NF-κB disinhibition contributes to dendrite defects in fly models of neurodegenerative diseases

Neurodegenerative diseases (NDs) are late-onset neurological disorders characterized by gradual loss of neuronal ability to maintain morphological integrity and functionality. Representative NDs include Alzheimer’s disease (AD), Parkinson’s disease (PD), polyglutamine (polyQ) diseases, and amyotrophic lateral sclerosis (ALS). The progressive impairment in various neuronal functions may be caused by the loss of function and/or gain of toxicity of associated proteins in the disease condition. Regarding the loss-of-function aspect, extensive studies have identified numerous proteins (either disease-responsible proteins or mediators of pathogenesis) whose loss of function contributes to various neuropathic features in NDs. In contrast, the gain-of-toxicity aspect has been studied with a primary focus on how accumulated misfolded proteins (e.g., amyloid β in AD, α-synuclein in PD, and polyQ proteins in polyQ diseases) gain protein toxicity in neurons upon genetic mutations or environmental factors (Chung et al., 2018). However, the involvement of additional pathogenic mediators for the gain of toxicity during disease progression has not been well studied.

Because neuronal cell death common to NDs occurs mainly at the late stages of disease progression, early neuropathic features such as morphological changes have been extensively studied to explain how disease initiates and progresses before massive neuronal cell death (Kweon et al., 2017). In line with this, dendrite defects have been frequently observed in patients and in animal models for AD (Knobloch and Mansuy, 2008; Spires et al., 2005), PD (Blumenstock et al., 2017; McNeill et al., 1988), polyQ diseases (Graveland et al., 1985; Lee et al., 2011), and ALS (Gorrie et al., 2014). Previous studies have identified several molecules whose loss of function contributes to dendrite defects in animal models for NDs originating from gain of toxicity of accumulated misfolded proteins. For example, May et al. (2014) reported that the amount of soluble UNC119 regulating dendrite morphology was reduced by poly-Gly-Ala aggregates produced by the expanded C9orf72 nucleotide repeats, which led to dendrite defects in primary hippocampal and cortical neurons of ALS rat models. In addition, misfolded polyQ proteins induce dendrite defects in fly models for polyQ diseases by perturbing the CREB-binding protein-CREB3L1/CrebA pathway (Chung et al., 2017) or Forkhead box, subgroup O (Kwon et al., 2018).

However, whether the gain of toxicity of additional mediators (hereafter called toxic mediators) contributes to dendrite defects in NDs remains to be elucidated. We therefore searched for toxic mediators by categorizing candidates using the following criteria: first, their abundance and/or functions are aberrantly up-regulated in the disease condition; second, their down-regulation or reduced expression is able to ameliorate dendrite defects in NDs. Using these criteria, we identified NF-κB as a novel toxic mediator of dendrite defects in animal models for two representative NDs, Machado-Joseph disease/spinocerebellar ataxia type 3 (MJD/SCA3) and ALS. Notably, several previous studies suggested that NF-κB harbors deleterious effects within the neuronal context in a cell-autonomous manner (Pizzi et al., 2002; Sarnico et al., 2009). For proper function of neurons, the latent toxicity of NF-κB should be under tight control to avoid unwanted damage from it. However, it remains elusive in what conditions and how neurons lose their control for NF-κB toxicity and what pathophysiological features in neuronal disorders involve dysregulation of NF-κB toxicity. Here, using Drosophila as a model system, we present disinhibition of NF-κB induced by perturbed Dredd/Caspase-8–mediated endoproteolytic cleavage of it as a novel pathogenic mechanism of action underlying dendrite defects in animal models for NDs and identify Tup and Pros as downstream mediators of NF-κB–induced neuronal toxicity.

Among the NDs, we focused on MJD, a neurological disorder belonging to the polyQ disease family that is caused solely by the genetic mutation of a Q-encoding CAG repeat expansion resulting in protein toxicity (Kawaguchi et al., 1994). To identify toxic mediators in MJD, we first selected candidates using the previous RNA-sequencing data collected from mouse MJD models (GSE117028, Zeng et al., 2018; GSE108069, Toonen et al., 2018) and patient-derived induced pluripotent stem cell neurons expressing toxic MJD proteins (GSE96826; Chuang et al., 2019). Based on our assumption that toxic mediators should not be down-regulated under the disease condition for their gain of toxicity, we selected 6,748 non–down-regulated genes (Fig. 1 A). Of them, 5,995 genes were identified to be Drosophila homologous genes, among which 99 genes were predicted to be involved in dendrite morphogenesis based on gene ontology biological process annotation. Among these 99 genes, we selected only those that can act as a transcription factor (TF), since the formation and maintenance of complex dendrite morphology require sophisticated regulation of multiple genes, which is commonly conferred by TFs (Jan and Jan, 2010). Indeed, as shown in our previous studies (Chung et al., 2017; Kwon et al., 2018), perturbed TF activities result in aberrant regulation of the many downstream genes that are important in dendrite morphology and, therefore, contribute to polyQ-induced dendrite defects. Besides, enrichment analysis of gene ontology molecular functions (GOMFs) for the selected 99 genes revealed that transcriptional activity was the most strongly associated (Fig. 1 B). Among the 15 genes with TF activity selected, only 12 were identified to have available RNAi lines. Thus, we focused on 12 genes for subsequent experimental validation.

We next experimentally validated whether reduced expression of each of the 12 selected genes could ameliorate MJD polyQ–induced dendrite defects. To this end, we individually knocked down each gene in Drosophila class IV dendritic arborization (C4da) neurons expressing toxic MJD polyQ proteins (MJD-78Q) and compared the consequent dendrite phenotypes to those caused by MJD-78Q alone. Previous studies showed that MJD-78Q expression in C4da neurons induces obvious dendrite defects characterized by highly reduced terminal dendrite branches (Lee et al., 2011). Among the 12 genes examined, knockdown of only 2 (Relish, a Drosophila homologue of NF-κB, and Ftztranscription factor1 [Ftz-f1], a Drosophila homologue of LRH-1 and SF-1) resulted in significant restoration of MJD-78Q–induced dendrite defects (Fig. 1, C and D; and Fig. S1 A). Because knockdown of Relish showed the largest effect, we focused on characterizing its potential role as a toxic mediator of dendrite defects. Furthermore, two additional Relish RNAi transgenes with validated efficiencies (Fig. S1 B) also showed the similar restoration effects in MJD polyQ–induced dendrite defects (Fig. S1, C and D). Because this restoration effect of Relish RNAi is not due to alteration in the amount of MJD-78Q proteins (Fig. S1, E–H), it can be suggested that the restoration effect of Relish RNAi may due to reduced amount of Relish in toxic MJD polyQ–expressing neurons.

To qualify as a toxic mediator, the function and/or abundance of Relish should be up-regulated by toxic MJD polyQ proteins. Thus, we first examined whether the molecular function of Relish as a TF is up-regulated by toxic MJD polyQ proteins in fly brain. To this end, we compared mRNA expression levels of well-known Relish target genes (CecA1, CecA2, CecB, AttA, AttB, and DptB; De Gregorio et al., 2002; Tanji et al., 2010) between control fly heads and those panneuronally expressing MJD-78Q. The mRNA levels of all target genes examined were significantly increased by MJD-78Q (Fig. 1 E), indicating that the molecular function of Relish is up-regulated by MJD-78Q. We then examined whether the amount of Relish is increased by MJD-78Q as well. For this, through Western blot analysis, we compared the amounts of endogenous Relish between control fly heads and those panneuronally expressing MJD-78Q. The amount of endogenous Relish was significantly (P < 0.05) increased in brains expressing MJD-78Q compared with control brains (Fig. 1, F and G). Taken together, these data suggest that Relish is a novel toxic mediator of polyQ-induced dendrite defects.

Next, we questioned how Relish abnormally gains the toxicity that contributes to dendrite defects induced by the expression of MJD polyQ proteins. To address this, we first examined whether overexpression of Relish has toxic effects on dendrite morphology. Unexpectedly, overexpression of Relish in C4da neurons did not show noticeable changes in dendrite morphology, determined by the number of branch points and dendritic complexity (Fig. 2, A and B; and Fig. S2 A). Because endoproteolytic cleavage of Relish by Dredd (a Drosophila homologue of Caspase-8) is known to be required for activation of its TF function (Stöven et al., 2000; Fig. S2 B), we then examined whether cleaved fragments of Relish have toxic effects on dendrite morphology. N-terminal (Rel68) and C-terminal (Rel49) fragments of Relish, generated by endoproteolytic cleavage, contain Relish homology domain with TF activity and an ankyrin repeat domain, respectively (Fig. S2 C; Stöven et al., 2000). Overexpression of Rel68 in C4da neurons significantly (P < 1.0 × 10⁻⁴) reduced both the number of dendrite branch points and dendritic complexity compared with control, whereas overexpression of Rel49 did not induce noticeable change in dendrite morphology (Fig. 2, A and B; and Fig. S2 A). These data indicate that Rel68, but not Rel49 nor uncleaved Relish, has toxicity inducing changes in dendrite morphology in C4da neurons.

Next, we wondered whether polyQ-induced dendrite defects involve endoproteolytic cleavage of Relish to generate Rel68. To address this, we attempted to examine whether toxic MJD polyQ proteins change the quantity of these two protein fragments through Western blot analysis. However, neither fragment was detected by commercially available Rel49 antibody or our newly generated antibody for Rel68 (not depicted). We instead turned to genetic analysis to determine whether Dredd-mediated endoproteolytic cleavage of Relish to generate Rel68 contributes to polyQ-induced dendrite defects. To this end, we knocked down Dredd in C4da neurons expressing MJD-78Q and compared the consequent dendrite phenotypes to those caused by expression of MJD-78Q alone. Knockdown of Dredd significantly (P < 1.0 × 10⁻⁴) suppressed polyQ-induced dendrite defects (Fig. 2, C and D), while knockdown of Dredd only in C4da neurons did not induce noticeable changes in dendrite morphology and dendrite branch points (Fig. S2, D and E). Consistently, overexpression of Caspar, an upstream inhibitor of Dredd (Fig. S2 B), significantly restored polyQ-induced dendrite defects in C4da neurons (Fig. 2, C and D), while overexpression of Caspar only in C4da neurons did not induce noticeable changes in dendrite morphology and dendrite branch points (Fig. S2, D and E). In addition, we found that knockdown of Dredd or Caspar overexpression did not induce significant changes in the amount of MJD-78Q proteins (Fig. S3, A–F), suggesting that the restoration effect of Dredd RNAi or Caspar overexpression does not result from reduced amounts of toxic MJD polyQ proteins. These data suggest that Dredd-mediated endoproteolytic cleavage of Relish contributes to polyQ-induced dendrite defects.

Then, we asked how toxic MJD polyQ proteins affect Dredd-mediated endoproteolytic cleavage of Relish. We first checked whether Dredd mRNA levels were up-regulated by expression of MJD polyQ proteins and found that Dredd mRNA was not up-regulated in the existing RNA-sequencing database (Chung et al., 2017). Given that toxic polyQ proteins have a high propensity to sequester their target molecules, resulting in their loss of function (Chung et al., 2017; Kwon et al., 2018), the toxic polyQ proteins may up-regulate Dredd activity by aberrantly interacting with an upstream negative regulator of Dredd, such as Caspar. A search for known negative regulators of Dredd, other than Caspar, revealed none, so we hypothesized that toxic MJD polyQ proteins may aberrantly interact with Caspar in neurons. To explore this, we overexpressed Caspar in C4da neurons, with or without MJD-78Q expression, and compared localization patterns of Caspar. Overexpressed Caspar alone in C4da neurons showed dispersed localization throughout the cytoplasm, with relatively scarce localization in the nucleus (Figs. 2 E and S3 G). By contrast, overexpression of Caspar in C4da neurons expressing MJD-78Q showed aberrant localization in the nucleus, with puncta formation that colocalized with MJD-78Q (Figs. 2 E and S3 G), supporting our hypothesis of aberrant interaction between Caspar and MJD polyQ proteins. To confirm this aberrant interaction, we performed a coimmunoprecipitation experiment in HEK293T cells expressing Caspar with or without MJD-76Q and found that Caspar was coimmunoprecipitated with MJD-76Q (Fig. 2 F). Collectively, these data suggest that disinhibition of Relish toxicity required for polyQ-induced dendrite defects results from up-regulation of Dredd-mediated endoproteolytic cleavage of Relish to generate Rel68 owing to aberrant interactions between Caspar and toxic MJD polyQ proteins.

Then, to further support that N-terminal fragment of Relish contributes to polyQ-induced dendrite defects, we searched for characteristic molecular features of Rel68 that are distinct from full-length Relish and Rel49. According to previous studies, Rel68 translocates from the cytoplasm to the nucleus to act as a TF (Wiklund et al., 2009). Thus, we examined subcellular localization of Relish and its cleaved fragments in C4da neurons. We found that only Rel68, but not Relish nor Rel49, preferentially localized in the nucleus of C4da neurons (Fig. S4, A and B). In addition, we found that Rel68, but not Rel49, transcriptionally regulates the expression of downstream targets of Relish (CecA1, CecA2, CecB, AttB, and DptB) in fly brain by performing RT-PCR experiments (Fig. 3 A). Based on these findings showing that Rel68, but not Rel49, predominantly localizes to the nucleus to function as a TF in neurons, we hypothesized that nuclear localization of Rel68 is critical for MJD-78Q–induced dendrite defects. Next, we examined whether MJD-78Q expression increases the nuclear localization of Relish. Because of the lack of antibodies detecting N-terminal fragment of endogenous Relish, we co-overexpressed dual-tagged full-length Relish (Flag-Relish-V5) and MJD-78Q in C4da neurons and detected N-terminus of Relish using anti-Flag antibody. Of note, both cleaved and full-length forms of Relish are detected by anti-Flag immunostaining. Consequently, overexpression of MJD-78Q resulted in significantly increased nuclear proportions of Relish proteins detected by anti-Flag antibody (Flag-positive Relish) in C4da neurons; however, this increase in nuclear localization of Flag-positive Relish was markedly decreased by coexpression of Dredd RNAi or Caspar (Fig. 3, B and C). Together with our above results, that dendrite defects caused by toxic polyQ proteins were significantly restored by coexpression of Dredd RNAi or Caspar (Fig. 2, C and D) and that Caspar proteins aberrantly interact with toxic polyQ proteins (Fig. 2, E and F), these data suggest that nuclear localization of N-terminal fragment of Relish is increased by polyQ toxicity owing to aberrant up-regulation of Relish cleavage pathway, thereby resulting in dendrite defects.

Next, we further investigated whether inhibition of nuclear localization of Rel68 ameliorates polyQ-induced dendrite defects. To address this, we prevented nuclear localization of Rel68 in C4da neurons by knockdown of Charon mediating nuclear localization of Rel68 (Ji et al., 2016) and confirmed that Charon knockdown markedly reduced the nuclear localization of Rel68 (Fig. 3, D and E). Then, we examined whether the inhibition of nuclear localization of Rel68 by Charon knockdown suppresses Rel68-induced dendrite defects. To this end, we compared dendrite morphology of C4da neurons overexpressing both Charon RNAi and Rel68 with that of C4da neurons overexpressing Rel68 alone. Notably, the Rel68-induced decrease in number of dendrite branch points was significantly (P < 1.0 × 10⁻⁴) restored by Charon knockdown (Fig. 3, F and G), indicating that the nuclear localization of Rel68 is required for Rel68-induced neuronal toxicity causing dendrite defects. Next, we examined whether nuclear localization of Rel68 is required for polyQ-induced dendrite defects as well. PolyQ-induced dendrite defects were also significantly (P < 1.0 × 10⁻³) ameliorated by knockdown of Charon in C4da neurons (Fig. 3, H and I). In addition, we confirmed that inhibition of nuclear localization of Rel68 resulted in decreased mRNA expression of Relish target genes (Fig. S4 C) up-regulated by MJD-78Q (Fig. 1 E). Taken together, these data suggest that increased nucleus-localized N-terminal fragment of Relish contributes to dendrite defects induced by toxic polyQ proteins.

We then wondered what transcriptional target genes of Rel68 mediate Rel68-induced dendritic toxicity. Target genes of Relish or NF-κB identified in previous studies have broad molecular functions, but are primarily associated with immune response, such as cytokines and immunoreceptors (Cao et al., 2006; Hinz et al., 2001; Hiscott et al., 1993). To identify novel Rel68 target genes involved in Rel68-induced dendritic toxicity, we selected candidates from 171 Drosophila proteins whose functions are dendrite morphogenesis based on gene ontology biological process annotation, and further 51 TFs among them, since transcriptional regulation is the most strongly associated with dendrite morphogenesis (Fig. 1 B). To determine whether 41 of 51 TFs with mammalian homologues available are transcriptional target genes of Rel68, we used RT-PCR analysis to compare their mRNA levels in adult fly heads panneuronally overexpressing Rel68 with those in control heads (Fig. 4 A). Among the 41 TFs, we found 8 genes up-regulated (log₂ fold-changes ≥ 0.4, 1.3-fold) by Rel68 overexpression (Prospero [Pros], Cut [Ct], Abnormal chemosensory jump 6 [Acj6], Pox meso [Poxm], Enhancer of Polycomb [E[Pc]], Hamlet [Ham], Tailup [Tup], and Sequoia [Seq]), as well as three genes down-regulated (log₂ fold changes ≤ −0.4) by Rel68 overexpression (Hand, Absent, small, or homeotic discs 2 [Ash2], and Domino [Dom]). These data raise a possibility that these dendrite morphology–associated genes may be novel transcriptional target genes of Rel68 and mediate Rel68-induced neuronal toxicity causing dendrite defects.

Using available fly lines, we first confirmed the roles of transcriptionally up-regulated or down-regulated target genes of Rel68 in dendrite morphogenesis by examining dendrite phenotypes of C4da neurons with their increased or decreased gene expression, respectively. Among eight transcriptionally up-regulated genes, five (Pros, Ct, Poxm, Tup, and Seq) were available for overexpression transgenes, whose overexpression all markedly reduced the number of dendrite branch points in C4da neurons (Fig. 4 B). For three transcriptionally down-regulated genes, knockdown of each of Dom, Hand, and Ash2 in C4da neurons caused significant (P < 1.0 × 10⁻³) reduction in the number of dendrite branch points (Fig. 4 B). Then, we examined whether Rel68-induced dendrite phenotypes can be restored by decreasing or increasing expression levels of transcriptionally up-regulated or down-regulated target genes, respectively. Among seven transcriptionally up-regulated genes tested, only knockdown of Tup and Pros in C4da neurons showed significant restoration of Rel68-induced dendrite phenotypes (Fig. 4, C and D). On the other hand, overexpression of two transcriptionally down-regulated genes tested (Dom and Ash2) did not significantly restore Rel68-induced dendrite phenotypes (Fig. 4 C). These data suggest Tup and Pros as novel transcriptional target genes of Rel68 that mediate Rel68-induced dendritic toxicity. Then, we wondered whether Tup and Pros are key downstream effectors of Rel68 toxicity that contribute to polyQ-induced dendrite defects. To address this, we knocked down Tup or Pros in C4da neurons expressing MJD-78Q and examined dendrite phenotypes. Knockdown of Tup or Pros significantly (P < 1.0 × 10⁻²) restored dendrite defects caused by MJD-78Q expression in C4da neurons (Fig. 4, E and F). Collectively, these data demonstrate that Tup and Pros are novel downstream effectors of Rel68 that mediate Rel68 toxicity and contribute to toxic polyQ-induced dendrite defects.

To examine whether Rel68 directly regulates the expression of Tup and Pros, we performed chromatin immunoprecipitation (ChIP)-PCR analysis to assess binding enrichment of Rel68 to promoter regions of Tup and Pros in fly heads panneuronally overexpressing Rel68. ChIP-PCR analysis showed significant enrichment of Rel68 in promoter regions of Tup (Fig. S5 A) and Pros (Fig. S5 B). These data collectively suggest that Rel68 directly binds to the promoters of Tup and Pros to up-regulate their expression and induces dendrite defects.

Having characterized Relish-induced dendrite defects in the context of MJD polyQ disease and identified Tup and Pros as downstream mediators, we next wondered whether Rel68 induces additional neuronal toxicity beyond dendrite defects, and if so, whether Tup or Pros is associated with these toxicities as well. Of note, overexpression of a strong allele of Rel68 (Rel68^strong) in C4da neurons frequently resulted in complete loss of dendrites and axon tract, indicative of neurodegeneration (Fig. 5, A and B). This observation is consistent with a previous study showing degenerative eye phenotypes caused by Rel68 overexpression in Drosophila (Chinchore et al., 2012). The ratio of degenerating C4da neurons by Rel68^strong overexpression was significantly (P < 0.05) reduced by knockdown of Tup or Pros (Fig. 5 B).

We next examined whether overexpression of Rel68 in dendritic arborization (da) neuronal clusters can lead to organism-level changes in larval motility assessed by measuring crawling and head-turning activities (Riedl and Louis, 2012). We first tested the crawling activity determined by the time for the larvae to reach the dish edge, as previously described (Kwon et al., 2018), using the following four different lines of larvae: control larvae and larvae expressing Rel68, Tup, or Rel68 + Tup RNAi in da neuronal clusters using the 109(2)80-Gal4 driver. All control larvae reached the edge of the dish within 100 s, whereas only 26.6% of larvae overexpressing Rel68 reached the edge in this time (Fig. 5, C and D). The fraction of larvae expressing Rel68 reaching the edge of the dish increased significantly from 26.6% to 57.6% when Tup was knocked down (Fig. 5, C and D). Of note, only 16.4% of larvae overexpressing Tup reached the edge of the dish, comparable to the value of larvae overexpressing Rel68 (Fig. 5, C and D), supporting that Tup mediates Rel68-induced behavioral defects. To measure head-turning activity, we monitored the changes in angle of a larval head for 1 min, as previously described (Kwon et al., 2018). Control larvae barely turned their heads (Fig. 5, E and F), whereas larvae overexpressing Rel68 turned their heads frequently (Fig. 5, E and F). The frequent head turning of larvae expressing Rel68 was significantly reduced by knockdown of Tup (Fig. 5, E and F). Larvae overexpressing Tup showed frequent head turning, comparable to larvae overexpressing Rel68 (Fig. 5, E and F), which is consistent with observations on crawling activity described above. Collectively, these data demonstrate that Rel68-induced neuronal cell death and behavioral abnormalities are mediated by Tup.

We next wondered whether Rel68-mediated neuronal toxicity contributes to neuropathic features of NDs other than polyQ diseases. To address this, we used the C9orf72-associated ALS (C9ALS) fly model expressing an expanded GGGGCC 30-repeat (G4C2)₃₀ (Xu et al., 2013) to induce dendrite (Burguete et al., 2015) and behavioral (Goodman et al., 2019; Xu et al., 2013) defects and examined whether knockdown of Relish or Tup can suppress these defects. First, we checked whether overexpression of (G4C2)₃₀ up-regulates TF activity of Relish by measuring mRNA levels of transcriptional target genes of Relish (CecA1, CecA2, CecB, AttA, AttB, and DptB) in adult fly heads. Overexpression of (G4C2)₃₀ significantly increased mRNA levels of all tested target genes (Fig. 6 A). Furthermore, as shown in C4da neurons expressing MJD-78Q (Fig. 3, B and C), significantly increased nuclear localization of Flag-positive Relish was observed in C4da neurons expressing (G4C2)₃₀ (Fig. S6, A and B). Collectively, these data suggest the increased TF activity of Relish due to the increase in nuclear localization of N-terminal fragment of Relish in the C9ALS model.

We then examined whether knockdown of Relish or Tup can suppress dendrite defects induced by (G4C2)₃₀. Overexpression of (G4C2)₃₀ in C4da neurons significantly reduced the number of dendrite branch points (Fig. 6, B and C) as previously reported (Burguete et al., 2015). These dendrite defects were significantly suppressed by knockdown of Relish or Tup (Fig. 6, B and C). In addition, we further tested whether G4C2 repeat expansion–mediated dendrite defects can be synergistically aggravated by Rel68. The reduced number of dendrite branch points by Rel68 or (G4C2)₃₀ overexpression was significantly aggravated by co-overexpression of (G4C2)₃₀ and Rel68 (Fig. 6, D and E). These data demonstrate that the Rel68-Tup pathway mediates the dendrite defects of C9ALS model as well. Next, we performed a negative geotaxis assay to compare climbing activity among control adult flies and adult flies panneuronally expressing (G4C2)₃₀, (G4C2)₃₀ + Relish RNAi, or (G4C2)₃₀ + Tup RNAi. Flies overexpressing (G4C2)₃₀ showed significantly reduced climbing activity compared with control flies (Fig. 6 F), consistent with previous reports (Goodman et al., 2019). The reduced climbing activity was significantly (P < 1.0 × 10⁻²) restored by knockdown of Relish or Tup (Fig. 6 F). Furthermore, we also assessed the lifespan of control flies and flies panneuronally overexpressing (G4C2)₃₀ or (G4C2)₃₀ + Relish RNAi (Fig. 6 G). Flies overexpressing (G4C2)₃₀ showed reduced lifespan compared with control flies, consistent with previous studies (Mizielinska et al., 2014; Moens et al., 2019). Knockdown of Relish in flies overexpressing (G4C2)₃₀ suppressed the G4C2 repeat expansion–mediated reduction in lifespan, while knockdown of Relish in control flies showed reduced lifespan (Fig. 6 G; Discussion). Taken together, these data demonstrate that Rel68-mediated neuronal toxicity contributes to dendrite and behavioral defects in the C9ALS model.

In this study, we identified disinhibition of latent Relish toxicity as a novel pathogenic mechanism of action underlying dendrite and behavioral defects in animal models for two NDs, MJD polyQ disease and C9ALS. Our data showed that, as a mechanism mediating disinhibition of Relish toxicity in MJD polyQ disease condition, Dredd-mediated endoproteolytic cleavage of Relish to generate N-terminal fragment of Relish was up-regulated by aberrant interaction between Caspar and MJD toxic polyQ proteins (Fig. 2, E and F; and Fig. 3, B and C). This is further supported by our observation that overexpression of Relish alone in C4da neurons did not induce dendrite phenotypes (Fig. 2 A), indicating that increased abundance of Relish may be not a key determinant of neuronal toxicity associated with dendrite defects. As for C9ALS condition, based on the following previous studies, we speculate that up-regulation of Dredd-mediated endoproteolytic cleavage of Relish is similarly applicable as well. Although the cleavage-based regulation of NF-κB in mammals is different from that of Relish in Drosophila, Caspase-8 and Dredd share a conserved function of activating NF-κB and Relish, respectively. Previous studies showed that Caspase-8, a mammalian homologue of Dredd, is up-regulated in patient samples of ALS, potentially by loss of function of optineurin that binds to Caspase-8 (Nakazawa et al., 2016), and that genetic mutations in optineurin have been observed in patients with ALS (Maruyama et al., 2010). It will be valuable to see whether documented changes in subcellular localization, activity, and intracellular expression level of NF-κB seen in ND neurons and/or glial cells (Table S1) are, in fact, associated with our proposed pathogenic model involving disinhibition of latent NF-κB toxicity.

NF-κB plays various well-characterized essential physiological roles, including regulation of immune response (Cao et al., 2006; Hinz et al., 2001; Hiscott et al., 1993) and cell death (Chinchore et al., 2012). In the nervous system, the most recognized role involves regulation of neuroinflammation in glial cells, such as microglia and astrocytes (Shih et al., 2015). Aberrant activation of neuroinflammation is often observed in the late stages of various NDs, which leads to nonselective degeneration of neurons throughout the entire brain. The dysregulation of NF-κB, therefore, has garnered great attention since it was reported that NF-κB contributes to non–cell-autonomous neurodegeneration in late stages of NDs (Heneka et al., 2015; Hirsch and Hunot, 2009; Philips and Robberecht, 2011). In contrast, in this study, we characterized, for the first time, cell-autonomous effects of NF-κB on an early pathogenic feature, dendrite defects, in animal models for NDs.

The up-regulation of quantity and/or activity of NF-κB has been consistently seen in neurons of both ND patients and animal models for ND (Table S1), supporting our finding of cell-autonomous contribution of NF-κB in neurons to the disease pathogenesis. Consistently, several previous studies showed that NF-κB induces neuronal toxicity in a cell-autonomous manner. Among five mammalian homologues of Relish, RelA/p65 is known to confer neuronal toxicity, inducing ectopic cell death by forming a heterodimer with p50 in ischemic conditions (Sarnico et al., 2009). In addition, knockdown of RelA reduced glutamate-mediated cell death of cerebellar granule neurons, which is indicative of a neuronal proapoptotic function of RelA (Pizzi et al., 2002). Likewise, overexpression of Rel68 in Drosophila eyes induced degenerative phenotypes (Chinchore et al., 2012). Interestingly, however, a neuroprotective role of NF-κB in a cell-autonomous manner has been also proposed. For example, Bhakar et al. (2002) showed that overexpression of RelA prevented cell death of primary cortical neurons induced by neurotoxins such as camptothecin and etoposide. In addition, overexpression of RelA can promote cellular survival of cultured nodose neurons (Gutierrez et al., 2005). Consistently, we observed that panneuronal expression of Relish RNAi alone reduced lifespan compared with the control (Fig. 6 G). Therefore, based on these findings, functional roles of NF-κB in neurons could be shifted between protective and toxic depending on the cellular and environmental context, and amounts and functions of NF-κB in neurons should be tightly regulated to sustain neuronal homeostasis and health. Future studies elucidating specific functions and modes of regulation of NF-κB in various cellular environmental contexts are expected to broaden our understanding of the basic principles of neuronal maintenance and pathophysiology of neurological disorders associated with NF-κB.

The main mechanism proposed in this study is the aberrant increase in the nucleus-localized N-terminal fragment of Relish by changes in cleavage machinery due to the toxic disease proteins. Even though this mechanism is supported by our genetic and immunohistochemical results (Fig. 2, C–F; and Fig. 3, B–I), Western blot analysis directly detecting N-terminal fragment of Relish is thought to be the most direct evidence related to this. However, when we performed Western blot analysis to detect N-terminal fragment of Relish using anti-Flag antibody, which detects N-terminus of dual-tagged full-length Relish (Flag-Relish-V5), for unknown reasons, it failed to detect any signals of cleaved N-terminal fragment of Relish. Consistent with our results, a previous study in Drosophila cell line reported that Flag-tagged N-terminal fragment of Relish is barely detected through Western blot analysis (Tanji et al., 2010). Alternatively, we showed the involvement of cleavage machinery of Relish on the nuclear localization of Flag-positive Relish in C4da neurons using immunohistochemistry accompanied by genetic modulation of cleavage machinery (Fig. 3, B and C). We expect that further studies surely can overcome this technical limitation and provide solid evidence supporting the conclusion of this study.

In this study, we present clear evidence for NF-κB–induced neuronal toxicity in the context of NDs. We also identified Tup and Pros as mediators of NF-κB–induced neuronal toxicity. Given that our findings propose therapeutic measures to intervene in polyQ disease and C9ALS before significant dendrite defects occur, we speculate that inhibitor-based therapeutics targeting Relish/NF-κB and/or downstream mediators (Tup and Pros) may be of particular use in modifying NDs, especially during early stages of disease.

Following fly stocks were obtained from Bloomington Drosophila Stock Center: UAS-Pros.L (BL32244), UAS-Tup (BL39661), UAS-Rel RNAi (BL33661), UAS-Dredd RNAi (BL34070), UAS-Pros RNAi (BL42538), UAS-Tup RNAi (BL51763), UAS-Shn RNAi (BL34689), UAS-Ftz-f1 RNAi (BL33625), UAS-Ci RNAi (BL64928), UAS-Smox RNAi (BL41670), UAS-Rfx RNAi (BL61847), UAS-RunxA RNAi (BL33353), UAS-D RNAi (BL34672), UAS-Foxo RNAi (BL32427), UAS-Cnc RNAi (BL32863), UAS-Usp RNAi (BL56729), UAS-Seq.B (BL9244), UAS-Dom RNAi (BL38385), UAS-Seq RNAi (BL51923), UAS-Ct RNAi (BL29625), UAS-E(Pc) RNAi (BL67921), UAS-Dom.A (BL64262), UAS-MJD-78Q (BL8150), UAS-MJD-78Q (BL8141), elav-gal4 (BL8760), UAS-Rel-V5.49 (BL55779), UAS-Flag-Rel.68 (BL55777), and UAS-Flag-Rel.68 (BL55778). UAS-Flag-Rel.68 (BL55777) has higher expression levels of Rel68 and therefore is denoted as Rel68^strong in this study. UAS-Flag-Rel68 (BL55778) has milder expression levels of Rel68 and is denoted as Rel68 in this study. UAS-Rel RNAi (KK108469), UAS-Rel RNAi (GD49413), UAS-Charon RNAi (KK106292), UAS-Hand RNAi (GD23306), UAS-Ash2 RNAi (GD7141), UAS-Ham RNAi (GD40763), and UAS-Acj6 RNAi (KK105292) were obtained from Vienna Drosophila Resource Center. The following fly lines were obtained from Fly ORF: UAS-Rel-3xHA (F001780), UAS-Poxm (F003333), and UAS-Ash2-3xHA (F003904). UAS-(GGGGCC)₃₀ was kindly provided by Peng Jin (Emory University School of Medicine, Atlanta, GA). Hsp70-Gal4 was kindly provided by Y.N. Jan (University of California San Francisco, San Francisco, CA). The following lines were used as previously described (Kwon et al., 2018): ppk-gal4 (III), ppk^1a-gal4 (III), 109(2)80-gal4 (II), UAS-CD4tdGFP (II), and UAS-CD4tdGFP (III). All fly stocks were raised at room temperature (25°C) with 60% humidity except for negative geotaxis assay and lifespan assay (temperature, 27°C; humidity, 60%).

Following transgenic fly lines were generated in this study: UAS-2xFlag-Relish-2xV5, UAS-2xHA-Rel68, and UAS-2xV5-Caspar. The transgene of UAS-2xFlag-Relish-2xV5 was synthesized by Biomatik. UAS-2xHA-Rel68 transgene was generated by engineering the UAS-2xFlag-Relish-2xV5 transgene. The UAS-2xV5-Caspar transgene was generated using LD03368 clone (Berkeley Drosophila Genome Project). All transgenes were incorporated into pACU2 vector, and each transgenic fly line was generated by BestGene.

Third instar larval fillet was fixed with 3.7% formaldehyde for 20 min at room temperature. After brief washing with 0.3% PBT (PBS containing 0.3% Triton X-100), dissected tissues were incubated in blocking buffer (5% normal donkey serum in 0.3% PBT) for 50 min at room temperature. The tissues were incubated at 4°C with primary antibody. The following primary antibodies were used in this study: rat anti-HA (3F10, Roche; 1:200 dilution) for detecting the HA epitope; mouse anti-V5 (R960-25; Thermo Fisher Scientific; 1:200) for detecting the V5 epitope; and mouse anti-Flag (1E6, Wako; 1:200) for detecting the Flag epitope. To detect primary antibodies, the following secondary antibodies were used: donkey anti-mouse cy3 (Jackson Immunoresearch Laboratories; 1:300), goat anti-mouse Alexa Fluor 647 (Invitrogen; 1:1,000), goat anti-HRP Alexa Fluor 488 (Jackson Immunoresearch Laboratories; 1:1,000), and goat anti-rat Alexa Fluor 647 (Jackson Immunoresearch Laboratories; 1:1,000).

All confocal images of C4da neurons were obtained from abdominal segments A4 and A5 using LSM700, LSM780, or LSM800 microscopes (Zeiss) at 20°C. For live imaging of C4da neurons genetically labeled with mCD4tdGFP, an individual third-instar larva was mounted in 1× PBS, and C4da neurons were captured using a 20×/0.5-NA objective. For imaging of immunostained samples, a larval fillet was mounted in PBG (70% glycerol in 1× PBS), and C4da neurons were captured using a 40× water/2.0-NA objective. The following fluorochromes were used in immunostaining experiments: Alexa Fluor 488, Alexa Fluor 647, cy3, and EGFP. After image acquisition, maximum-intensity projections of each image were obtained using Zeiss Zen software, and additional image processing (pseudocolor filters or deconvolution) was applied in Adobe Photoshop CC.

Western blot analysis was performed as previously described (Kwon et al., 2018). Protein lysates were extracted from adult fly heads using lysis buffer (150 mM NaCl, 1% Triton X-100, and 50 mM Tris-HCl, pH 7.5). Next, 4× Laemmli sample buffer (161-0747; Bio-Rad) and β-mercaptoethanol were mixed with 30 µg of total proteins of each extracted sample to be 1× and 2.5%, respectively. The mixture was incubated for 5 min at 95°C and loaded into Mini-PROTEAN TGX Stain-Free, 4–15% gel (BR456-8083; Bio-Rad) for electrophoresis. The separated proteins were transferred to the nitrocellulose membrane (BR170-4270; Bio-Rad), and the nitrocellulose membrane was incubated in blocking buffer (TBST [50 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 0.1% Tween-20] containing 5% skim milk) for 1 h at room temperature. To detect endogenous Relish and β-tubulin, mouse anti-Relish-C (21F3; Developmental Studies Hybridoma Bank; 1:1,000) and mouse anti-β-tubulin (E7; Developmental Studies Hybridoma Bank; 1:1,000) were used, respectively. For detecting HA-MJD-78Q protein, rat anti-HA (3F10; Roche; 1:1,000) antibody was used. To detect primary antibodies, the following secondary antibodies were used: anti-mouse IgGκ BP HRP conjugated (sc-516102; Santa Cruz; 1:3,000) and goat anti-rat IgG-HRP (sc-2006; Santa Cruz; 1:3,000).

To identify protein interactions between MJD-76Q and Caspar through coimmunoprecipitation, HEK293T cells were transfected with 2xV5-Caspar and 2xHA-MJD-76Q subcloned into EGFP-N1 vector and cultured for 24 h before sample preparation. Harvested HEK293T cells were homogenized in lysis buffer (150 mM NaCl, 1% Triton X-100, and 50 mM Tris-HCl, pH 7.5), and whole-cell lysates containing equivalent protein amounts were incubated with 1 µg of rat anti-HA (3F10; Roche) at 4°C overnight. After the incubation, the precleaned protein A/G plus agarose beads (sc-2003; Santa Cruz) with lysis buffer were coincubated with the whole-cell lysates containing HA antibody for 3 h at 4°C in a shaking incubator. After the incubation, the beads were washed with lysis buffer three times, and immunoprecipitated proteins were eluted in 1× Laemmli sample buffer containing 2.5% β-mercaptoethanol. These immunoprecipitated proteins were separated through electrophoresis and transferred to nitrocellulose membrane as described above. To detect Caspar and MJD-76Q through Western blot analysis, mouse anti-V5 (R960-25; Thermo Fisher Scientific; 1:1,000) and rat anti-HA (3F10; Roche; 1:1,000) antibodies were used, respectively. To detect primary antibodies, the following secondary antibodies were used: anti-mouse IgGκ BP HRP conjugated (sc-516102; Santa Cruz; 1:3,000) and goat anti-rat IgG-HRP (sc-2006; Santa Cruz; 1:3,000).

RT-PCR analysis was performed as previously described (Chung et al., 2017). Total RNAs were extracted from adult fly heads using Easy-Blue system (iNtRON Biotechnology). cDNAs were synthesized from 3 µg of total RNAs using GoScript Reverse Transcription (A2791; Promega) according to the manufacturer’s standard protocol. For RT-PCR analysis, each target gene was amplified with the corresponding primer set (Table S2 and Table S3) using GoTaq G2 Master Mixes (M7823; Promega) in a C1000 Thermal Cycler, C1000 Touch Thermal Cycler, or T100 Thermal Cycler system (Bio-Rad).

Because of the lack of available ChIP-grade antibody that specifically precipitates endogenous Rel68, we panneuronally overexpressed HA-Rel68 and examined whether overexpressed HA-Rel68 is enriched in the promoter regions of Tup or Pros. ChIP-PCR experiments were performed as previously described (Chung et al., 2017). First, fly heads were collected in ice-cold 1× PBS. After elimination of the PBS, the collected fly heads were homogenized in fixation buffer (1.5% formaldehyde, 0.143 M NaCl, 1.43 mM EDTA, 71.45 mM Hepes-KOH, pH 7.5, and 1× protease inhibitor [87786; Thermo Fisher Scientific]) and incubated for 20 min at room temperature to cross-link target TFs with chromatin. Then, glycine and Tris were added to the fixation buffer to be 370 mM and 2.5 mM, respectively, and incubated for 5 min at room temperature to terminate the cross-linking reaction. After 5-min centrifugation at 4,000 g, the pellet was washed with Tris-buffered saline (20 mM Tris-HCL, pH 7.5, and 150 mM NaCl), FA lysis buffer (50 mM Hepes-KOH, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% SDS, and 0.1% Na-deoxycholate), FA lysis buffer/0.5% SDS (50 mM Hepes-KOH, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.5% SDS, and 0.1% Na-deoxycholate), and FA lysis buffer sequentially. The pellet was then resuspended in FA lysis buffer.

For the chromatin shearing, the resuspended sample was sonicated for 20 min using Covaris M220 with following conditions: peak power, 70.0; duty factor, 25.0; cycles/burst, 200; average power, 17.5. The sheared nucleosomes were acquired after centrifugation for 10 min at 17,000 g. After Bradford assay, input control was prepared from the lysates, and remaining lysates were incubated with 3 µg of anti-HA antibody (C29F4; Cell Signaling Technology) at 4°C overnight. Next, precleaned protein A/G plus agarose beads (sc-2003; Santa Cruz) with FA lysis buffer were added into the lysates containing HA antibody and incubated for 4 h at 4°C. The beads were precipitated and washed with FA lysis buffer and washing buffer (10 mM Tris-HCl, pH 8.0, 0.25 M LiCl, 1 mM EDTA, 0.5% NP-40, and 0.5% Na-deoxycholate) sequentially. Sheared chromatin precipitated with the beads were eluted using elution buffer (50 mM Tris-HCl, pH 7.5, 10 mM EDTA, and 1% SDS) at 65°C for 20 min. The eluted samples and prepared input control were incubated with RNase (1 mg/ml) at 37°C for 1 h. After 42°C incubation with proteinase K (20 mg/ml) for 1 h, the samples were incubated at 65°C for 5 h. The DNA was then extracted using phenol-chloroform-isoamyl alcohol.

Next, quantitative PCR (qPCR) experiments were performed using the extracted DNA with QuantiSpeed SYBR Green kit (QS105-10; PhilKorea) and CFX96 TouchTM Real-Time PCR Detection System (Bio-Rad). For PCR amplification of samples, the following primers were used: Tup promoter, forward 5′-AGT​TCG​ACA​AGC​GCA​TTA​CAA​CAG-3′ and reverse 5′-TGC​CAA​TTT​CGC​TTT​TGG​TTT​TCG-3′; Tup ORF, forward 5′-GGC​ATG​ACT​CCC​CAC​TGA​ATC​T-3′ and reverse 5′-GTG​TGG​GTA​TTG​ATC​GCT​CCA​TT-3′; Pros promoter, forward 5′-CCT​TTT​TGG​ACC​GCG​CTT​CAT​T-3′ and reverse 5′-CCC​CCT​CCG​TAT​GTA​AAA​GCA​TCT-3′; Pros ORF, forward 5′-CCC​CAG​AAT​TTC​CGC​TTC​GT-3′ and reverse 5′-CGA​CTT​GAA​GTA​CTC​GGG​CAC-3′.

The larval motility assay was performed as previously described (Kwon et al., 2018). 109(2)80-gal4 driver was used to express each transgene in da neuronal clusters. Each larva was washed in 1× PBS and placed in the center of a 90-mm Petri dish. The motility assay was performed inside a dark box with indirect lighting. After 1-min acclimation, we recorded total elapsed time for larvae to reach the edge of a 90-mm Petri dish up to 100 s. Then, we estimated a cumulative density function by using the Gaussian kernel density estimation method (Bowman and Azzalini, 1997) and evaluated statistical significance of difference between the cumulative density function of each genotype using Matlab.

To analyze larval motility from acquired video files, we used EthoVision XT (v11.5; Noldus Information Technology) video tracking system. In Fig. 5 C, the entire trace of larva was converted into a heatmap in which color represents elapsed time that larva spent in the specific location. For all heatmaps, the amount of time was normalized to those during 100 s. For head-turning behavior of larva, we used the Heading tool in EthoVision XT and calculated changes in head angle of each larva every second for 1 min. We evaluated the statistical significance of the difference in head-turning angles of each genotype using Matlab.

Right after eclosion from pupa, 15 flies of each genotype were collected and reared for 3 d at 27°C with 60% humidity. After 3 d, these flies were transferred to an acrylic cylinder closed at one end (3-cm diameter, 18-cm height) without CO₂ anesthesia, and the top of the cylinder was sealed with sponges to block escape. After 30-min acclimation at room temperature, the flies were placed on the bottom of the cylinder by tapping the cylinder five times against a table, and climbing was recorded for 1 min. Climbing ability was measured with a climbing index (proportion of 15 flies climbing >10 cm from the bottom of cylinder within 10 s) in each experimental group of flies.

Statistical analysis was performed using Prism 7 (GraphPad Software) with Student’s unpaired t test, one-way ANOVA (with Tukey post hoc test), or two-way ANOVA (with Tukey post hoc test). Data distribution was assumed to be normal, but this was not formally tested. Values are presented as mean ± SEM. P < 0.05 was considered statistically significant (*, P < 0.05; **, P < 1.0 × 10⁻²; ***, P < 1.0 × 10⁻³; and ****, P < 1.0 × 10⁻⁴).

Fig. S1 shows that the restoration effects of Relish RNAi on polyQ-induced dendrite defects are not attributed to the changes in the amount of MJD-78Q proteins. Fig. S2 shows that knockdown and overexpression of the regulators of endoproteolytic cleavage of Relish (Dredd and Caspar, respectively) have no effect on dendrite morphology. Fig. S3 shows that the restoration effects of Dredd knockdown or Caspar overexpression on polyQ-induced dendrite defects are not attributed to the changes in the amount of MJD-78Q proteins. Fig. S4 supports Fig. 3 by showing that the increased transcriptional activity of Relish by toxic polyQ proteins is significantly reduced by the inhibition of its nuclear localization by Charon RNAi. Fig. S5 shows enrichment of Rel68 in the promoters of Tup and Pros. Fig. S6 shows increased nuclear proportion of Flag-positive Relish in C4da neurons co-overexpressing (G4C2)₃₀ and dual-tagged full-length Relish (Flag-Relish-V5). Table S1 lists studies reporting association of NF-κB with NDs and aging. Table S2 and Table S3 include detailed information about primers used for RT-PCR.

This work was supported by Basic Science Research Program through the National Research Foundation of Korea, funded by the Ministry of Science and ICT, South Korea (2018R1A2B6001607 and 2019R1A4A1024278 to S.B. Lee); the Development of Platform Technology for Innovative Medical Measurements Program from the Korea Research Institute of Standards and Science (KRISS-2019-GP2019-0018 to S.B. Lee); and Korea Brain Research Institute basic research program, funded by Ministry of Science and ICT, South Korea (20-BR-04-02 to S.B. Lee).

The authors declare no competing financial interests.

Author contributions: M.H. Han, M.J. Kwon, B.S. Ko, and S.B. Lee conceived the project. M.H. Han, M.J. Kwon, B.S. Ko, D.Y. Hyeon, and D. Hwang performed experiments and analyzed the data. M.H. Han, M.J. Kwon, B.S. Ko, D. Lee, H.-J. Kim, and S.B. Lee interpreted the data and wrote the manuscript.