N-glycanase NGLY1 regulates mitochondrial homeostasis and inflammation through NRF1

The N-glycanase 1 (NGLY1), also known as peptide:N-glycanase (PNGase), is an evolutionarily conserved enzyme that removes N-glycans from glycosylated proteins in the cytosol (Suzuki et al., 2016). NGLY1 plays a crucial role in quality control for newly synthesized N-glycoproteins. For example, during ER-associated degradation, misfolded N-glycoproteins are retrotranslocated from the ER lumen to the cytosol (Hirsch et al., 2003). NGLY1 removes N-glycans from these misfolded proteins before they can be degraded by the proteasome. Mutations in the NGLY1 gene are associated with a rare congenital disorder affecting mostly young children (IMIM 615273 and 610661; Enns et al., 2014; Lam et al., 2017). Almost all NGLY1 missense and nonsense mutations reduce the NGLY1 protein level and enzymatic activity (He et al., 2015). Clinical features of NGLY1-deficient patients include global developmental delay, neurological disorders, and liver disease (Enns et al., 2014), clearly indicating the physiological importance of the enzyme.

One important substrate of NGLY1 is the ER-associated glycosylated transcription factor nuclear factor, erythroid 2 like 1 (Nfe2l1, also known as Nrf1). Nrf1 requires deglycosylation by NGLY1 and proteolysis by DNA Damage Inducible 1 Homologue 2 (DDI2) before nuclear translocation, where the nuclear form of Nrf1 drives expression of proteasome subunit genes (Tomlin et al., 2017; Owings et al., 2018). Nrf1 plays an important role in the proteasome “bounce-back” mechanism where it drives expression of proteasome subunit genes in response to proteotoxic stress (Radhakrishnan et al., 2010). Interestingly, Nrf2, a close homologue of Nrf1, is not glycosylated, responds to oxidative stress, and regulates similar proteasome subunit genes, as does Nrf1 (Kwak et al., 2003). Nrf2 also regulates expression of autophagy and mitophagy genes (East et al., 2014; Pajares et al., 2016).

Despite severe developmental and neurological disabilities in NGLY1 children, interestingly, parents often report that these affected NGLY1 children “oddly” appear to be resistant to common childhood viral infections (Mnookin, 2014). Similar viral resistance phenotypes were also reported in patients with chronically activated type I IFN response that are caused by mutations in nucleic acid metabolizing enzymes such as TREX1 and RNaseH2 (Hasan et al., 2013; Pokatayev et al., 2016). TREX1 and RNaseH2 mutations were associated with a neuro-inflammatory disease called Aicardi-Goutières syndrome (AGS). We and others have shown that TREX1- and RNaseH2-associated AGS is caused by accumulation of self-DNA in the cytosol and activation of the innate immune DNA-sensing pathway (Crow and Manel, 2015). Innate immune detection of microbial or self-DNA by the DNA sensor cGAS activates downstream STING–TBK1–IRF3 signaling cascade, leading to type I IFN and inflammatory responses and induction of IFN-stimulated genes (ISGs) that resembles an antiviral state (Wu and Chen, 2014). A parallel cytosolic RNA-sensing pathway begins with RIG-I or MDA5 recognizing the RNA ligand, then signals through MAVS–TBK1–IRF3 to activate IFN response. Gain-of-function mutations in MDA5 are also associated with AGS (Crow and Manel, 2015). Mitochondrial and nuclear genomic DNA are major sources of self-DNA, which can accumulate in the cytosol during mitochondrial stress (West et al., 2015), DNA damage (Härtlova et al., 2015), or deficiency in cytosolic nucleic acid metabolizing enzymes, such as TREX1 (Stetson et al., 2008; Wolf et al., 2016), and cause a broad range of autoimmune and inflammatory diseases, including AGS and systemic lupus erythematosus (SLE; Yan, 2017). In this study, we report novel innate immune and mitochondrial defects associated with NGLY1 deficiency. We also present the underlying molecular mechanism of these defects and an unexpected therapeutic strategy for treating NGLY1 patients.

We first investigated whether Ngly1 deficiency is associated with increased expression of ISGs, also known as the IFN gene signature, that is commonly observed in autoimmune diseases such as type I interferonopathy and cells at an antiviral state (Hasan et al., 2013; Pokatayev et al., 2016). We found that several ISGs were up-regulated in Ngly1^−/− murine embryonic fibroblasts (MEFs), compared with littermate WT cells (Fig. 1, A and B). We also knocked down Ngly1 expression using shRNA and observed increased ISG expression in two independent lines of Ngly1 shRNA-expressing cells (Fig. 1, C and D). To further determine whether the increased ISG expression is caused by the lack of Ngly1 enzymatic activity, we stably expressed WT or an enzymatically inactive Ngly1 mutant C306A (Hirsch et al., 2003) in Ngly1^−/− MEFs using a retroviral vector (Fig. 1, E and F). WT Ngly1 expression in Ngly1^−/− MEFs significantly reduced ISG expression down to close to normal levels, but Ngly1 mutant C306A did not (Fig. 1, E and F), demonstrating that the N-glycanase activity is critical for inhibiting ISG expression. We previously showed that cell-intrinsic activation of ISGs renders cells to become resistant to viral infections (Hasan et al., 2013; Pokatayev et al., 2016). We challenged WT and Ngly1^−/− MEFs with vesicular stomatitis virus (VSV), and we found that VSV replication was significantly inhibited in Ngly1^−/− MEFs, consistent with an overall antiviral state (Fig. 1 G).

To determine whether NGLY1 deficiency also cause innate immune activation in human cells, we generated NGLY1 knockout THP-1 cells (a human monocytic cell line) using CRISPR/Cas9 (Fig. 1 H). Three independent clones of NGLY1^KO THP-1 cells all showed robustly increased ISG expression compared with WT and two nontargeting control cell lines (Fig. 1 H). We next examined lymphoblastoid cell lines from NGLY1 patients and four healthy controls. NGLY1 patient cells show reduced or undetectable NGLY1 protein and elevated expression of several ISG transcripts compared with healthy controls (Fig. 1 I). Collectively, our data showed that NGLY1 deficiency in both human and mouse cells activate innate immune signaling leading to increased expression of ISGs that resemble an antiviral state.

We next sought to determine which innate immune signaling pathway is involved in inducing ISG expression in NGLY1-deficient cells. We first used inhibitors that block key kinases required for innate immune signaling. TBK1 inhibitors (Compound II [Hasan et al., 2015] and BX-795) and IKK-ε inhibitor (CAY10576) completely suppressed ISG expression in Ngly1^−/− MEFs, while IKK-α/β inhibitor (BAY11-7082) partially decreased ISGs expression (Fig. 2 A). We next used siRNAs to knockdown key components of the cytosolic DNA- (via cGAS–STING–TBK1–IRF3) and RNA- (via RIG–I/MDA5–MAVS–TBK1–IRF3) sensing pathways (Fig. 2 B). We found that knockdown of cGas, Sting, or Irf3 completely abolished the elevated ISG expression in Ngly1^−/− MEFs (Fig. 2 C). Knockdown of MDA5 or Mavs, which are required for sensing cytosolic RNA, also reduced overall ISG expression, although residual ISG expression in Ngly1^−/− MEFs remained higher than WT controls, suggesting that RNA-sensing pathways are partially required for innate immune activation in Ngly1^−/− MEFs (Fig. 2 D). We next generated shRNA-mediated stable knockdown of Sting or cGas, and both completely suppressed ISG expression in Ngly1^−/− cells (Fig. 2 E). We further measured phosphorylation of STING (Ser365) and found increased phospho-STING in Ngly1^−/− cells compared with littermate Ngly1^+/+ cells (in which low baseline phosphorylation was detected), providing direct evidence for the activation of the cGAS–STING pathway (Fig. S1).

Ngly1^−/− mice are reported to be embryonic lethal (on the C57BL/6 background; Fujihira et al., 2017). We next bred Ngly1^+/− mice to Sting^−/−, Mavs^−/− or Ifnar^−/− mice to test whether ablation of either cytosolic DNA or RNA sensing or type I IFN signaling pathway could eliminate innate immune activation and rescue embryonic lethality of Ngly1^−/− mice. We found that all double knockout neonates were still-born or died soon after birth, similar to Ngly1^−/− alone (Fig. 2 F; Table 1). We then isolated primary MEFs from each double knockout embryos at day E13.5 and found that Ngly1^−/−Sting^−/− completely suppressed ISGs expression to the low levels of Ngly1^+/+Sting^−/−, whereas Ngly1^−/−Mavs^−/− MEFs showed substantial but incomplete suppression of ISGs (Fig. 2, G and H), consistent with the RNAi knockdown data. Together, these data suggest that NGLY1 deficiency leads to aberrant activation of cytosolic nucleic acid–sensing pathways, especially the cGAS–STING pathway. The genetic evidence also suggests that innate immune activation is unlikely to be the main cause of embryonic lethality of Ngly1^−/−, thus the process upstream of immune activation needs to be explored.

We next explored the cause of immune activation and potential molecular pathogenesis of Ngly1 deficiency. Nuclear genomic and mitochondrial DNA (mtDNA) are major sources of endogenous ligand for the cGAS–STING DNA-sensing pathway (Rongvaux et al., 2014; White et al., 2014; Härtlova et al., 2015). Since cGAS is the main cytosolic DNA sensor and is required for ISG induction in Ngly1^−/− MEFs, we stably expressed FLAG-tagged cGAS in WT or Ngly1^−/− MEFs using a retroviral vector, immunoprecipitated (IP) FLAG-cGAS, and isolated bound DNA (Fig. 3 A). We found that mtDNA were enriched in FLAG-cGAS pulldown in Ngly1^−/− MEFs compared with that in the WT cells (Fig. 3 B). FLAG-IP efficiency was similar in both cells (Fig. 3 A). We further depleted mtDNA using ethidium bromide in both WT and Ngly1^−/− cells and found that mtDNA depletion substantially decreased ISGs expression in Ngly1^−/− cells (Fig. S2, A and B). Ethidium bromide treatment alone slightly increased baseline ISG levels in WT cells. These data suggest that increased cytosolic mtDNA is responsible for cGAS–STING activation in Ngly1^−/− cells. In comparison, genomic DNA was not detectable in cGAS-IP complex, likely due to the mild detergent we used to isolate the cytosolic content for the IP (data not shown). We also did not detect any evidence of DNA damage in Ngly1^−/− MEFs, and both WT and Ngly1^−/− MEFs responded similarly to exogenous DNA damage caused by etoposide (Fig. S2 C). These data suggest that mitochondrial DNA is detected by cGAS in Ngly1^−/− MEFs.

The leakage of mtDNA into the cytosol prompted us to examine the mitochondrion in more detail. Using immunofluorescence microscopy, we found that mitochondria in Ngly1^−/− MEFs are severely fragmented compared with the tubular structure observed in WT cells (Fig. 3, C and D). Stable expression of WT Ngly1 in Ngly1^−/− MEFs corrected fragmented mitochondria phenotype, suggesting that the defect was indeed caused by Ngly1 deficiency (Fig. 3, E and F). Ngly1^−/−Sting^−/− and Ngly1^−/−Mavs^−/− primary MEFs, despite having reduced ISGs, exhibited similarly fragmented mitochondria compared with Ngly1^−/− MEFs, suggesting that mitochondrial damage occurs upstream of immune activation in Ngly1^−/− cells (Fig. S3, A and B). NGLY1-deficient patient fibroblasts also showed more fragmented mitochondria compared with the health control (Fig. 3, G and H). We next examined mitochondrial respiratory functions. The oxygen consumption rate (OCR), which is an indicator of mitochondrial respiration, is decreased in both basal mitochondrial respiration and maximal respiratory capacity in Ngly1^−/− compared with WT MEFs (Fig. 3 I). Together, our data suggest that NGLY1 deficiency causes mitochondrial fragmentation, leading to impaired mitochondrial function and leakage of mtDNA and possibly RNA into the cytosol, which activate innate immune signaling.

The mitochondrion is a dynamic organelle, and fragmented mitochondria can be caused by defects in either mitochondrial fusion or mitophagy, the latter of which is required for clearance of damaged mitochondria (Fig. 4 A; Seo et al., 2010). We first examined the expression of mitochondrial fusion proteins MFN1, MFN2, and OPR1, and found no difference between WT and Ngly1^−/− cells (Fig. S4 A). We also stressed Ngly1^−/− cells with cycloheximide or actinomycin D, both of which induce hyperfusion of mitochondria (Tondera et al., 2009). Both treatments induced elongated tubular mitochondria in Ngly1^−/− cells, suggesting that mitochondrial fusion machinery is intact in Ngly1^−/− cells (Fig. S4 B).

We next examined the effect of Ngly1^−/− deficiency on mitophagy. We used HeLa cells stably expressing E3 ligase Parkin (HeLa/Parkin cells) to analyze mitophagy-mediated clearance of damaged mitochondria (Wei et al., 2017). We knocked down NGLY1 expression with three different siRNAs and treated control and NGLY1-knockdown HeLa/Parkin cells with a combination of oligomycin and antimycin (OA) that activates mitophagy by inducing mitochondria damage (Fig. 4 B). While control siRNA-treated cells cleared damaged mitochondria effectively (note the clearance of cytoplasmic mitochondrial signal), NGLY1 knockdown led to significantly impaired mitophagy as measured by both mtDNA (Fig. 4, C and D) and HSP60 (Fig. 4, E and F) staining. Similarly, we observed impaired mitophagy in NGLY1^KO THP-1 cells (Fig. 4, G and H). We further analyzed expression of mitophagy-related genes, and found a global decrease in expression in Ngly1^−/− MEFs compared with WT cells regardless of mitophagy inducers treatment (Fig. 4 I). Collectively, these results demonstrate that NGLY1 deficiency impairs mitophagy leading to accumulation of damaged mitochondria. The change in mitophagy-related gene expression profile in Ngly1^−/− cells raises the possibility that NGLY1 regulates mitochondrial homeostasis through a transcription factor.

Recent studies in both mammalian and Drosophila melanogaster cells revealed that the ER-anchored transcription factor Nrf1 requires deglycosylation by NGLY1 in the cytosol before translocation into the nucleus and activation of proteasome genes in response to proteotoxic stress (Tomlin et al., 2017; Owings et al., 2018). Nrf1 has not been implicated in mitochondrial homeostasis, but its homologue Nrf2, which binds to similar proteasome gene targets, does play a role in autophagy and mitophagy (Digaleh et al., 2013). Nrf2 is not glycosylated and does not require NGLY1 for its function. We thus hypothesized that Nrf1 regulates mitophagy through a transcriptional program, and that NGLY1 deficiency causes cytosolic retention of Nrf1 due to misprocessing, which then leads to impaired mitophagy.

Nrf1 is usually undetectable at steady-state due to continuous degradation by the proteasome. After proteasome inhibitor bortezomib treatment, processed Nrf1 was observed in WT but not Ngly1^−/− cells (Fig. 5 A). Expression of proteasome subunit genes was decreased in Ngly1^−/− compared with WT cells (Fig. 5 B). After bortezomib treatment, proteasome subunit gene expression increased in WT but not Ngly1^−/− cells, consistent with Ngly1^−/− cells lacking functional Nrf1 (“processed”) and proteasome “bounce-back” response (Fig. 5 B).

Since global deletion of Ngly1 is embryonic lethal, we generated loxP-flanked Ngly1^fl/fl mice and bred to LysM-Cre to generate myeloid-specific Ngly1-deficient mice. Ngly1^fl/flLysM-Cre mice appear healthy, with oldest ones approaching 3 mo of age. Ngly1 protein in Ngly1^fl/flLysM-Cre bone marrow–derived macrophages (BMDMs) was markedly reduced compared with littermate Ngly1^fl/fl (Fig. 5 C). Similar to Ngly1^−/− MEFs, Ngly1^fl/flLysM-Cre BMDMs also showed reduced amount of processed Nrf1 after bortezomib treatment (Fig. 5 D). Ngly1^fl/flLysM-Cre BMDMs showed a global decrease in mitophagy-related genes (Fig. 5 E) and proteasome subunit genes (Fig. 5 F). Ngly1 processing of Nrf1 is critical for cell to withstand proteotoxic stress (Tomlin et al., 2017). We treated Ngly1^fl/fl and Ngly1^fl/flLysM-Cre BMDMs with increasing amount of proteasome inhibitor bortezomib or carfilzomib to induce proteotoxic stress, and we found that Ngly1^fl/flLysM-Cre BMDMs are more susceptible to cell death (Fig. 5 G). These data suggest that Ngly1 regulates transcriptional programs of proteasome and mitophagy through Nrf1 in vitro and ex vivo. We also measured ISG expression in primary peritoneal macrophages from Ngly1^fl/flLysM-Cre mice, and we found that expressions of several ISGs were significantly increased in Ngly1 knockout primary peritoneal macrophages (Fig. 5, H and I).

To further test the role of Nrf1 in mitophagy defect in Ngly1^−/− cells, we stably expressed full-length or N-terminus–deleted Nrf1 (Nrf1-ΔN; Fig. 6 A), the latter of which lacks ER-anchoring sequence and mimics the cleaved and transcriptionally active fragment (Widenmaier et al., 2017). Western blot analysis showed reduced processing of ectopically expressed Nrf1 in Ngly1^−/− cells compared with WT cells and no difference for Nrf1-ΔN (Fig. 6 B), suggesting that Nrf1-ΔN is no longer regulated by NGLY1. Endogenous Nrf1 was undetectable by immunofluorescence microscopy in untreated cells. After bortezomib treatment, in vector-transduced cells, endogenous Nrf1 accumulated in the nucleus of WT cells while a substantial amount of Nrf1 remained in the cytosol in Ngly1^−/− cells (Fig. 6 C). Stable expression of full-length Nrf1 increased nuclear presence of Nrf1 in WT cells and to a lesser extent in Ngly1^−/− cells. Nrf1-ΔN was readily detectable in the nucleus of both WT and Ngly1^−/− cells even in the absence of bortezomib treatment (Fig. 6 C). Similar to the extent of nuclear localization, Nrf1 expression increased proteasome subunit gene expression in Ngly1^−/− cells, although the extent of rescue is less potent compared with Nrf1-ΔN (Fig. 6 D). We further examined mitophagy-related gene expression, and found that Nrf1 or Nrf1-ΔN expression restored many mitophagy-related gene expression in Ngly1^−/− cells with Nrf1-ΔN being more potent than full-length Nrf1 (Fig. 6, E and F). Mitochondria fragmentation phenotype of Ngly1^−/− cells was also partially rescued by Nrf1 or Nrf1-ΔN expression (Fig. 6, G and H). Moreover, elevated ISGs were also decreased in Nrf1 or Nrf1-ΔN–expressing Ngly1^−/− cells (Fig. 6, I and J). Collectively, these data suggest that NGLY1 regulates mitochondrial homeostasis through transcription factor Nrf1. Our data also suggest that boosting nuclear activity of Nrf1 by using a transcriptionally active fragment Nrf1-ΔN can overcome the lack of NGLY1 activity and rescue both mitochondria and immune detects of Ngly1^−/− cells.

Both Nrf1 and Nrf2 bind to similar cis-regulatory anti-oxidant response element (ARE) and transcriptionally activate expression of proteasome subunit gene targets. However, these two transcription factors differ in response cues and protein processing. Nrf1 responds to proteotoxic stress, while Nrf2 responds to oxidative stress. Nrf1 resides on the ER and requires deglycosylation and cleavage for activation, while Nrf2 resides in the cytosol and does not required NGLY1-mediated deglycosylation to become active (Kensler et al., 2007). Under normal condition, Nrf2 binds to cysteine-rich Keap1, which mediates ubiquitination and degradation of Nrf2. Pharmacological compounds (e.g., Keap1 inhibitors, also known as Nrf2 inducers) have been identified to disrupt that interaction to enhance Nrf2 activity. Therefore, we next explored whether we can exploit Nrf2 biology and harness existing Nrf2 inducing compounds for rescue of Ngly1^−/− cells. We first stably expressed Nrf2 in either WT or Ngly1^−/− cells. Nrf2 protein accumulated at similar levels in both WT or Ngly1^−/− cells at steady-state, and both increased to similar levels after bortezomib treatment (Fig. 7 A). Enhanced Nrf2 expression also led to exclusive nuclear localization in both WT and Ngly1^−/− cells (Fig. 7 B). Ngly1^−/− cells transduced with Nrf2 showed increased expression of mitophagy-related genes (Fig. 7, C and D), reduced mitochondrial fragmentation (Fig. 7, E and F), and decreased expression of ISGs (Fig. 7 G). Enhanced Nrf2 expression as well as Nrf1-ΔN expression also reduced cGAS-bound mtDNA in Ngly1^−/− cells (Fig. S5, A and B). These data indicate that enhancing Nrf2 expression can bypass the need for NGLY1, activate transcriptional program of mitophagy and restore mitochondrial and immune homeostasis in Ngly1^−/− cells.

Lastly, we tested whether inducing Nrf2 activity by targeting Keap1 could rescue mitochondrial defects in Ngly1^−/− cells (Fig. 8 A). Naturally derived Keap1 inhibitor sulforaphane (a natural product) robustly increased Nrf2 protein level in both WT and Ngly1^−/− cells (Fig. 8 B). Expression of proteasome subunit genes and mitophagy-related genes were broadly enhanced in Ngly1^−/− cells after sulforaphane treatment (Fig. 8, C and D). Sulforaphane treatment also markedly decreased fragmented mitochondria (Fig. 8, E and F), expression of ISGs (Fig. 8 G), and cGAS-bound mtDNA (Fig. 8 H) in Ngly1^−/− cells. These data demonstrate a novel therapeutic approach and proof-of-concept for harnessing pharmacological Nrf2 inducers to correct abnormalities of NGLY1 disease (a model in Fig. 9).

NGLY1 deficiency is associated with a rare congenital disorder leading to global developmental delay and neurological abnormalities. The mechanism of disease is unclear and no effective therapy is available. We discovered that NGLY1 deficiency impairs mitophagy, leading to severely fragmented mitochondria and reduced mitochondrial function. These damaged mitochondria release mtDNA and possibly also mtRNA into the cytosol, which activates cytosolic innate immune DNA- and RNA-sensing pathways, respectively, leading to chronic activation of type I IFN signaling. Depletion of mtDNA using ethidium bromide decreases ISG expressions in Ngly1^−/− cells, suggesting mtDNA contributes to innate immune activation. Interestingly, ethidium bromide treatment also slightly increased baseline ISGs expression in WT cells likely due to other mitochondrial stresses that may indirectly activate immune signaling. The ability of mtDNA to activate the cGAS–STING DNA-sensing pathway has been largely confined to acute external stimulations so far such as inhibition of caspases (White et al., 2014) and Dengue virus infection (Aguirre et al., 2017). mtRNA and the cytosolic RNA-sensing pathways have not been analyzed carefully in these settings. In our study, we detected strong induction of ISGs in Ngly1^−/− cells and less robust but significant increase of cGAS-bound mtDNA, suggesting that other endogenous ligands beside mtDNA also contribute to innate immune signaling activation. In fact, a previous study demonstrated that mtRNA is the most potent immunostimulatory mammalian RNA type (Karikó et al., 2005). It is thus possible that impaired mitophagy and damaged mitochondria in Ngly1^−/− cells lead to leakage of both mtDNA and mtRNA into the cytoplasm that activate DNA- and RNA-sensing pathways, respectively. Importantly, we present genetic evidence for the involvement of both DNA- and RNA-sensing pathways by breeding Ngly1^−/− mice to Sting^−/− and Mavs^−/− mice. The exact mechanism for RNA-sensing pathway activation in Ngly1^−/− cells and potential involvement of mtRNA require future investigation.

Unchecked hyperactivation of type I IFN signaling has been described in several monogenic autoimmune and autoinflammatory diseases that are collectively known as type I interferonopathies (Crow and Manel, 2015). Clinical features of type I interferonopathies are diverse, involving multiple organs depending on underlying gene mutations, although there is a striking overlap of involvement of central nervous system (CNS) and the skin (Rodero and Crow, 2016). The most severe example is AGS (associated with mutations in nucleases TREX1, RNASEH2A, and others), which is a neuro-inflammatory disease characterized by excess IFN-α in the cerebrospinal fluid (CSF) and brain calcification detected by magnetic resonance imaging (Rodero and Crow, 2016). Increased IFN signaling in SLE is also recently reported to drive synapse loss of neurons, providing an explanation for the neurological symptoms observed in some SLE patients (Bialas et al., 2017). Abnormal brain imaging, progressive loss of neurons, and evidence of neuropathy have been reported for many NGLY1 patients (Lam et al., 2017), although NGLY1 deficiency has not been formally considered as an autoinflammatory disease. IFNs and inflammatory cytokines also have not been measured in the CSF of NGLY1 patients in limited clinical studies to date. However, “odd” resistance to common viral infections in NGLY1-deficient children has been noted by parents, and is consistent with our findings of elevated ISGs and reduced viral replication in Ngly1^−/− cells.

AGS gene deficiencies in mouse present drastically different survival phenotypes, which can be rescued to various extents after ablation of the underlying innate immune signaling pathway. For example, Trex1^−/− mice develop systemic autoinflammation due to the cGAS–STING pathway activation with an average survival of 2–3 mo; Sting^−/− fully rescues the overall survival (Gall et al., 2012; Ahn et al., 2014). In contrast, we showed that RNASEH2A-G37S mice also chronically activate the cGAS–STING pathway; however, these mice are still-born lethal and Sting^−/− completely eliminates the immune signaling, but only rescues 2% of the lethality (Pokatayev et al., 2016). Other functions of RNASEH2, such as maintaining genome stability, may be the main cause of lethality in mice. We found here that Ngly1^−/− mice on the C57BL/6 background is still-born lethal, chronically activates the cGAS–STING pathway, and Sting^−/− ablates the innate immune gene signature but does not rescue the overall lethality. Ngly1^−/−-mediated immune activation occurs downstream of mitochondrial defects. NGLY1 also regulates proteostasis (Tomlin et al., 2017) and mitochondrial homeostasis (this study), both of which could be important for embryonic development and survival in mice. Nonetheless, molecular immune features of the NGLY1 disease discovered here could be important for clinical features in human patients and need to be further investigated.

Our study also uncovers an unexpected connection of transcription factor Nrf1 in regulating mitophagy. Nrf1 is known for its role in regulating proteasome subunit gene expression during proteotoxic stress. Whole body Nrf1 knockout is embryonic lethal, highlighting the importance of Nrf1 in early development. CNS-specific Nrf1 knockout causes progressive motor neuron dysfunctions (Kobayashi et al., 2011), a clinical feature also observed in NGLY1 patients (Enns et al., 2014; Lam et al., 2017). Two recent studies using tissue-specific Nrf1 knockouts also implicated Nrf1 in cholesterol homeostasis and brown adipose tissue thermogenic adaptation (Widenmaier et al., 2017; Bartelt et al., 2018), suggesting an expanded view of Nrf1 beyond its sole function in proteasome bounce-back response. Brown adipose tissue–specific Nrf1 knockout causes mitochondrial damage and tissue inflammation (Bartelt et al., 2018), which remarkably mirror the immune activation and mitochondrial defects we observed in Ngly1^−/− cells and Ngly1-deleted myeloid tissues. Clinically, NGLY1 patients show abnormal mitochondrial physiology in muscle and liver (Kong et al., 2018). An integrated meta-transcriptome analysis of mitochondrial respiratory chain disease also identified NGLY1 as one of the most dysregulated genes (Zhang and Falk, 2014; Kong et al., 2018). NGLY1 is essential for Nrf1 deglycosylation and activation (Tomlin et al., 2017). Thus, we propose that the NGLY1–Nrf1 axis regulates proteostasis and mitochondrial homeostasis with a broad impact in inflammation, metabolism, and tissue health.

As for therapeutic strategy, we showed that ectopic expression of a transcriptionally active nuclear form of Nrf1 (that bypass the need for processing) potently rescues Ngly1^−/−-associated mitochondrial defects. Better yet, pharmacologically enhancing Nrf2 activity, which boosts expression of mitophagy-related genes, also restored mitochondrial integrity and suppressed immune activation in Ngly1^−/− cells. Although both Nrf1 and Nrf2 bind to ARE cis-regulatory element to induce target gene transcription, genetic studies show that Nrf1^−/− is embryonic lethal, while Nrf2 is dispensable for development of mice, clearly demonstrating that two transcription factors have unique target genes despite sharing overlapping ones. Nrf2 is undetectable at steady-state due to proteasome degradation mediated by E3 ligase adaptor protein Keap1, which could be the reason why Nrf2 could not compensate for lack of functional Nrf1 in Ngly1^−/− cells. Taking advantage of the Keap1-Nrf2 biology as well as existing pharmacological compounds that inhibit Keap1, we provide important proof-of-principle evidence that one such compound almost fully rescued mitochondrial and immune defects in Ngly1^−/− cells. Together, our findings not only reveal novel functions of NGLY1–Nrf1 axis in mitochondrial homeostasis and inflammation; we also demonstrate that Nrf2-enhancing drugs could have exciting potential for treating NGLY1 patients.

BX795 was purchased from Invivogen. CAY10576, BAY 11–7082, and Ruxolitinib were from Selleck Chemicals. Bortezomib and carfilzomib were from Cayman Chemical. Sulforaphane, cycloheximide, actinomycin D, oligomycin, and antimycin were obtained from Sigma-Aldrich. Mouse monoclonal Ngly1 antibody was a gift from A. Zuberi (Jackson Laboratories). Other antibodies used as follows: anti-STING (13647; Cell Signaling, 1:1,000 dilution), phospho-STING (Ser365; D8F4W) Rabbit mAb (72971; Cell Signaling; 1:1,000 dilution), anti-cGAS (31659; Cell Signaling; 1:1,000 dilution), anti–RIG-I (3743; Cell Signaling; 1:1,000 dilution), anti-MDA5 (5321; Cell Signaling; 1:1,000 dilution), anti-tubulin (B-5-1-2; Sigma-Aldrich; 1:20,000 dilution), anti-HMGB1 (ab18256; Abcam; 1:2,000 dilution), anti-GAPDH (2118; Cell Signaling; 1:2,000 dilution), anti-SQSTM1/p62 (ab56416; Abcam; 1:1,000 dilution), anti-phosphorylated p70 S6K (Thr389; 9205; Cell Signaling; 1:1,000 dilution), anti-p70 S6K (2708; Cell Signaling; 1:1,000 dilution), anti–phospho-ATM (Ser1981) antibody (clone 10H11.E12; EMD Millipore; 1:1,000 dilution), phospho-Histone H2A.X (Ser139) antibody (2577; Cell Signaling; 1:1,000 dilution), Nrf1 (8052; Cell Signaling; 1:1,000 dilution), Nrf2 (12721; Cell Signaling; 1:1,000 dilution).

Ngly1^+/− C57BL/6J mice were purchased from the Jackson Laboratories (C57BL/6J-Ngly1^em4Lutzy/J, Stock No. 027060). Sting^−/− mice were obtained from G. Barber (University of Miami, Miami, FL) and Mavs^−/− mice were from Z. Chen (UT Southwestern Medical Center, Dallas, TX). Infar^−/− mice were from Taconic Biosciences. All mice were housed in specific pathogen-free animal facility of UT Southwestern Medical Center. Genotyping of Sting^−/−, Mavs^−/−, and Infar1^−/− were performed as described previously (Pokatayev et al., 2016). Ngly1^−/− mice were genotyped using Taqman probes. Primers and probes used were as follows: Fwd, 5′-TGTGTTCTGTAACTGATTCCAGGT-3′; Rev, 5′-TCTTCTTTAACCTTGGTCCTCCT-3′; WT probe, Hex-5′-TCACCTGGCGATACTCTTGTAAA-3′-Black Hole Quencher 1; Ngly1 mutant probe, 6-FAM-5′TTGATGTCACCTTGTAAACATGATG-3′-Black Hole Quencher 1. Ngly1^fl/fl C57BL/6J mice were gifts from T. Suzuki (RIKEN, Japan) and crossed with hemizygous B6.129P2-Lyz2^tm1(cre)Ifo/J transgenic mice (Jackson Laboratories) expressing LysM promoter driven Cre recombinase to generate myeloid-specific Ngly1-deficient mice. Ngly1^fl/fl mice were genotyped using PCR with the following primers: Fwd, 5′-GCATCACTGTCATCAGAAGCGT-3′; Rev, 5′-TGACCCACATTCAAACTTCCAGG-3′. Ngly1^fl/flLysM-Cre mice and Ngly1^fl/fl littermate controls were used for experiments. Animal protocol was approved by the Institutional Animal Care and Use Committee at UT Southwestern Medical Center (assurance no. APN2017-101968).

WT and Ngly1^−/− MEFs were gifts from Dr. T. Suzuki (RIKEN, Japan) and generated and maintained as described previously (Huang et al., 2015). NGLY1-deficient patient fibroblasts (GM25343 and GM25990) and lymphoid cell lines (GM25347 and GM25348) and healthy control cells were purchased from Coriell Institute. HeLa cells stably expressing Parkin (HeLa/Parkin) was generated and maintained as described previously (Sumpter et al., 2016). Primary MEFs were isolated from E13.5 embryos of indicated genetic background and maintained in DMEM supplemented with 20% FBS. BMDMs were generated as described previously (Hasan et al., 2013). Peritoneal macrophages were obtained from peritoneal cavity lavage of 6–8-wk-old Ngly1^fl/flLysM-Cre and Ngly1^fl/fl littermate control mice. All cells used in the study were tested negative for mycoplasma contamination.

Ngly1, FLAG-tagged cGAS, FLAG-tagged Nrf1, FLAG-tagged N-ternimus–deleted Nrf1 (Nrf1-ΔN), and Nrf2 were cloned into retroviral pMRX-ires-bsr vector (a gift from S. Akira) using EcoRI and NotI sites. Enzymatically inactive C306A mutant Ngly1 was generated by site-directed mutagenesis (Agilent Technologies). Retroviruses were packaged in HEK293T cells and used for transduction followed by selection with 15 µg/ml blasticidine. For stable RNA interference, shRNA oligos were synthesized and cloned into a pLKO.1-TRC cloning vector (Addgene). Lentiviral particles were packaged in Lenti-X 293T cells following standard protocol. Cells were transduced with shRNA-harboring lentiviruses and selected with puromycin (2 µg/ml) for several days before subsequent analysis. Ngly1 shRNA target sequences are as following: sh-Ngly1-1, 5′-TTAGGGTTTGAAGCTCGATAT-3′; sh-Ngly1-2, 5′-GCTCCAATGGTTGTTGGTGAT-3′. Sting and cGas shRNA target sequences are used as reported previously (Pokatayev et al., 2016).

Predesigned siRNAs were purchased from Sigma-Aldrich. For transient RNA interference, cells were grown on 12-well plates and reversely transfected with 50 nM siRNA using Lipofectamine RNAiMAX Reagent per the manufacturer’s instruction (Thermo Fisher Scientific). Cells were collected after 48 h and then used for quantitative RT-PCR or Western blot analysis.

To generate NGLY1 knockout THP-1 cells by CRISPR/Cas9-mediated gene editing, guide sequences targeting NGLY1 were designed (Zhang Lab, MIT, crispr.mit.edu [accessed April 26, 2018]) and cloned into LentiCRISPRv2 vector (Sanjana et al., 2014). The targeting guide sequences of NGLY1 were used as following: Target 1 (T1), 5′-CTTGACGGTTCCTTGTGTGC-3′; Target 3 (T3), 5′-ACTAGACTCTTGCCTGTCAG-3′. Lentiviruses carrying sgRNA and Cas9 were generated in Lenti-X 293T cells with packaging plasmids psPAX2 and an envelope plasmid pMD2.G (Addgene). THP-1 cells were transduced with lentiviruses, followed by puromycin (2 µg/ml) selection for several days. Single cell-derived clones were confirmed with Western blot for knockout of NGLY1.

Total RNA was isolated from cultured cells using TRI reagent (Sigma-Aldrich) per manufacturer’s instruction, and cDNA was synthesized with iScript cDNA Synthesis kit (Bio-Rad). iTaq Universal SYBR Green Supermix (Bio-Rad) was used to quantify mRNA expression with CFX96 Real-Time PCR Detection System. Primer sequences used were described previously (Pokatayev et al., 2016) and available upon request. Gapdh was used as internal control to normalized gene expression. Heat map of gene expression was generated using Multiple Experiment Viewer (MeV4).

Western blots were performed as described previously (Hasan et al., 2013). In brief, cell lysate was quantified using bicinchoninic acid assay, and equal amount of proteins were separated on SDS-polyacrylamide gel and transferred to nitrocellulose membrane. Membranes were blocked with 5% nonfat milk in 1× TBS-T and incubated with diluted primary antibodies at 4°C overnight per manufacturers’ instructions. Membranes were incubated with HRP-conjugated secondary antibody (Bio-Rad) diluted for 1 h at room temperature. SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific) was used to develop the blots on film.

Cells were grown on glass coverslips and treated as indicated. Cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and blocked with 5% normal donkey serum. Slides were incubated with primary antibody for 1 h at room temperature followed by fluorescent-conjugate secondary antibody incubation, and mounted with VECTASHIELD Mounting Media containing DAPI (Vector Laboratories). Primary antibodies used were mouse monoclonal anti-HSP60 (sc-13115; Santa Cruz; 1:50 dilution), mouse anti-DNA (CBL186; Millipore; 1:25 dilution), rabbit anti-Nrf1 (8052; Cell Signaling; 1:100 dilution), and rabbit anti-Nrf2 (12721; Cell Signaling; 1:100 dilution). Secondary antibodies were donkey anti-mouse IgG Alexa Fluor 488 (A21202; Invitrogen; 1:1,000 dilution), donkey anti-rabbit IgG Alexa Fluor 488 (A21206; Invitrogen; 1:1,000 dilution), and anti-mouse IgM Alexa Fluor 488 (406522; BioLegend; 1:500 dilution). Slides were visualized with a Zeiss AxioImager Z2 microscope equipped with AxioVision software. Mitochondrial morphology was categorized and enumerated as reported (Chen et al., 2003).

Cells were treated with ethidium bromide (150 ng/ml; Sigma-Aldrich) for 6 d. During ethidium bromide treatment, cells were maintained in DMEM supplemented with 10% FBS, 1 mM sodium pyruvate, and 100 µg/ml uridine. To measure mtDNA depletion efficiency, cellular DNA was isolated using QIAamp DNA mini kit, and mitochondrial ND4 gene was quantified using real-time PCR and normalized to genomic GAPDH gene. The primer pair used is as following: ND4 forward, 5′-AACGGATCCACAGCCGTA-3′; ND4 reverse, 5′-AGTCCTCGGGCCATGATT-3′. mtDNA-depleted cells were replated, cultivated overnight, and total RNA was extracted for ISGs expression measurement by real-time RT-PCR.

Mitophagy assay and automated quantification were performed as described previously (Wei et al., 2017). In brief, HeLa/Parkin cells were treated with a combination of two specific mitochondrial respiration inhibitor oligomycin (2.5 µM) and antimycin (250 nM) for 16 h. Cells were fixed and stained with primary antibody as above. For following cell segmentation, cells were stained with CellMask Deep Red Plasma membrane stain (C10046; Invitrogen; 1:2,000 dilution) during secondary antibodies incubation. Z-stack images were acquired with Zeiss AxioImager Z2 microscope equipped with a Photometics CoolSnap HQ2 camera using 20×/0.8 numerical aperture air. The same acquisition parameters were used for samples stained with identical primary antibody. Z stacks were further deconvolved with AutoDeBlur (Bitplane) before automated quantification of mitophagy using Imaris version 8.2 (Bitplane). Cell segmentation was performed based CellMask signal, and all cells touching the border of image were excluded with additional filtering. The same parameters were applied to all images within a given experiment using Imaris Batch. Data for number of cytoplasmic HSP60 or mtDNA vesicle were exported for further statistical analysis.

FLAG-cGAS IP followed by quantitative PCR was performed as described previously (White et al., 2014). In brief, WT and Ngly1^−/− MEFs were transduced or transiently transfected with a plasmid encoding FLAG-tagged cGAS. Dynabeads Protein G (10004D; Thermo Fisher Scientific) were conjugated with mouse monoclonal anti-FLAG M2 antibody (F1804; Sigma-Aldrich) or isotype mouse-IgG (I5381; Sigma-Aldrich) at 4°C overnight. Immunoprecipitation was performed following standard protocol. Before DNA extraction, a fraction of beads was eluted using loading buffer and boiled for 5 min to confirm FLAG-cGAS precipitation by Western blot. Protein G beads were treated with 200 µg/ml protease K (Ambion) at 55°C for 30 min to release bound DNA. Co-precipitated DNA were examined by quantitative PCR and normalized to IP FLAG-cGAS. Primer sequences of mitochondrial genes are used as described previously (West et al., 2015).

For real-time analysis of the OCR, cells were seeded in Seahorse 24-well XF Cell Culture Microplates. Real-time changes in OCR were analyzed with XF-24 extracellular flux analyzer (Seahorse Bioscience) during subsequent sequential treatment with oligomycin (inhibitor of ATP synthase), FCCP (carbonyl cyanide 4-(trifluoromethoxy) phenylhydrazone), and rotenone (inhibitors of the electron-transport chain) according to the manufacturer’s instruction. The metabolism assay was performed at UT Southwestern Medical Center Metabolic Phenotyping Core.

Graphpad Prism was used for statistical analysis. Statistical tests performed were indicated in figure legend. Numerical data were shown as mean ± SEM. P values of less than 0.05 were considered statistically significant.

Supplemental figures associated with this study include phosphorylation of STING (Fig. S1), mtDNA depletion and DNA damage response (Fig. S2), mitochondrial morphology in Ngly1^−/−Sting^−/− and Ngly1^−/− Mavs^−/− MEFs (Fig. S3), mitochondrial fusion (Fig. S4), and cGAS-bound mtDNA in Ngly1-deficient cells (Fig. S5).

We thank Aamir Zuberi (Jackson Labs) for NGLY1 antibody, Beth Levine (UT Southwestern) for mitophagy assay reagents and technical advice, Prashant Mishra (UT Southwestern) for mitochondrial fusion assay advices, and Kevin Lee and Matt Wilsey at Grace Science Foundation for communications and coordination of reagent sharing.

This work is supported by UT Southwestern Medical Center Endowed Scholar Award (N. Yan), Grace Science Foundation (N. Yan and T. Suzuki), and Mr. Hiroshi Mikitani (T. Suzuki).

The authors declare no competing financial interests.

Author contributions: N. Yan and K. Yang conceived of and designed the study. K. Yang performed most experiments and analyzed data. R. Huang assisted with mouse breeding, including crosses between Ngly1 mice and Sting^−/−, Mavs^−/− and Infar1^−/− mice. T. Suzuki and H. Fujihira provided critical reagents and generated the Ngly1-flox mice. N. Yan supervised the study. N. Yan and K. Yang wrote the paper.