suppressive effect on autophagy than that of WIPI3 (Fig. 1), and the rescue effect of re-expression of WIPI4 in WIPI1-4 QKO was greater than that of WIPI3 (in the presence of WIPI2) (Fig. 2). This may be because ATG2 interacts with WIPI4 more strongly than with WIPI3 (28). ATG2 was detected in the interactome of WIPI4 but not WIPI3 (26). Consistent with these results, we detected interaction between WIPI4 and ATG2A/B, but not between WIPI3 and ATG2A/B (Fig. 3C). In addition to the redundant functions, it is still possible that WIPI3 and WIPI4 have different and specific functions.
The next issue regards the relationship between the two functional units, WIPI1/2 and WIPI3/4. It was reported that, in C. elegans, ATG-18 (a homolog of WIPI1/2) appears to function upstream of EPG-6 (a homolog of WIPI3/4) (24). In mammals, WIPI1 and WIPI2 also function upstream of WIPI3 and WIPI4 (26). Our data are consistent with these previous reports; WIPI1/2 DKO cells showed an almost complete block in autophagic f lux, whereas WIPI3/4 DKO cells maintained partial autophagic activity (Fig. 1) and were able to form small autophagosomes (Fig. 3) as previously shown in neuronal cells (30), suggesting that WIPI1 and WIPI2 are important for an early step of autophagosome formation and that WIPI3 and WIPI4 are important for the elongation step. Our data showing the additive effect of WIPI2 and WIPI4 further suggest that WIPI2 and WIPI4 have distinct functions. ATG16L1 and ATG2A/B are known downstream factors of WIPI2 and WIPI4, respectively (1,2). It was previously reported that ATG8s, which are conjugated to PE by the ATG12-ATG5-ATG16L1 complex, recruit ATG2A/B via interaction with the LIR domain in ATG2A/B (32), but the relative contribution of ATG8s and WIPI3/4 to ATG2 recruitment has been unknown. We suggest that both are redundantly important, as the LIR mutation and WIR deletion showed an additive effect (Fig. 4B and C). Thus, the mechanism that recruits ATG2 in mammals is different from that in yeast (Fig. 8). Atg18 (a homolog of WIPI1/2) recruits Atg2 in yeast (14,15), whereas WIPI4 (and possibly WIPI3) recruits ATG2 in mammals. Although it remains unknown how much WIPI2 contributes to the direct recruitment of ATG2, WIPI2 can recruit ATG2 indirectly through ATG12-ATG5-ATG16L1 and ATG8-PE (Fig. 8). It should also be noted that the inhibition of Halo-GFP processing in WIPI2 KO cells was greater than that in ATG16L1 KO cells (Fig. 1B and C). The data suggest that WIPI2 has a yet unknown fundamental function beyond the recruitment of ATG16L1 (Fig. 8). A previous report has shown the interaction between WIPI2 and ATG2A/B (28), and we also detected a possible interaction between WIPI2B and ATG2A/B (Fig. 3C). The mutations of ATG2A, which reduce the interactions with GABARAP and WIPI3/4 and suppress the recruitment of ATG2A to phagophores, did not impair the interaction between  Figure 1B. (E) Data from (D) were quantified, and the results are shown as described in Figure 1C. Solid bars indicate the means, and dots indicate the data from three independent experiments. Data were statistically analyzed by one-way ANOVA with Dunnet's test. Figure 6. Some of the disease-associated WIPI4 mutants retain autophagic activity. (A) Halo-GFP processing assay of WT and WIPI3/4 DKO HEK293T cells with or without 3 × FLAG-tagged WIPI3/4 or their disease-related mutants was performed as in Figure 1B. (B) Data from (A) were quantified, and the results are shown as described in Figure 1C. Solid bars indicate the means, and dots indicate the data from three independent experiments. Data were statistically analyzed by one-way ANOVA with the Holm-Šídák test.
ATG2A and WIPI2B (Supplementary Material, Fig. S4C). Therefore, the interaction between ATG2 and WIPI2 may be important for a potential mechanism other than the recruitment of ATG2. This hypothesis may be consistent with our finding that WIPI2 and WIPI4 function in autophagy through different mechanisms (Fig. 2).
BPAN patients exhibit only neuronal symptoms (33)(34)(35), and Wipi3/4 DKO mice also demonstrate p62 accumulation only in the brain (31), leading to the hypothesis that the autophagic functions of WIPI3 and WIPI4 are specific to neuronal tissues. In this study, we revealed WIPI4 functions in autophagy, even in non-neuronal cells. Therefore, the tissue specificity of BPAN and related mouse models may be explained by the vulnerability of neuronal tissues to mild autophagy defects rather than tissue-specific functions, as seen in other autophagy-related diseases (63)(64)(65)(66)(67). Another possibility is that the relative importance of WIPI4 and ATG8s in recruiting ATG2A/B may differ between neuronal and nonneuronal tissues.
The expression levels of some of the BPAN-related WIPI4 mutants were reduced, suggesting that these mutations affect Figure 7. The magnitude of autophagy defects in WIPI4 mutants is possibly associated with the severity of neurodevelopmental symptoms but not with the age of neurodegeneration onset. The autophagic activity (%WT in Fig. 6B) of corresponding WIPI4 mutants and the age of the onset of neurodegeneration symptoms, such as parkinsonism and dementia, were plotted for seven BPAN cases. The parenthetical after each mutation indicates the age of the onset of these symptoms. The Pearson's correlation coefficient (r) between the degree of autophagic defect and the age of the onset of these symptoms is shown.
the stability of WIPI4. In particular, the expression levels of the R13P, V54del, V66E, F100S and R134P mutants were reduced. These residues are all located close to the interaction surface with ATG2A/B (28,29) and most of these mutations change the polarity of these amino acids. In addition, R13, V54, V66 and F100 are likely involved in β-sheet formation. Thus, mutations that alter the interaction between WIPI4 and ATG2A/B, the polarity of amino acids, and β-sheet formation may affect the stability of WIPI4.
Halo-based reporter assays revealed various effects of BPANrelated missense mutations in WIPI4 on their autophagic activity. We found a possible association between the magnitude of autophagy defect and the severity of neurodevelopmental symptoms, which is consistent with the previous findings that neurodevelopmental defects are core symptoms in autophagy-related diseases with mutations in EPG5, ATG5 and ATG7 (63)(64)(65)(66)(67). In contrast, we detected no correlation between the magnitude of autophagy defect and the age of the onset of neurodegeneration symptoms; some mutations found in BPAN patients with iron deposition and neurodegeneration symptoms showed almost normal autophagic activity. These findings suggest that the development of iron deposition and neurodegeneration in BPAN is attributable to unknown non-autophagic functions of WIPI4, for example, in retrograde transport from endosomes to the Golgi apparatus, as was also suggested for its most closely related yeast PROPPIN, Hsv2 (Fig. 8) (19). Elucidating the non-autophagic functions of WIPI4 is necessary to develop therapeutics for the prevention of neurodegeneration in the disease.
One of the limitations of this study is that we used HEK293T cells, which may under-or overestimate the effect of BPAN-related missense mutations in WIPI4 in the central nervous system. It would be ideal to carry out Halo-based reporter assays using neuronal cell lines or patient-derived cells. In addition, we used language ability as an indicator of the severity of intellectual disability in this study. However, to accurately assess intellectual disability, we need more quantitative indicators, such as intelligence quotient (53). A more detailed analysis of the relationship between the autophagic activity in the central nervous system and the clinical severity of the patients would provide a better understanding of the pathogenesis of BPAN.

Cell lines
HEK293T cells and HeLa cells were incubated in Dulbecco's modified Eagle's medium (DMEM; D6546, Sigma-Aldrich, St. Louis, MO) supplemented with 10% fetal bovine serum (FBS, 173012, Sigma-Aldrich) and 2 mm l-glutamine (25030-081, Gibco, Waltham, MA) at 37 • C in a 5% CO 2 incubator. For starvation, after being washed twice with phosphate-buffered saline (PBS), cells were cultured in amino acid-free DMEM (048-33575, Fujifilm Wako Pure Chemical Corporation, Osaka, Japan) without serum (starvation medium) for the time indicated for each experiment. For bafilomycin A 1 treatment, cells were incubated in a starvation medium with 100 nm bafilomycin A 1 (B1793, Sigma-Aldrich) for the time indicated for each experiment.

Preparation of retrovirus for stable expression
For the preparation of retrovirus, HEK293T cells were transfected with a retroviral vector together with pCG-VSV-G and pCG-gagpol (gifts from Dr T. Yasui), using Lipofectamine 2000 (11668019, Thermo Fisher Scientific). Two days after transfection, the supernatant was passed through a 0.45 μm syringe filter unit (SLHV033RB, Merck Millipore, Billerica, MA) and collected. Then, the retrovirus was applied to cells, and stable transformants were selected by application of 2 μg/ml puromycin (P8833, Sigma-Aldrich), 5 μg/ml blasticidin (022-18713, Wako) and 100 μg/ml hygromycin B (10687-010, Thermo Fisher Scientific). For the generation of ATG2A/B DKO HEK293T cells stably expressing ATG2A or its variants, ATG2A/B DKO HEK293T cells were infected with the retrovirus carrying ATG2A or ATG2A variants and the internal ribosomal entry site drive SNAP-tag. Stable transformants were labeled with 100 nm SNAP-Cell 647-SiR ligand (New England Biolabs, S9102S). The f luorescence signals from SNAP-Cell 647-SiR were detected, and cells with the same f luorescence intensities were collected by a cell sorter (MoFlo AstriosEQ; Beckman Coulter).

Protein extraction and immunoblotting
Cells were lysed with 50 mm Tris-HCl, pH 7.5, 150 mm NaCl, 1 mm MgCl 2 , 0.2% n-dodecyl-β-d-maltoside and Complete EDTAfree protease inhibitor cocktail (05056489001, Roche, Basel, Switzerland) for 10 min on ice and then treated with 0.1% benzonase (9025-65-4, Sigma-Aldrich) for 10 min on ice. The protein concentrations in the whole cell lysates were measured by the BCA method. The whole cell lysates were mixed with 6× sample buffer (46.7 mm Tris-HCl, pH 6.8, 5% glycerol, 1.67% SDS, 1% β-mercaptoethanol and 0.02% bromophenol blue) and heated to 95 • C for 5 min. The protein concentrations were adjusted with 1 × sample buffer. Samples were subjected to sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to a polyvinylidene dif luoride (PVDF) membrane (IPVH00010, Millipore) using the Trans-Blot Turbo Transfer System (BioRad, Hercules, CA). Immunoblotting analysis was performed with the antibodies indicated in the figures. The signals were visualized with Immobilon Western Chemiluminescent HRP Substrate (P90715, Millipore) and were detected on a Fusion SOLO.7S.EDGE (Vilber-Lourmat, Collégien, France). Contrast and brightness were adjusted and quantified using the Fiji image processing package (70).

Halo-GFP/LC3 processing assay
Cells were incubated with 100 nm TMR-conjugated HaloTag ligand (G8251, Promega) for 30 min. After being washed twice with PBS, cells were cultured in the starvation medium for the time indicated. Then, cells were harvested, and proteins were obtained. Proteins were separated by SDS-PAGE, and the gel was immediately visualized for in-gel f luorescence from TMR with a Fusion SOLO.7S.EDGE (Vilber-Lourmat). For the calculation of Halo-LC3 processing, the images of gels were captured with an exposure time of 30 s. The Halo-LC3 processing rate was calculated as the intensity of the free Halo band divided by the sum of the intensities of the free Halo and unprocessed Halo-LC3 bands. In the Halo-GFP processing assay, Halo processing is usually so low that the intensity of free Halo and unprocessed Halo-GFP bands cannot be visualized in the same image because the appropriate exposure time for each is different. Therefore, the images of gels were obtained twice with exposure times of 2 s and 10 min. The Halo-GFP processing rate was calculated as the intensity of the free Halo band in the 10-min exposed image divided by the intensity of the unprocessed Halo-GFP band in the 2-s exposed image (total Halo-GFP [processed + unprocessed] can be approximated as the intensity of unprocessed Halo-GFP in a 2 s exposed image, in which processed Halo can be hardly observed).

Fluorescence imaging
HeLa cells expressing muGFP-ATG2A and WIPI2B-HaloTag7 were seeded onto glass-bottom dishes (3911-035, IWAKI) 48 h before observation. Cells were cultured in starvation medium with 200 nm SF650-conjugated HaloTag ligand (A308-02, Goryo Chemical) and LipidTOX Red (H34476, Thermo Fisher Scientific) for the time indicated. Then, cells were observed using a confocal laser microscope (FV3000, Olympus, Tokyo, Japan) equipped with a 60× oil immersion objective lens (1.42 NA, Olympus). Obtained images were processed using the Fiji image processing package (70). Each cell was isolated in a single image and processed individually. After being convolved with the Gaussian Blur filter in Fiji (sigma = 2), images were processed with the Top Hat filter (radius = 10). A threshold was determined for each channel and applied to all images in each channel to binarize images. Binarized images were processed by the Watershed filter, and then puncta were identified and counted by the Analyze Particles tool. Colocalized puncta were visualized by applying the Multiply function in Fiji's Image Calculator to two binarized images of interest and then counted using Analyze Particles again.

3D correlative light and electron microscopy
To observe and measure autophagosomes, 3D correlative light and electron microscopy (3D-CLEM) analysis was performed as described previously (46). Brief ly, gridded coverslip-bottom dishes (TCI-3922-035R-1CS, a custom-made product based on 3922-035, with cover glass attached in the opposite direction, IWAKI, Sizuoka, Japan) were coated with carbon by a vacuum evaporator (IB-29510VET, JEOL, Tokyo, Japan) and treated with 10 μg/ml poly-l-lysine for 15 min, followed by UV irradiation. Cells were seeded onto the custom-made glass-base dishes 48 h prior to the experiment. After incubation in the starvation medium for 1 h, cells were fixed with 2% paraformaldehyde (26126-54, Nacalai Tesque, Kyoto, Japan) and 0.5% glutaraldehyde (G018/1, TAAB) in 0.1 M PB, pH 7.4 for 1 h at 4 • C. Then, cells were observed using a confocal laser microscope (FV3000, Olympus) equipped with a 60× oil immersion objective lens (1.42 NA, Olympus). After obtaining f luorescence images, the cells were embedded in epoxy resin for observation with an electron microscope as described previously (46). After cells were embedded and removal of the coverslip, the resin block was trimmed into a 150 × 150 μm square, which included the area observed by the confocal laser microscope (FV3000, Olympus). Then, the block was cut to a 25-nm thickness to create a ribbon of 50-200 serial sections using a ultramicrotome (EM UC7, Leica, Wetzlar, Germany) equipped with a diamond knife (Ultra JUMBO, 45 degrees, DiATOME, Hatfield, PA). Those sections were observed by a scanning electron microscope (JSM7900, JEOL) with the assistance of Array Tomography Supporter software (System in Frontier, Tokyo, Japan). The images were stacked in order using Stacker NEO TEMography.com 3.3.4.0 software (System in Frontier). The alignment of the images obtained with the confocal laser microscope and the electron microscope was done using Fiji software. The maximum cross-sectional areas of autophagosomes were also measured using Fiji software. The observation of autophagosomes, the quantification of autophagosome sizes, and the statistical analysis were performed in a double-blinded manner.

Statistical analysis
Statistical analyses were performed with GraphPad Prism 8 software (GraphPad Software, Boston, MA). The statistical method used for each experiment is described in each figure legend.

Supplementary Material
Supplementary Material is available at HMG online.