A Drosophila model of neuronal ceroid lipofuscinosis CLN4 reveals a hypermorphic gain of function mechanism

The autosomal dominant neuronal ceroid lipofuscinoses (NCL) CLN4 is caused by mutations in the synaptic vesicle (SV) protein CSPα. We developed animal models of CLN4 by expressing CLN4 mutant human CSPα (hCSPα) in Drosophila neurons. Similar to patients, CLN4 mutations induced excessive oligomerization of hCSPα and premature lethality in a dose-dependent manner. Instead of being localized to SVs, most CLN4 mutant hCSPα accumulated abnormally, and co-localized with ubiquitinated proteins and the prelysosomal markers HRS and LAMP1. Ultrastructural examination revealed frequent abnormal membrane structures in axons and neuronal somata. The lethality, oligomerization and prelysosomal accumulation induced by CLN4 mutations was attenuated by reducing endogenous wild type (WT) dCSP levels and enhanced by increasing WT levels. Furthermore, reducing the gene dosage of Hsc70 also attenuated CLN4 phenotypes. Taken together, we suggest that CLN4 alleles resemble dominant hypermorphic gain of function mutations that drive excessive oligomerization and impair membrane trafficking.


Introduction
NCLs comprise a group of progressive neurodegenerative diseases with 14 known disease-associated genes, termed CLN1-14 (Haltia, 2003;Haltia and Goebel, 2013;Jalanko and Braulke, 2009;Mole and Cotman, 2015). NCLs have mostly an infantile or juvenile symptomatic onset and are characterized by loss of vision, gait abnormalities, seizures, dementia, and premature death. In general, NCLs are considered lysosomal storage diseases due the accumulation of lipofuscin and are typically caused by recessive loss of function mutations with one exception, CLN4. Accordingly, lysosomal dysfunction, dysregulated ER-lysosomal trafficking, or aberrant lipid modifications are thought to be the basis for most NCLs, consistent with the known function of the mutated genes (Bennett and Rakheja, 2013;Cárcel-Trullols et al., 2015;Mole and Cotman, 2015;Warrier et al., 2013). Understanding disease mechanisms of NCLs has implications as well for more prevalent diseases since mutations in a growing number of CLN genes also cause other diseases like frontotemporal lobar degeneration, progressive epilepsy with mental retardation, spinocerebellar ataxia, retinitis pigmentosa, juvenile cerebellar ataxia, or Parkinson disease 9 (Bras et al., 2012;Mole and Cotman, 2015;Yu et al., 2010).
The autosomal dominantly inherited NCL CLN4 has an adult onset between 25 and 46 years. lysosomal dysfunction and neurotoxicity remains unclear. Therefore, an animal model may provide critical insight into molecular and cellular mechanisms underlying CLN4 disease.
Here, we generated two Drosophila models for CLN4 by expressing either CLN4 mutant hCSPa or dCSP in fly neurons. The humanized fly model has the unique advantage of being able to differentially visualize CLN4 mutant hCSPa and WT dCSP. We show that both fly models replicate all key pathogenic biochemical properties of CLN4, including decreased monomeric CSP levels, increased levels of high-molecular weight, and ubiquitinated CSP oligomers. Further analysis revealed novel insights into mechanisms underlying CLN4 pathology.

Generation of a Drosophila CLN4 model
We generated a Drosophila model of CLN4 by expressing the disease-causing human proteins hCSPa-L115R, hCSPa-L116D ( Figure 1A; denoted as L115 and L116 from now on) and the corresponding WT hCSPa control from a common genomic phi31-attP insertion site under the transcriptional control of the yeast Gal4-UAS expression system (Brand and Perrimon, 1993). Unless otherwise indicated, we used the pan-neuronal elav-Gal4 driver C155 (Lin and Goodman, 1994) to express these proteins exclusively in otherwise WT neurons (w 1118 ).
To evaluate whether WT hCSPa is at least partially functional in fly neurons, we first determined whether neuronally expressed hCSPa is properly palmitoylated and targeted to SVs. This analysis was aided by the availability of species-specific antibodies that discriminate between hCSPa and dCSP (Figures 1B-C and 2A; Figure 1-figure supplement 1A). Neuronally expressed hCSPa was efficiently trafficked to axon terminals and co-localized with endogenous dCSP in the neuropil of the larval ventral nerve cord (VNC; Figure 1B) and larval neuromuscular junctions (NMJs; Figure 1C). Consistent with the normal trafficking of hCSPa, the majority of neuronally expressed hCSPa was palmitoylated, which was confirmed by treating larval protein extracts with 0.5 M hydroxylamine (Figure 1-figure supplement 1A). Signals were normalized to loading control and plotted as n-fold change from 1-copy expression of WT hCSPa (N = 6). (J) Levels of monomeric dCSP shown as n-fold change from control (N = 3). Graphs display mean ± SEM. Statistical analysis used one-way ANOVA (C-F, J) or two-tailed unpaired t test (G-I); *, p<0.05; **, p<0.01; ***, p<0.001. DOI: https://doi.org/10.7554/eLife.46607.004 Next, we tested whether hCSPa can functionally replace endogenous dCSP and restore the premature lethality of flies lacking dCSP . Flies and mice lacking CSP due to a gene deletion exhibit progressive neurodegeneration, paralysis and premature death (Fernández-Chacó n et al., 2004;Zinsmaier et al., 1994). Pan-neuronal elav-driven expression of normal hCSPa from one transgenic copy in homozygous dcsp deletion mutants significantly restored adult lifespan from~4-5 days to 15-20 days (LD50 p<0.001; Figure 1D). In comparison, expression of fly dCSP2 also restored adult lifespan only partially ( Figure 1D). Even though this rescue was only partial, it confirms that hCSPa is functional in flies, which made a CLN4 Drosophila model using human proteins tenable.
Notably, neuronal expression of hCSP-L115 and -L116 in a dcsp null background was able to partially rescue the lifespan deficit (p<0.001, LD50), although not nearly to the extent of WT hCSPa ( Figure 1D). This indicates that CLN4 mutant hCSPa proteins are at least partially functional.
To test whether expression of CLN4 mutant hCSPa in Drosophila neurons recapitulates the pathological features of post-mortem human brains, we analyzed the properties of CLN4 mutant hCSPa by immunoblotting, using protein extracts from larval VNCs. When expressed from a single transgene with an elav-Gal4 driver, lipidated monomeric protein levels of WT hCSPa were estimated to be 0.98 ± 0.23 times of endogenous dCSP levels (data not shown). In comparison, levels of lipidated monomeric hCSP-L115 and -L116 were significantly reduced to 13% and 40% of WT hCSPa levels, respectively (p<0.003, Figure 2A,D). Levels of lipidated hCSP-L115 monomers were significantly lower than hCSP-L116 (p<0.01, Figure 2D). Levels of non-lipidated monomeric hCSP-L115 and -L116 were comparable to WT hCSPa levels (p>0.2; Figure 2A,E).
Both CLN4 mutations induced the formation of SDS-resistant, high-molecular weight hCSPa oligomers in Drosophila neurons (p<0.002; Figure 2A,C). In contrast, WT hCSPa oligomers were barely detectable (Figure 2A,C). The mutation L115 triggered oligomerization to a significantly larger degree than L116 (p<0.006; Figure 2C), which is consistent with the lower levels of hCSP-L115 monomers ( Figure 2D). Increasing levels of DTT in the buffer to reduce disulfide bonds had little to no effect on the levels and size of hCSP-L115 and -L116 oligomers, which varied widely from~250 kDa to more than 500 kDa (not shown).
To determine whether CLN4 mutant hCSPa oligomers are ubiquitinated, we immunoprecipitated hCSPa from larval brain cell lysates and probed western blots with an antibody that specifically detects K29-, K48-, and K63-linked mono-and poly-ubiquitinated proteins but not free ubiquitin monomers (Fujimuro et al., 1994). hCSPa antibodies immunoprecipitated both hCSPa monomers and oligomers ( Figure 2B) but not endogenous dCSP (not shown). The anti-ubiquitinated protein antibody only recognized a strong~250 kD protein band in precipitates from L115 and L116 mutant brains, which correlated in size with high-molecular weight hCSPa oligomers ( Figure 2B). No ubiquitin-positive signals corresponding to CLN4 mutant hCSPa monomers, WT hCSPa monomers or oligomers were detected ( Figure 2B). Taken together, these data suggest that expression of CLN4 mutant hCSPa in fly brains reproduces the critical biochemical pathological features of post-mortem human CLN4 brains (Greaves et al., 2012;Henderson et al., 2016;Nosková et al., 2011).

Oligomerization of CLN4 mutant hCSPa is dose-dependent
Oligomerization of purified CLN4 mutant hCSPa proteins is dose-dependent in vitro (Zhang and Chandra, 2014). To test whether this is also the case in vivo, we doubled the gene dosage of WT and CLN4 mutant hCSPa by expressing two copies of the respective transgenes with a single elav-Gal4 driver. In comparison to the expression from one transgene, levels of lipidated and non-lipidated monomeric WT hCSPa increased~2.4 fold and~3.3 fold, respectively (p<0.01, Figure 2A,D-E,H-I). Yet, WT hCSP oligomer levels increased only 1.5-fold (p<0.005; Figure 2G) and remained low such that the oligomer/monomer ratio was not altered (p=0.9; Figure 2C).
Doubling expression disproportionally increased levels of non-lipidated hCSP-L115 and -L116 monomers 8.6-and 4.6-fold, respectively (p<0.002, Figure 2I). The increase of non-lipidated mutant monomers is unlikely due to a rate limiting effect on mechanisms mediating palmitoylation because levels of non-lipidated WT monomers increased much less than mutant monomers, even though overall levels of lipidated WT monomers were much higher ( Figure 2E).
Doubling gene dosage of hCSP-L115 and -L116 expression increased total overall protein levels 3.0 and 3.9-fold, respectively (p<0.002; Figure 1-figure supplement 1D). It also preserved the relative difference in overall protein levels between WT and CLN4 mutants ( Figure 2F). Since lipidated mutant monomer levels remained unaltered (p>0.1; Figure 2D), the increase in overall protein levels of CLN4 mutant hCSPa is essentially due to increased levels of oligomers and non-lipidated monomers (p<0.03; Figure 2C,E).
Levels of endogenous dCSP monomers were not affected by low or high expression of CLN4 mutant hCSPa ( Figure 2J). Notably, high-molecular weight oligomers of endogenous WT dCSP were not detectable after expression of CLN4 mutant hCSPa ( Figure 1-figure supplement 1C), which contrasts a previous study detecting small amounts of overexpressed GFP-tagged WT hCSPa in CLN4 mutant hCSPa oligomers (Greaves et al., 2012).

Oligomerization of neuronally expressed CLN4 mutant hCSPa precedes lethality
Elav-driven neuronal expression of WT or CLN4 mutant hCSPa from a single transgene had no effect on viability during development ( Figure 3A) and adult lifespan (not shown). However, doubling gene expression levels severely reduced viability during development of hCSP-L115 and -L116 animals in comparison to WT hCSPa expression (p<0.0001; Figure 3A). Adult L115-and L116-mutant escapers were initially sluggish in comparison to WT hCSPa but improved within~10 days and exhibited a normal lifespan (not shown), which may be due to a decrease in the elav-driven expression. There was no significant difference in the viability between L115 and L116 mutant animals (p=0.91; Figure 3A). The sharp drop in viability induced by 2-copy expression of hCSP-L115 and -L116 indicates a tight threshold of neurotoxicity, which might be linked to the more than 3.6-fold increase of Figure 3. CLN4 mutations cause dose-dependent lethality and eye degeneration. (A) Viability of animals expressing WT, L115, or L116 mutant hCSPa pan-neuronally from one or two transgenes (2x) with an elav driver (N ! 3, n ! 740). (B) Images of adult fly eyes expressing WT hCSPa or hCSPa-L116 with a GMR-Gal4 driver at 23˚C and 28˚C. (C) Semi-quantitative assessments of CLN4-induced eye phenotypes (N ! 9). Graphs display mean ± SEM. Statistical analysis used one-way ANOVA (A) and Kruskal-Wallis test (C); *, p<0.05; **, p<0.01; ***, p<0.001. DOI: https://doi.org/10.7554/eLife.46607.005 CLN4 mutant CSPa oligomers and/or the more than 4.6-fold increase of non-lipidated monomers ( Figure 2G,I).
Expression of CLN4 mutant hCSPa with the pan-neuronal nSyb-Gal4 driver also caused a dosedependent lethality, which was similar to that of the elav-Gal4 driver (data not shown). In contrast to the elav-driven expression, CLN4 mutant adult escapers were very sluggish and hardly moved. They were also unable to properly inflate their wings (Figure 1-figure supplement 1B) and died prematurely within 1-3 days.
To further examine potential degenerative effects in adults, we exclusively expressed hCSP-L116 in the eye by using the eye-specific GMR-Gal4 driver (Hay et al., 1997). High-level expression of WT hCSPa from a line with multiple transgene insertions (see Materials and methods) had essentially no effect at 23˚C or 28˚C in comparison to non-expressing control (p>0.9; Figure 3B-C). In comparison, expression of hCSP-L116 at 23˚C severely impaired the size, integrity, and pigmentation of the eye (p<0.0001; Figure 3B-C). Raising flies at 28˚C to increase Gal4 activity and thereby gene expression enhanced the severity of the degenerative phenotype of L116-mutant eyes (p<0.0001; Figure 3B-C). This enhancement correlates with the dose-dependent increase in hCSP-L116 oligomerization ( Figure 2C,G) but the higher temperature may also increase misfolding and/or oligomerization of mutant hCSPa.

CLN4 mutations impair hCSPa's synaptic localization
Previously, it has been suggested that CLN4 mutations may disrupt anterograde trafficking of CSPa due to either oligomer formation or impaired palmitoylation of monomers (Benitez et al., 2011;Greaves et al., 2012;Nosková et al., 2011). However, whether CLN4 mutations indeed affect the synaptic localization of CSPa remained unclear. To assess this, we triple-immunostained 3 rd instar larval NMJs with antibodies against hCSPa, endogenous dCSP marking SVs, and HRP marking the neuronal plasma membrane. In comparison to WT hCSPa, levels of mutant hCSP-L115 and -L116 were significantly reduced at synaptic boutons (p<0.013; Figure 4A,H). Expression of CLN4 mutant or WT hCSPa had no effect on synaptic levels of endogenous WT dCSP at NMJs (p>0.6; Figure 4A, I). Despite the reduced synaptic levels, mutant hCSPa still partially rescued the lifespan deficit of dcsp deletion mutants, although not nearly to the extent of WT hCSPa ( Figure 1D). This indicates that CLN4 mutant hCSPa proteins are at least partially functional.
While hCSPa co-localized with endogenous dCSP uniformly in the periphery of synaptic boutons ( Figure 4A,C), hCSP-L115 and -L116 were enriched in brightly stained clusters that were more distant from the presynaptic membrane ( Figure 4A,C). In comparison to control, CLN4 mutant hCSPa also accumulated abnormally in axons of segmental nerves ( Figure 4G) and the larval brain (p<0.009; Figure  . The size of CLN4 mutant hCSPa accumulations in the larval brain was heterogeneous, ranging from~500 nm in diameter to~3 mm. Occasionally, extreme accumulations of mutant hCSPa were observed at synaptic boutons of larval NMJs (Figure 4-figure supplement 1C). The occurrence of abnormal hCSP-L115 and -L116 accumulations was dose-dependent (not shown), which was most pronounced in axons of segmental nerves ( Figure 4G). Hence, CLN4 mutations reduce synaptic levels of hCSPa and cause a severe mislocalization.

Abnormal accumulations of CLN4 mutant hCSPa contain WT dCSP and ubiquitinated proteins
A fraction of the abnormal accumulations of hCSP-L115 and -L116 in axons (not shown) and somata ( Figure 4E and Figure 4-figure supplement 1A) contained significant amounts of endogenous WT dCSP. Since WT dCSP was not detected together with high-molecular weight oligomers of CLN4 mutant hCSPa on Western blots (Figure 1-figure supplement 1C), this indicates that CLN4 mutant hCSPa and WT dCSP may co-accumulate on intracellular membranes.
Theoretically, CLN4 mutant hCSPa could accumulate on ER, Golgi, or abnormal SVP membranes during secretory trafficking. Alternatively, mutant hCSPa could accumulate on endosomal membranes that are targeted for degradation. To test whether mutant hCSPa accumulations contain ubiquitinated proteins indicative of prelysosomal membranes, we used antibodies that exclusively detect ubiquitinated proteins but not monomeric ubiquitin (Fujimuro et al., 1994). Indeed, essentially all CLN4 mutant hCSPa accumulations were immunopositive for ubiquitinated proteins in   . In comparison to larvae expressing WT hCSPa, the amount of hCSPa accumulations containing ubiquitinated proteins was significantly increased in CLN4 mutants (p<0.001; Figure 4K). Hence, the abnormal accumulations of CLN4 mutant hCSPa are either enriched in ubiquitinated hCSPa oligomers and/or contain other ubiquitinated proteins that are potentially destined for protein degradation.
The co-immunostainings against hCSPa and ubiquitinated proteins also revealed a substantial increase in the levels of ubiquitinated proteins that was not associated with accumulations of CLN4 mutant hCSPa ( Figure 4D and Figure 4-figure supplement 1B). Brains of control (w 1118 ) and larvae expressing WT hCSPa contained a similar amount of ubiquitinated protein foci that were negative for hCSPa ( Figure 4L), indicating that expression of WT hCSPa exerts little to no effect on protein ubiquitination. In comparison, the amount of ubiquitinated protein foci that were not associated with mutant hCSPa was significantly increased in brains expressing hCSPa-L115 or -L116 (p<0.03; Figure 4L). Hence, CLN4 mutant hCSPa may directly or indirectly affect a step of protein homeostasis that that leads to excessive protein ubiquitination of unrelated proteins.

CLN4 mutations in fly and human CSP have similar effects
To validate the observed phenotypes of CLN4 mutant hCSPa and exclude that they are not an artifact of expressing mutant human proteins in fly neurons, we expressed CLN4 mutant dCSP2 (dCSP2) containing the mutations V117R and I118D, which are analogous to the mutations L115R and L116D in hCSPa ( Figure 1A). All fly transgenes were inserted at the same transgenic insertion site as the human transgenes. Elav-driven neuronal expression of dCSP2, dCSP-V117, or -I118D from a single transgene had no effect on viability during development at 23˚C (not shown) and 27˚C (p>0.4; Fig

CLN4 mutant hCSPa accumulates on prelysosomal endosomes
To further define the nature of the co-accumulations of CLN4 mutant hCSPa with ubiquitinated proteins, we tested a number of organelle markers for a potential colocalization. Mutant hCSPa accumulations did not colocalize with the endoplasmic reticulum (ER), cis-, or trans-Golgi complexes ( Figure 5A), which excluded a major defect in ER or Golgi trafficking.
A large fraction of mutant hCSPa co-accumulated with hLAMP1-GFP ( Figure 5B-C), which labels a heterogeneous population of organelles ranging from pre-degradative endosomal species to degradative lysosome (Cheng et al., 2018;Saftig and Klumperman, 2009;Yap et al., 2018). A similar large fraction of mutant hCSPa co-accumulated with hepatocyte growth factor regulated tyrosine kinase substrate (HRS, Figure 5C,E), which is a critical component of 'endosomal sorting complexes required for transport' (ESCRT) mediating the transition from early to late endosomes (Raiborg and Stenmark, 2009). Only a small fraction co-localized with coexpressed Rab5-GFP (not shown) or the autophagosomal marker ATG8/LC3-GFP ( Figure 5D). No co-localization was observed with lysosomal Spinster-GFP ( Figure 5B; Rong et al., 2011;Sweeney and Davis, 2002) or the late endosomal protein Rab7 (not shown; Guerra and Bucci, 2016).
Notably, endosomes accumulating mutant hCSPa did not co-localize with the co-expressed phosphatidylinositol 3-phosphate (PI 3 P) sensor FYVE-GFP ( Figure 5B), even though HRS requires PI 3 P for membrane association (Mayers et al., 2013;Raiborg et al., 2001a). Hence, this suggests that all PI 3 P binding sites may be occupied on the abnormal endosomes accumulating mutant hCSPa. Alternatively, PI 3 P may be absent from these endosomes, which would imply an abnormal retention of HRS. Taken together, these data suggest that CLN4 mutant hCSPa accumulates on prelysosomal endosomes that are inefficiently processed for lysosomal fusion.
Consistent with the colocalization of mutant hCSPa with ubiquitinated proteins at axon terminals ( Figure 4F), HRS also colocalized with mutant hCSPa at synaptic boutons of NMJs ( Figure 5F). This raised the possibility that hCSPa-positive endosomes may originate from synaptic terminals. Consistently, hCSPa-positive accumulations co-localized with SV-associated Synaptobrevin-GFP (Syb-GFP; Figure 5C,G) while a small subset co-localized with the synaptic plasma membrane protein Syntaxin 1A (not shown). The CSPa chaperone clients Dynamin and SNAP-25 (Sharma et al., 2011;Zhang et al., 2012) did not co-localize with hCSPa-positive endosomes (not shown).
CLN4 mutations cause various ultrastructural endo-membrane abnormalities in axons and neuronal somata Since CLN4 mutant hCSPa abnormally accumulates on prelysosomal endosomes, we used electron microscopy to detect potential ultrastructural defects in membrane trafficking. Expression of hCSPa-L115 and -L116 induced highly abnormal membrane structures in neuronal somata, the neuropil of the larval VNC, and axons of segmental nerves ( Figure 6). The most prominent and frequent abnormal structures were multilamellar 'membrane whirls' of various shapes and density that contain highly electron-dense membranes (arrowheads, Figure 6A-B,C-D,F-G). Like endosomal hCSPa-positive accumulations detected by confocal microscopy, membrane whirls were most frequently observed in the neuropil of the VNC and axons of segmental nerves ( Figure 6C-D). To a lesser degree, they were present in the cytoplasm of neuronal somata ( Figure 6A-B,F-G) and occasionally the nucleoplasm (not shown). Sporadically, whirls were found on opposite sides of plasma membranes of neighboring cells (arrowhead, Figure 6B).
Next to abnormal membranes structures, homogenous electron-dense accumulations of unknown nature were frequently present in cellular regions of CLN4 mutant larval brains (arrow, Figure 6B,H). A limiting membrane was not detectable for these EM-dense accumulations, which are reminiscent but not identical of granular osmophilic deposits (GRODs) seen in post-mortem human tissue of CLN4 patients (Anderson et al., 2013;Burneo et al., 2003;Nosková et al., 2011;Virmani et al., 2005). Since these structures were absent from the VNC neuropil and segmental nerves, they are unlikely a correlate for the abnormal endosomal accumulations of mutant hCSPa in axons.
Finally, a number of neurons that contained abnormal membrane structures also exhibited a severe fragmentation of the nuclear envelope and bloated Golgi cisternae ( Figure 6B), which together likely indicate a late stage of neuronal dysfunction and neurodegeneration (Nixon, 2006).

Loss of ESCRT function causes endosomal accumulations of endogenous dCSP but no oligomerization
To verify that WT CSP is normally trafficked through the endo-lysosomal pathway, we impaired ESCRT function by an RNAi-mediated knock down (KD) of tumor suppressor gene 101 (TSG101), which is part of the ESCRT-I complex acting downstream of HRS and required for late endosome formation (Doyotte et al., 2005;Razi and Futter, 2006). Neuronal KD of TSG101 caused a mislocalization of endogenous dCSP, which accumulated on HRS-positive endosomes in the neuropil and neuronal somata of the larval VNC ( Figure 5-figure supplement 1A). Abnormal dCSP accumulations were also present in central regions of synaptic boutons of larval NMJs (Figure 5-figure supplement 1B). Hence, WT CSP is likely being degraded though the endo-lysosomal pathway. This conclusion is consistent with the low amounts of lysosome-associated CSPa in human fibroblasts, N2A cells, mouse neurons, and lysosome-enriched fractions (Benitez and Sands, 2017;Chapel et al., 2013;Schrö der et al., 2007;Tharkeshwar et al., 2017). Notably, KD of TSG101 essentially mimicked the abnormal endosomal accumulation of CLN4 mutant hCSPa-L115 and -L116 with one important exception. In contrast to hCSPa-L115 and -L116 expression, KD of TSG101 did not cause the formation of high-molecular weight dCSP oligomers ( Figure 5-figure supplement 1C), indicating that CSP oligomer formation is independent of impaired ESCRT function. In contrast to a recent report showing that impaired lysosomal activity in mouse neurons decreases CSPa's palmitoylation (Sambri et al., 2017), impaired prelysosomal ESCRT function had no effect on the palmitoylation state of dCSP.
The CLN4 alleles hCSP-L115 and -L116 are hypermorphic gain of function mutations Several hypotheses regarding the genetic nature of the dominant CLN4 mutations have been suggested. Mutant hCSPa monomers and/or oligomers could act as dominant-negatives that progressively sequester WT hCSPa and deplete the functional hCSPa pool, which in turn could trigger neurodegeneration (Greaves et al., 2012;Nosková et al., 2011;Zhang and Chandra, 2014). Alternatively, CLN4 mutant alleles may induce toxicity via a different gain of function (GOF) mechanism (Greaves et al., 2012;Henderson et al., 2016;Zhang et al., 2012). Such a GOF mutation could increase a normal activity of the protein (hypermorphic mutation) or introduce a new activity (neomorphic mutation). In theory, either type of mutation could trigger neurotoxicity.
To genetically address the genetic nature of CLN4 mutations, we examined to what degree alterations of WT dCSP or hCSPa levels may affect L115 and L116 phenotypes. The rationale behind these genetic experiments is simple: lowering WT levels should enhance dominant-negative phenotypes, reduce hypermorphic phenotypes due to an increased normal activity, or have no effect on neomorphic phenotypes due to a new protein activity (Muller, 1932;Wilkie, 1994).
Reducing endogenous WT dCSP levels by expressing two transgenic copies of CLN4 mutant hCSP in heterozygous dcsp deletion mutants significantly suppressed the lethality induced by CLN4 mutations such that significantly more animals reached adulthood (p<0.02, Figure 7A). Conversely, increasing levels of WT hCSPa by co-expressing WT hCSPa with either one copy of hCSP-L115 or -L116 significantly reduced viability to~54% and 48%, respectively (p<0.001, Figure 7B-C). The lethality induced by co-expression of WT and CLN4 mutant hCSPa was significantly higher than the lethality induced by two copy expression of WT hCSPa (p<0.03, Figure 7B-C). Hence, the modulating effects of altered WT dCSP or hCSPa levels on L115-and L116-induced lethality essentially exclude the possibility that either mutation acts as a dominant-negative. Instead, these effects are consistent with the genetic characteristics of a hypermorphic GOF mutation (Muller, 1932;Wilkie, 1994).
Next, we determined whether altered WT CSP levels may modulate the tendency of CLN4 mutant hCSPa to form abnormal high-molecular weight oligomers. Reducing WT dCSP levels by expressing CLN4 mutant hCSPa in hetero-or homozygous dcsp deletion mutants significantly attenuated the levels of SDS-resistant hCSP-L116 and -L115 oligomers (p<0.05; Figure 7D,F). Levels of hCSP-L116 and -L115 oligomers were attenuated to a similar degree (p>0.5; Figure 7F). Conversely, increasing WT dCSP levels by co-expressing a WT transgene together with a single CLN4 mutant transgene increased the amount of hCSP-L115 and L116 oligomers (p<0.007, Figure 7D,I). Coexpression of WT hCSPa also significantly increased oligomerization of both hCSP-L116 and -L115 (p<0.008; Figure 7E,L). In comparison, high-level overexpression of WT hCSPa from two transgenes did not induce significant oligomerization ( Figures 7E and 2A,C).
In contrast to the modulatory effects of WT CSP on mutant hCSPa oligomer levels, reducing or even abolishing WT dCSP levels had no effect on the levels of lipidated monomeric hCSP-L115 and -L116 (p>0.4; Figure 7G). Coexpressing WT dCSP2 or hCSPa with CLN4 mutant hCSPa increased lipidated mutant hCSPa monomers (p<0.05; Figure 7J,M). Notably, coexpression of WT hCSPa increased lipidated hCSP-L115 and -L116 monomers to levels that were similar to those of two copy WT hCSPa expression ( Figure 7M). This indicates that WT hCSPa may stabilize CLN4 mutant hCSPa monomers.
Levels of non-lipidated monomeric CLN4 mutant hCSP were unaffected by reducing the gene dosage of endogenous WT dCSP2 (p>0.3; Figure 7H). However, coexpression of either WT dCSP2 or hCSPa with CLN4 mutant hCSPa increased levels of non-lipidated mutant hCSPa monomers (p<0.05; Figure 7K,N). At least the increase induced by WT hCSPa was non-additive since it was significantly larger than the increase induced by doubling WT hCSPa expression (p<0.013; Figure 7N).
Hence, increased levels of WT CSP may outcompete CLN4 mutant hCSPa for palmitoylation or, alternatively, promote its depalmitoylation.
Finally, we tested whether altering levels of endogenous dCSP also affects other phenotypes of the CLN4 fly model. Indeed, reducing dCSP levels by one gene copy significantly suppressed the endosomal accumulation of both hCSP-L115 and -L116 in larval VNCs (p<0.05; Figure 8A-B). Reducing dCSP gene dosage also decreased the amount of accumulating ubiquitinated proteins in the larval VNC (p<0.04; Figure 8C). However, reduced dCSP levels significantly enhanced the already reduced synaptic protein levels of hCSP-L115 and -L116 at larval NMJs (p<0.03; Figure 8D). The underlying cause for the latter effect is unclear.
Taken together, the modulating effects of altering WT fly and human CSP levels on L115-and L116-induced phenotypes in viability, mutant hCSPa oligomerization, accumulation on endosomes and ubiquitinated proteins suggest that the hCSP-L115 and -L116 alleles are hypermorphic gain of function mutations that increase an intrinsic activity. This conclusion is consistent with a recent study showing similar gene dosage effects of WT CSPa on hCSP-L115R phenotypes in mouse CSPa-deficient fibroblasts (Benitez and Sands, 2017). In addition, these findings uncover a correlation between the reduced viability of CLN4 mutant flies, the degree of mutant hCSPa oligomerization and the failure of degrading oligomers.

Partial loss of CSPa's chaperone partner Hsc70-4 suppresses CLN4induced phenotypes
The hypermorphic nature of CLN4 mutations raised the possibility that at least some of the induced phenotypes are caused by an abnormally increased co-chaperone activity of CSP with the molecular chaperone Hsc70, which normally ensures efficient SV exo-and endocytosis by likely chaperoning SNARE proteins and Dynamin (Chandra et al., 2005;Nie et al., 1999;Sharma et al., 2012;Sharma et al., 2011;Zhang et al., 2012). To address this possibility, we tested whether altered levels of SNARE proteins or Hsc70 may modify the eye phenotypes induced by GMR-driven expression of CLN4 mutant hCSPa ( Figure 9A).
To further test whether Hsc4 may at least in part contribute to CLN4 phenotypes, we examined effects of altered Hsc4 levels on the lethality induced by pan-neuronal expression of CLN4 mutant hCSPa. Reducing the gene dosage of hsc4 by expressing mutant hCSP-L115 or L116 in  heterozygous hsc4 D356 deletion mutants significantly suppressed their lethality (p<0.0001; Figure 9F). Hence, Hsc4 contributes to the toxicity of CLN4 mutations.
Next, we examined to what degree altered Hsc4 levels may attenuate the oligomerization of CLN4 mutant hCSPa. Hsc4 OE had no significant effects on the amount of hCSP-L115 and -L116 oligomers, monomers and overall protein levels (p>0.05; Figure 9-figure supplement 1A-D). However, reducing Hsc4 levels by expressing mutant hCSP-L115 or L116 in heterozygous hsc4 D356 deletion mutants significantly reduced the amount of hCSP-L115 and -L116 oligomers (p<0.02; Figure 9B-C,G) but had no differential effects (p>0.8). It also reduced the levels of lipidated and non-lipidated hCSP-L115 and -L116 monomers, as well as overall protein levels (p<0.05; Figure 9H-J). The overall reduction of CLN4 mutant hCSPa could have been due to a general dependence of hCSPa expression on normal Hsc4 levels. However, this is unlikely the case since reducing Hsc4 levels had no effect on the levels of monomeric WT hCSPa ( Figure 9K), which was consistent with the normal dCSP levels observed in hsc4 deletion mutants (Bronk et al., 2001).
Reducing Hsc4 levels also suppressed the amount of hCSP-L115 and -L116 protein accumulations on endosomes in the larval VNC (p<0.05; Figure 9L and Figure 9-figure supplement 1E). It also suppressed the amount of ubiquitinated protein accumulations induced by hCSP-L116 (p<0.05) but had no effect on L115-induced levels (p=0.8; Figure 9M). Reducing hsc4 gene dosage had no effect on the synaptic levels of hCSP-L116 at larval NMJs (p=0.7; Figure 9N). However, it further lowered the synaptic levels of hCSP-L115 (p<0.01; Figure 9N). The underlying cause for these differential effects is unclear.
In conclusion, the suppression effects of a reduced hsc4 gene dosage on the oligomerization and abnormal accumulation of CLN4 mutant hCSPa on endosomes suggest that Hsc4 significantly contributes to this pathology. Since HA-tagged Hsc4 colocalized with the abnormal accumulations of hCSP-L115 and -L116 ( Figure 9O), it may interfere with the prelysosomal trafficking and subsequent degradation of mutant hCSPa. Consistently, lower Hsc4 levels reduced the overall protein levels of mutant but not WT hCSPa ( Figure 9J-K).

Discussion
To better understand pathological mechanisms underlying CLN4, we developed two Drosophila models by expressing either CLN4 mutant hCSPa or dCSP in neurons. Neuronal expression of either mutant hCSPa or dCSP mirrored key pathological features of post-mortem human CLN4 brains, including reduced levels of palmitoylated hCSPa monomers and excessive formation of oligomers. Consistent with the idea that ubiquitination is likely a consequence of oligomerization, only ubiquitinated mutant hCSPa oligomers were detected but no ubiquitinated monomers. In general, the L115 mutation had stronger effects than L116, which has also been observed in post mortem brains and mammalian cell cultures (Diez-Ardanuy et al., 2017;Greaves et al., 2012;Henderson et al., 2016;Zhang and Chandra, 2014). Accordingly, expression of CLN4 mutant hCSPa or dCSP in fly neurons provides a valid CLN4 disease model.

Dose-dependency of phenotypes in the Drosophila CLN4 model
To assess the dose-dependency of CLN4 phenotypes in the fly model, we controlled the gene dosage of transgenically expressed CLN4 mutant hCSPa by neuronally expressing either one or two mutant transgenes in the presence of two endogenous WT dCSP gene copies.
Single copy expression of CLN4 mutant hCSPa did not cause detectable degeneration or lethality, even though significant levels of ubiquitinated hCSPa oligomers and abnormal prelysosomal accumulations of hCSPa were present. However, doubling the gene dosage reduced the viability of CLN4 mutant flies, which is likely due to neurodegeneration as indicated by fragmentated nuclear envelopes and 'bloated' Golgi cisternae of larval neurons ( Figure 6B). The significant drop in viability indicated a sharp threshold of neurotoxicity, which was correlated with an increase in the levels of hCSPa oligomers and prelysosomal accumulations. Consistent with a neuronal failure, adult escapers showed reduced locomotor activity.
A similar dose-dependent effect was observed for the eye phenotype induced by GMR-driven expression of CLN4 mutant hCSPa. The build-up of these pathological features in the CLN4 fly model agrees with the assumed general concept of late onset neurodegenerative diseases, which typically show a progressive accumulation of protein aggregates and defects in protein homeostasis over a long period that is met by a declining age-dependent capacity of protein homeostasis (Labbadia and Morimoto, 2015).

CLN4 mutations impair hCSPa's synaptic localization
Previous studies speculated that CLN4 mutations may disrupt the trafficking and/or SV localization of hCSPa (Benitez et al., 2011;Greaves et al., 2012;Nosková et al., 2011), mainly because they reside in CSP's palmitoylated CS domain. The CLN4 fly model confirms this prediction in two ways: First, CLN4 mutations reduced the synaptic levels of mutant hCSPa. Second, they caused abnormal accumulations of mutant hCSPa in axons and neuronal somata. Analogous CLN4 mutations in fly dCSP had similar effects excluding the possibility that these defects are the consequence of expressing a human protein in fly neurons.
In principle, there are two possibilities how CLN4 mutations may mislocalize CSP: 1) CLN4 mutations may primarily impair palmitoylation of the CS domain, which is required for ER and Golgi exit, and for a stable association with SVPs and/or SVs (Greaves and Chamberlain, 2006;Greaves et al., 2008;Ohyama et al., 2007;Stowers and Isacoff, 2007). However, this possibility is unlikely the case since mutant hCSPa was not detectable in the ER or Golgi of fly neurons. In addition, levels of non-lipidated CLN4 mutant hCSPa were similar to WT hCSPa when expressed from one transgenic copy. 2) CLN4 mutations may primarily induce oligomerization by exaggerating the dimerization properties of the CS domain (Xu et al., 2010) such that self-association leads mostly to misfolded high-molecular weight oligomers. Notably, oligomerization of CLN4 mutant hCSPa can be suppressed by alanine or leucine substitutions of a cluster of palmitoylated cysteines (C122-125) located next to the L115/116 mutations (Diez-Ardanuy et al., 2017), which is consistent with the idea that CLN4 mutations primarily induce oligomerization.
In fly neurons, the oligomerization of mutant hCSPa is apparently triggered after exiting the Golgi, either during its association with SVPs, or after arriving at axon terminals and associating with SVs. The transition of SVP cargo onto SVs or endosomes is not well understood but in principle SVPs can fuse with SVs, the synaptic plasma membrane, and endosomes (Rizzoli, 2014). In contrast, SVs are linked to endosomes either by ultrafast endocytosis, or the ability of freshly endocytosed individual SVs to fuse with endosomes (Hoopmann et al., 2010;Soykan et al., 2017;Watanabe and Boucrot, 2017;Watanabe et al., 2013a;Watanabe et al., 2013b). Mutant hCSPa oligomers are potentially re-routed from both SVPs and SVs because some mutant hCSPa colocalizes with SVs at axon terminals. Nevertheless, re-routing of ubiquitinated oligomers is likely the main factor reducing synaptic levels of CLN4 mutant hCSPa.
The ubiquitination of CLN4 mutant hCSPa oligomers likely indicates that they are misfolded and thus targets for ubiquitination and degradation like other misfolded proteins (Wang et al., 2017). Consistently, only ubiquitinated mutant hCSPa oligomers but no ubiquitinated monomers were detected in fly neurons. Since WT dCSP ( Figure 5-figure supplement 1) and hCSPa are mostly degraded by lysosomes (Benitez and Sands, 2017;Chapel et al., 2013;Sambri et al., 2017;Schrö der et al., 2007;Tharkeshwar et al., 2017), one expects that ubiquitinated mutant hCSPa oligomers are sorted into the endolysosomal pathway if they remain associated with membranes. This appears to be the case since the majority of abnormal mutant hCSPa accumulations in fly neurons represented prelysosomal endosomes, which were defined by the colocalization of ubiquitinated proteins and the prelysosomal markers LAMP1-GFP and HRS. Finally, the suppression of mutant hCSPa phenotypes by reduced dCSP or Hsc4 levels revealed a strong link between the oligomerization of mutant hCSPa, its re-routing onto prelysosomal membranes, and its induced lethality (Figures 7-8). Hence, these data suggest that CLN4 mutations may primarily induce oligomerization of membrane-associated hCSPa, which causes ubiquitination and re-routing into lysosomal pathways.

CLN4 mutations impair prelysosomal trafficking of hCSPa
Axonal prelysosomal endosomes containing mutant hCSPa are likely retrogradely trafficked for lysosomal processing since degradative lysosomes containing cathepsin B/D are essentially absent from distal axons (Cai et al., 2010;Cheng et al., 2018;Gowrishankar et al., 2015). Both the axonal trafficking as well as the maturation of prelysosomal endosomes containing mutant hCSPa are potentially impaired. A physical defect in axonal trafficking is indicated by the large size of the hCSPa accumulations and by the frequent occurrence of large and abnormal EM-dense membrane structures in axons of CLN4 mutants, which have the potential to 'clog' axons.
A reduced efficiency in the maturation of mutant hCSPa-positive prelysosomal endosomes is indicated by their abnormally large size and by their failure to co-label with the PI 3 P lipid reporter FYVE-GFP. The latter may be due to two alternatives: First, hCSPa-positive endosomes may have normal PI 3 P levels but most of it is bound by PI 3 P-binding proteins. At least in part, this could be facilitated by HRS (Raiborg et al., 2001a), which strongly co-labels with hCSPa accumulations. Alternatively, PI 3 P could be depleted from these endosomal membranes, which then would imply that HRS is abnormally retained.
Since only a small number of mutant hCSPa accumulations co-localized with the early endosome marker Rab5 but essentially no accumulations colocalized with the late endosomal or lysosomal markers Rab7 and spinster, it is possible that mutant hCSPa may impair the transition from early to late endosomes.
The frequent occurrence of abnormal, multi-laminar structures containing EM-dense membranes and abnormal autophagosome-like structures further indicate that CLN4 mutations impair steps of membrane trafficking in axons and somata. Since similar ultrastructural defects were observed after disruption of ESCRT function (Doyotte et al., 2005;Lee et al., 2007), the abnormal endosomal membranes of CLN4 mutants suggest a defect in prelysosomal trafficking. This is further supported by the accumulation of mutant hCSPa on prelysosomal membranes including ATG8/LC3-GFP positive autophagosomes. Alternatively, at least some of the abnormal membrane structures of CLN4 mutants could be the consequence of defects in the unconventional secretion of misfolded cytosolic proteins, which is facilitated by CSP proteins at least in non-neuronal cells (Xu et al., 2018).
The CLN4 fly model also provides evidence for a general protein ubiquitination defect. Both of the CLN4 mutations caused an increase in ubiquitinated protein accumulations in axons and neuronal somata that were not associated with accumulations of CLN4 mutant hCSPa. Consistent with an altered protein homeostasis, significant proteomic and lipidomic changes were found in CLN4 mutant post-mortem brains (Henderson et al., 2016). Even though CLN4 mutant CSPa's ability to stimulate the ATPase activity of Hsc70 progressively deteriorates in vitro (Zhang and Chandra, 2014), the excessive ubiquitination phenotype of the fly model is unlikely due to a reduced CSP/ Hsc70 chaperone function since reducing WT dCSP levels suppressed the ubiquitination effects. Hence, the excessive ubiquitination may be a secondary consequence of the prelysosomal trafficking defects induced by CLN4 mutations.

CLN4 alleles genetically resemble hypermorphic gain of function mutations
Previously, it has been suggested that the progressive oligomerization of CLN4 mutant hCSPa may cause a dominant-negative effect by sequestering and depleting WT CSPa (Greaves et al., 2012;Henderson et al., 2016;Nosková et al., 2011;Zhang and Chandra, 2014). However, CLN4 mutant hCSPa had no effect on the synaptic and overall levels of endogenous WT dCSP in the fly model. In addition, decreasing or even abolishing WT dCSP reduced oligomer formation of mutant hCSPa. Both effects are inconsistent with a dominant-negative effect of L115 and L116 mutations. In addition, genetic analysis of other phenotypes found no evidence supporting dominant-negative effects in the CLN4 fly model. Instead, this analysis suggests that most of the observed phenotypes are due to gain of function effects.
Reducing or abolishing WT dCSP levels suppressed the oligomerization and endosomal accumulation of CLN4 mutant hCSPa, the elevated protein ubiquitination, and the premature lethality. Increasing WT dCSP or hCSPa levels enhanced the oligomerization and lethality of CLN4 mutations. Taken together, these results suggest that the respective phenotypes are due to a hypermorphic gain of function mutation that increases an activity of hCSPa. A similar dependence of CLN4 phenotypes on increased WT CSPa levels was observed in primary fibroblasts of CSPa-deficient mice after co-expressing WT CSPa with CSPa-L115R, which increased levels of CSPa oligomers and lysosomal lysotracker signals (Benitez and Sands, 2017).
Notably, there are also CLN4 phenotypes of the fly model that could not be genetically classified in an unambiguous manner. These phenotypes include the effects of altered dCSP levels on monomeric and synaptic levels of CLN4 mutant hCSPa, which were either not affected or further reduced by lower dCSP levels.
The primary effect of the gain of function mutation induced by L115 and L116 is not known. However, the fly model revealed a correlation among the degree of mutant hCSPa oligomerization, accumulation on endosomes, and lethality. Accordingly, it seems possible that CLN4 mutations primarily drive the excessive oligomerization of hCSPa by exacerbating the known dimerization properties of the CS domain (Xu et al., 2010). Such an effect would explain the excessive routing of the likely misfolded oligomers into prelysosomal pathways and the depletion of mutant hCSPa at synapses. The impaired processing of the endosomes containing accumulating mutant hCSPa is then potentially a secondary effect of misfolded oligomers.
Another possibility for the gain of function effect arises from the recently described role of hCSPs on late endosomes for an unconventional secretion pathway, termed misfolding-associated protein secretion (MAPS) (Xu et al., 2018). Even though it is unclear whether neuronal hCSPa participates in MAPS, the gain of function effect could exacerbate the trafficking of hCSPa onto late endosomes for MAPS after exiting the Golgi. Together, an excessive routing onto endosomes and oligomerization of mutant hCSPa may then impair membrane trafficking pathways requiring late endosomes like the lysosomal and MAPS pathways.
The dependence of key CLN4 phenotypes on WT CSP levels paralleled the modulatory effects of hCSPa's main interaction partner, Hsc70. Similar to reducing WT dCSP levels, reducing levels of Hsc4 by one gene copy suppressed the lethality and eye phenotypes induced by CLN4 mutant hCSPa, its oligomerization and endosomal accumulation. These modulatory effects are driven either by an individual activity of Hsc70 itself, or by a synergistic activity of the CSP/Hsc70 complex. Neither of these activities are necessarily localized to SVs. In support of the latter, co-expressed HAtagged Hsc70-4 co-localized with the accumulations of CLN4 mutant hCSPa on endosomes, which may interfere with the dissociation of clathrin coats that are anchored by the ESCRT component HRS (Raiborg et al., 2001b). Alternatively, the respective Hsc70 activity may be associated with endosomal microautophagy (Sahu et al., 2011;Uytterhoeven et al., 2015) or the unconventional secretion of misfolded cytosolic proteins (Xu et al., 2018). At least for this fly model, an activity of Hsc70 associated with chaperone-mediated autophagy (Kaushik and Cuervo, 2012) can be excluded since the critical lysosomal translocator LAMP-2 is absent in flies (Uytterhoeven et al., 2015).

Concluding remarks
This initial study of the CLN4 fly model revealed a number of novel and important insights into the pathology of CLN4, most notably, the unexpected genetic nature of CLN4 mutations, the mislocalization of mutant hCSPa, a general ubiquitination defect, and prelysosomal processing defects. The modulating effects of altered WT CSP or Hsc70 levels on key CLN4 phenotypes provided a strong correlation between the premature lethality of CLN4 mutant flies, oligomerization and accumulation of mutant hCSPa on prelysosomal endosomes. Accordingly, the neurotoxicity of CLN4 mutant hCSPa may be at least in part due to the formation of ubiquitinated hCSPa oligomers that then progressively accumulate on prelysosomal endosomes and interfere with their processing for lysosomal fusion and potentially their retrograde axonal trafficking.
The pathological mechanisms underlying CLN4 appear to be different from those of most other NCLs, which are typically due to loss of function mutations in genes that mediate lysosomal function, ER-lysosomal trafficking, or protein lipidation (Cárcel-Trullols et al., 2015;Cooper et al., 2015;Cotman et al., 2013;Mole and Cotman, 2015;Warrier et al., 2013). As such, the pathological mechanisms underlying CLN4 appear similar to those of 'classical' neurodegenerative diseases that are induced by a progressive build-up of protein oligomers/aggregates that then leads to a failure of protein homeostasis and/or lysosomal pathways (Abramov et al., 2009;Labbadia and Morimoto, 2015;Lansbury and Lashuel, 2006;Muchowski and Wacker, 2005;Neefjes and van der Kant, 2014). While we are just beginning to understand the complicated nature of CLN4, the fly model provides a valuable tool for future work to dissect pathological mechanisms underlying CLN4. Continued on next page Transgenic animals were generated by fC31-based integration (Bischof et al., 2007) of the UAS transgenes into the attP site at 22A2 on chromosome 2 (2L:1476459..1476459). pBID-UASC-CSP-x plasmids were injected into y 1 M[vas-int.Dm]ZH-22A w*; M[3xP3-RFP.attP']ZH-22A (BDSC #24481) embryos (Rainbow Transgenic Flies, Camarillo, CA). At least two independent recombinant strains were obtained for each transgene and out-crossed to exchange non-recombinant chromosomes. The 3xP3-RFP cassette was removed by loxP-mediated recombination as described (Bischof et al., 2007). Homozygous strains containing UAS transgenes were established in a genetic background representing WT control (w 1118 ) and dcsp deletion mutants (w 1118 ; dcsp X1 ).

Materials and methods
Crosses

Immunostainings
Wandering 3 rd instar larvae were dissected in Sylgard-coated dishes containing cold Ca 2+ free HL3 solution (in mM: 70 NaCl, 5 KCl, 20 MgCl 2 , 10 NaHCO 3 , 5 Trehalose, 115 sucrose, 5 HEPES, pH 7.2). Dissected larvae were fixed for 20 (VNC) or 45 min (NMJs) in 4% formaldehyde solution (Electron Microscopy Sciences, Hatfield, PA) in phosphate-buffered saline (PBS), pH 7.3 at room temperature (RT). PBS (pH 7.3) supplemented with 0.2% Triton X-100 (PBST) was used for immunostainings of larval NMJs while PBS (pH 7.3) supplemented with 0.4% Triton X-100 was used for immunostainings of larval VNCs. After washing 3 times for 10 min in PBST at RT, larvae were incubated with primary antibodies diluted in PBST overnight at 4˚C, washed 3x for 15 min at RT, incubated with secondary antibodies diluted in PBST for 2 hr at RT or overnight at 4˚C and finally washed 3x with PBST for 15 min at RT. Confocal images were acquired the same day, otherwise preparations were post-fixed.
The following antibodies and dilutions were used: rabbit anti-CSPa
For quantification of fluorescence signals, control and mutant larvae were dissected in the same dish such that fixation and antibody incubation were performed identically. All samples were imaged with the same laser settings. Fluorescence intensity per area was determined from a region of interest (ROI) encompassing single synaptic boutons by using Image J Software.
For quantification of CSP accumulations in VNC, z-projections with maximal intensity of optical sections were generated and cropped to a defined volume of 60 Â 80 Â 45 mm (216,000 mm 3 ). These image stacks represented the dorsal-most part of hemi-segments A4 and A5. After thresholding for background subtraction, ROIs were drawn around all visible CSP-positive punctae; for those appearing in multiple optical sections only the section with the brightest signal was used. ROI areas were compiled to compare the cumulative area among genotypes.

Hydroxylamine treatment
Adult heads were homogenized in 2% SDS, 100 mM DTT, 60 mM Tris pH 6.8, and then treated with 0.5M hydroxylamine pH 7.0 or 0.5 M Tris pH 7.0 (final concentration) for 24 hr at RT before being diluted 3-fold using 1x Laemmli buffer. Samples were boiled for 5 min before use for SDS-PAGE.

Immunoprecipitation
Young adult flies (1-3 day post-eclosion) were flash frozen in liquid N 2 and stored at À80˚C. Heads were enriched by brushing frozen heads on double metal sieves under liquid N 2 . Approximately 100 mL of heads were homogenized in 500 mL IP buffer containing: 1% Triton X-100, 30 mM Tris-HCl (pH 7.2), 150 mM KCl, 0.5 mM MgCl 2 . Homogenates were repeatedly centrifuged at 22,000 rpm to remove insoluble debris. The cleared supernatant was incubated with rabbit anti-CSPa at 1:200 (f.c.) for 1 hr on a rotator at 4˚C and then incubated with 50 mL protein A/G PLUS-Agarose (#sc-2003, Santa Cruz Biotechnology, Dallas, Texas) for 2 hr at 4˚C. Agarose beads were spun down with a tabletop centrifuge at 2000 g and the supernatant was removed and stored. The beads were washed 4x with IP buffer by repeated resuspension and centrifugation. Beads were finally resuspended in Laemmli buffer, boiled for 5 min, centrifuged for 2 min and transferred to a fresh tube for SDS-PAGE.

Quantification of eye phenotypes
This analysis was essentially done as described by others with some modifications (Papanikolopoulou and Skoulakis, 2011;Santa-Maria et al., 2015). Briefly, eye phenotypes of 1-2 day old adults were digitally imaged using a stereomicroscope. A semi-quantitative phenotype assessment was achieved by serially coding the obtained images and blind scoring by several naive researchers. Each individual eye was given a relative roughness score in comparison with known scoring classes: (1) normal WT-like eye. (2) Slightly 'rough' eye surface, slight dis-colorization, and/or least reduction in eye size. (3) Rough eye surface, significant dis-colorization, disorganized ommatidia, and/or reduced in size. (4) Rougher eye surface; loss of organized ommatidia, malformed and fused; more pronounced reduction in eye size. (5) Roughest eye surface, most pronounced loss of organization, and/or most pronounced reduction in eye size. Scores were collected for at least eight individual flies per genotype.

Data and statistical analysis
Data from at least three independent animals or experimental trials were used for statistical analysis. Data are represented as mean, and error bars represent SEM. Gaussian distribution of data was assessed using a D'Agostino and Pearson omnibus or Shapiro-Wilk normality test using Prism software (GraphPad Software). Statistical significance was assessed by either a paired or unpaired t test, Mann-Whitney, or one-way ANOVA (Kruskal-Wallis for non-parametric data) test with appropriate post-hoc tests using Prism software). P values < 0.05,<0.01, and <0.001 are indicated in graphs with one, two, and three asterisks, respectively.