MLKL-USP7-UBA52 signaling is indispensable for autophagy in brain through maintaining ubiquitin homeostasis

ABSTRACT

Individuals with genetic elimination of MLKL (mixed lineage kinase domain like pseudokinase) exhibit an increased susceptibility to neurodegenerative diseases like Alzheimer disease (AD). However, the mechanism is not yet fully understood. Here, we observed significant compromise in macroautophagy/autophagy in the brains of mlkl knockout (KO) mice, as evidenced by the downregulation of BECN1/Beclin1 and ULK1 (unc-51 like autophagy activating kinase 1). We identified UBA52 (ubiquitin A-52 residue ribosomal protein fusion product 1) as the binding partner of MLKL under physiological conditions. Loss of Mlkl induced a decrease in ubiquitin levels by preventing UBA52 cleavage. Furthermore, we demonstrated that the deubiquitinase (DUB) USP7 (ubiquitin specific peptidase 7) mediates the processing of UBA52, which is regulated by MLKL. Moreover, our results indicated that the reduction of BECN1 and ULK1 upon Mlkl loss is attributed to a decrease in their lysine 63 (K63)-linked polyubiquitination. Additionally, single-nucleus RNA sequencing revealed that the loss of Mlkl resulted in the disruption of multiple neurodegenerative disease-related pathways, including those associated with AD. These results were consistent with the observation of cognitive impairment in mlkl KO mice and exacerbation of AD pathologies in an AD mouse model with mlkl deletion. Taken together, our findings demonstrate that MLKL-USP7-UBA52 signaling is required for autophagy in brain through maintaining ubiquitin homeostasis, and highlight the contribution of Mlkl loss-induced ubiquitin deficits to the development of neurodegeneration. Thus, the maintenance of adequate levels of ubiquitin may provide a novel perspective to protect individuals from multiple neurodegenerative diseases through regulating autophagy.Abbreviations: 4HB: four-helix bundle; AAV: adeno-associated virus; AD: Alzheimer disease; AIF1: allograft inflammatory factor 1; APOE: apolipoprotein E; APP: amyloid beta precursor protein; Aβ: amyloid β; BECN1: beclin 1; co-IP: co-immunoprecipitation; DEGs: differentially expressed genes; DLG4: discs large MAGUK scaffold protein 4; DUB: deubiquitinase; EBSS: Earle’s balanced salt solution; GFAP: glial fibrillary acidic protein; HRP: horseradish peroxidase; IL1B: interleukin 1 beta; IL6: interleukin 6; IPed: immunoprecipitated; KEGG: Kyoto Encyclopedia of Genes and Genomes; KO: knockout; MAP1LC3/LC3: microtubule associated protein 1 light chain 3; MLKL: mixed lineage kinase domain like pseudokinase; NSA: necrosulfonamide; OPCs: oligodendrocyte precursor cells; PFA: paraformaldehyde; PsKD: pseudo-kinase domain; SYP: synaptophysin; UB: ubiquitin; UBA52: ubiquitin A-52 residue ribosomal protein fusion product 1; UCHL3: ubiquitin C-terminal hydrolase L3; ULK1: unc-51 like autophagy activating kinase 1; UMAP: uniform manifold approximation and projection; UPS: ubiquitin-proteasome system; USP7: ubiquitin specific peptidase 7; USP9X: ubiquitin specific peptidase 9 X-linked.

Introduction

AD is a complex neurodegenerative disorder that primarily affects the elderly. APOE (apolipoprotein E) ε4 is recognized as a major genetic risk factor for late-onset AD [1]. We previously identified a novel heterozygous variant in MLKL gene, leading to the decay of MLKL mRNA in an APOE ε4-negative Chinese AD population [2]. Two years later, Faergeman and coworkers reported a significant discovery regarding MLKL in the context of neurodegenerative disorders. They identify a homozygous frameshift mutation in MLKL that encoded a truncated (and ultimately degraded) MLKL protein from two brothers in a British family [3]. The mutation carriers display significant neurodegenerative symptoms, including acute global cerebral volume loss, and atrophy in both the cerebellum and brainstem, accompanied by cognitive impairment [3]. These findings highlight the critical role of MLKL in the central nervous system and its involvement in neurodegenerative diseases, including AD. However, the molecular basis of this relationship remains unclear and requires further investigation.

MLKL is known as the downstream effector of necroptosis [4,5]. As a pseudokinase, MLKL consists of an N-terminal four-helix bundle (4HB) domain plus a brace region (amino acids 1–180) and a C-terminal pseudo-kinase domain (PsKD) (amino acids 181–471) [6,7]. The 4HB domain is responsible for inducing membrane permeabilization, while the PsKD normally prevents the 4HB domain from executing necroptosis. However, upon receptor interacting serine/threonine kinase 3 (RIPK3)-mediated phosphorylation, the PsKD undergoes a conformational change, releasing the constrain on the 4HB domain and allowing it to initiate necroptosis [6,8–14]. The two-helix brace region is crucial for MLKL oligomerization and acts as a linker between the 4HB domain and the PsKD [7,15,16]. In addition to necroptosis, it has been observed that MLKL also regulates macroautophagy/autophagy in vitro [17–19] and is involved in the sorting of endocytosed proteins and bacteria [20–22]. However, the precise function of MLKL in non-necroptotic cellular events is not yet fully understood.

Aberrant autophagy has been increasingly recognized as a significant factor in the pathogenesis of neurodegenerative diseases, including AD [23–26]. A recent study utilizing a transgenic mouse model expressing a specific neuronal RFP-GFP-LC3 reporter revealed that impaired autolysosomal acidification plays a crucial role in neuronal amyloid β (Aβ) generation [27]. Several key autophagy-related molecules, including BECN1/Beclin1, PIK3C3/VPS34 (phosphatidylinositol 3-kinase catalytic subunit type 3), ULK1, SQSTM1/p62 (sequestosome 1) and MAP1LC3/LC3 (microtubule associated protein 1 light chain 3), have been found to be dysregulated in the brains of AD patients [28–32]. Dysfunctional autophagy has been implicated in the pathogenesis of AD [33–37], while enhanced autophagy has been shown to ameliorate AD pathologies [38–42].

To understand why MLKL loss confers susceptibility to neurodegeneration, we initially investigated whether autophagy in the brain is altered in the absence of Mlkl. Our results showed that the absence of Mlkl led to a significant inhibition of autophagy in brain, evidenced by the down-regulation of autophagy-initiation proteins BECN1 and ULK1. Furthermore, UBA52, a ubiquitin precursor protein, was identified as a novel MLKL-binding partner under physiological conditions. MLKL was shown to be indispensable for the maintenance of ubiquitin homeostasis by regulating UBA52 processing, which was mediated by the DUB USP7 in brain. Loss of Mlkl led to ubiquitin deficits, and BECN1 and ULK1 levels were accordingly reduced because of diminished lysine 63 (K63)-linked polyubiquitination. Furthermore, single-nucleus RNA sequencing (snRNA-seq) of hippocampus indicated that Mlkl loss-disturbed signaling pathways were correlated with multiple neurodegenerative diseases, including AD. Consistently, mlkl KO mice displayed marked cognitive impairment, and AD pathologies were exacerbated in an AD mouse model with mlkl deletion. Collectively, our study reveals that, in addition to necroptosis, MLKL is physiologically required to maintain ubiquitin homeostasis, which is executed through interaction with UBA52 and USP7, to support functional autophagy in the brain. Loss of Mlkl contributes to neurodegeneration, including AD.

Results

Autophagy is suppressed in the absence of Mlkl

Previous studies have suggested that the role of MLKL in autophagy may be tissue-dependent, as inconsistent results have been reported in cell lines derived from blood vessel, skin and liver [17–19]. Here, to validate the physiological function of MLKL in autophagy, we firstly established a stable Mlkl knockdown (KD) model in the mouse neuroblastoma cell line N2a using lentivirus expressing short hairpin RNA (shRNA) targeting different regions of Mlkl mRNA (Figure 1A). Results showed that the marker of autophagosome, LC3-II, was significantly reduced under normal (nutrient-rich) conditions and Earle’s balanced salt solution (EBSS)-induced amino acid starvation after Mlkl KD (Figure 1B-C). When cells were starved for 3 h, LC3-II levels in Mlkl KD group recovered to levels comparable to those in the scrambled (Scr) control group (Figure 1B), suggesting that autophagosome accumulation was inhibited in the absence of Mlkl but not completely abolished. Additionally, we applied the well-established human MLKL-specific inhibitor necrosulfonamide (NSA) [4] to treat HEK293T cells. LC3-II levels were also significantly declined compared to the vehicle (Veh) control (Figure 1D-E). These findings demonstrate that mlkl deletion or inhibition diminishes autophagosomes in vitro, similar to that reported in mouse dermal fibroblasts/MDFs [17] and hepatocytes [19]. Cell viability was not significantly altered by Mlkl KD or NSA treatment (1 µM) (Figure S1A). Intriguingly, LC3-I was also markedly decreased by Mlkl KD or inhibition (Figure B-E). Unlike LC3-II, which is degraded in lysosomes, LC3-I degradation is mediated by the ubiquitin-proteasome system (UPS) [43,44]. The dysregulation of LC3-I suggests that the UPS may also be disturbed in the absence of Mlkl.

Figure 1.

Autophagy is suppressed in the absence of Mlkl. (A) Immunoblotting of endogenous MLKL in N2a cells stably expressing Mlkl shRNAs (shMlkl) or Scr control. sh1 and sh2, two shRNAs targeting the mRNA of mouse Mlkl. (B) Representative blots of LC3 in Mlkl stably KD N2a cells under nutrient-rich (NR) condition and ebss-induced starvation. (C) Quantification of normalized LC3 (mean ± SEM) under NR or starvation conditions in (B). P values are calculated using one-way ANOVA, three independent experiments. *: P < 0.05, **: P < 0.01. ***: P < 0.001. (D) Representative blots of LC3 in HEK293T cells treated by the MLKL inhibitor NSA or veh control. (E) Quantification of normalized LC3 (mean ± SEM) in (D). P values are calculated using two-tailed t-test, four independent experiments. **: P < 0.01. (F) Analysis of Mlkl mRNA levels in hippocampus of 10-month-old male WT and mlkl KO mice using quantitative reverse transcription PCR (qRT-PCR). Data are mean ± SEM and each data point represents an individual mouse. (G) Representative blots of LC3 in hippocampus from 10-month-old male WT and mlkl KO mice. (H) Quantification of LC3 (mean ± SEM) in mlkl KO and control WT mice in (G). Each data point represents an individual mouse. P values are calculated using two-tailed t-test. *: P < 0.05. ***: P < 0.001. (I) Representative blots of LC3 in Mlkl stably KD N2a cells treated with or without Baf (50 nM, 3 h). (J) Quantification of normalized LC3 in the presence of Baf (mean ± SEM) in (I). P values are calculated using two-tailed t-test, five independent experiments. *: P < 0.05, **: P < 0.01. ***: P < 0.001. (K) Representative images of HT22 cells stably expressing tfLC3 reporter followed by transfection with Mlkl shRNAs (shMlkl, sh1+sh2) or Scr control. The cells were starved for 2 h in EBSS before analysis by confocal microscopy. Scale bar: 20 µm (left), 10 µm (right, enlarged). (L) Quantification of the number of LC3 puncta (yellow puncta: RFP-positive and GFP-positive. Red-only puncta: RFP-positive and GFP-negative) in (K). Quantification was performed across 3 independent experiments, with over 50 cells analyzed for each group (mean ± SEM). P values are calculated using two-tailed t-test. *: P < 0.05. (M) Representative images of LC3 puncta in mouse brains infected by AAV containing tfLC3 reporter and Mlkl shRNAs (shMlkl, sh1+sh2) or Scr control. The mice were starved for 24 h before analysis by confocal microscopy. Scale bar: 200 µm (left), 20 µm (left, enlarged), 10 µm (right, enlarged). (N) Quantification of the number of LC3 puncta (yellow puncta: RFP-positive and GFP-positive. Red-only puncta: RFP-positive and GFP-negative in (M). Quantification was performed from 3 mice, with over 50 cells analyzed for each group (mean ± SEM). P values are calculated using two-tailed t-test. **: P < 0.01.

To investigate the role of MLKL in autophagy in vivo, we established an mlkl KO mouse strain using CRISPR-Cas9 technology (Figure 1F). Likewise, LC3-II was significantly down-regulated in the hippocampus of mlkl KO mice (Figure 1G-H). Besides, LC3-I was also strikingly reduced. Autophagy is a highly dynamic process. Insufficient autophagosome accumulation may result from inhibited autophagic flux or accelerated autophagosome degradation via fusion with lysosomes [45]. Here, we used bafilomycin A₁ (Baf) to block autophagosome-lysosome fusion and then detected LC3-II. Results showed that LC3-II was still reduced after Mlkl KD in the presence of Baf (Figure 1I-J). To further investigate the effects of Mlkl deficiency on autophagic flux, the mouse hippocampal neuronal cell line HT22 [46] was used to establish a stable cell line expressing the tandem fluorescence mRFP-GFP-LC3 (tfLC3) reporter. Confocal microscopy revealed that both the yellow (representing autophagosomes) and the red-only puncta (representing autolysosomes) were significantly reduced in response to Mlkl KD (shMlkl, sh1 and sh2) after 2 h of starvation in EBSS (Figure 1K-L). In addition, to dissect the autophagic flux in brain neurons, we introduced tfLC3 complementary DNA (cDNA) and Mlkl shRNA (sh1 and sh2) sequences into adeno-associated virus (AAV) vectors. In these constructs, tfLC3 was driven by the CAG promoter, while the shRNAs were driven by the U6 promoter. The constructs were used to generate AAV particles of serotype AAV.CAP-B10 given their ability to cross the blood brain barrier and high specificity to neurons [47,48]. Purified virus containing tfLC3 (AAV-tfLC3) and Mlkl shRNAs (shMLKL, sh1 and sh2) or Scr control were intravenously delivered into 3-month-old wild-type (WT) mice via the tail veins. At one-month post-injection, the mouse brains were harvested for immunostaining, with the mice starved for 24 h before collection [49]. Similarly, the imaging showed a significant reduction in autophagosomes (yellow puncta) after Mlkl KD (Figure 1M-N). Taken together, these results indicate that Mlkl loss-induced defective autophagosome accumulation is due to suppressed autophagic flux, which is in accordance with findings in the liver [19]. However, the reason why MLKL is required for autophagosome generation remains to be addressed.

Autophagy-initiation proteins BECN1 and ULK1 are down-regulated after Mlkl loss

To investigate the mechanism by which Mlkl deficiency inhibits autophagosome generation, we screened the molecules responsible for autophagosome formation [45] using brain lysates from mlkl KO mice and WT littermates. Compared to WT controls, the two key proteins for autophagy initiation, BECN1 and ULK1, were constantly down-regulated in the hippocampus of female mlkl KO mice at 5 months (Figure 2A-B), male mice at 6 months (Figure S2A-B), male mice at 10 months (Figure S2C-D), and in the whole brain at postnatal 7 days of age (Figure S2E-F). Consistently, BECN1 and ULK1 were both reduced in Mlkl KD N2a (Figure 2C-D) and in NSA-treated HEK293T cells (Figure 2E-F).

Figure 2.

Autophagy-initiation proteins BECN1 and ULK1 are down-regulated after Mlkl loss. (A) Immunoblotting of autophagy-related proteins in the hippocampus from 5-month-old female mlkl KO mice and WT controls. (B) Quantification of examined targets (mean ± SEM) in (A). Each data point represents an individual mouse. P values are calculated using two-tailed t-test. *: P < 0.05. **: P < 0.01.***: P < 0.001. (C) Representative blots of BECN1 and ULK1 in Mlkl stable KD N2a cells. (D) Quantification of normalized BECN1 and ULK1 (mean ± SEM) in (C). P values are calculated using two-tailed t-test, five (ULK1) or four (BECN1) independent experiments. *: P < 0.05. ***: P < 0.001. (E) Representative blots of BECN1 and ULK1 in HEK293T cells treated by NSA (2 μM) or veh. (F) Quantification of normalized BECN1 and ULK1 (mean ± SEM) in (E). P values are calculated using two-tailed t-test, three (ULK1) or five (BECN1) independent experiments. *: P < 0.05. ***: P < 0.001.

Under physiological conditions, MLKL is primarily a cytosolic protein [50,51]. Our findings revealed a decrease in the levels of two other cytosolic proteins, BECN1 and ULK1, in response to mlkl deletion. We thus speculated that MLKL may exert its effects indirectly through its binding partner. However, few MLKL-interacting proteins in the absence of pro-necroptotic stimuli have been reported [50,51]. Therefore, we next explored physiological MLKL-association proteins using proteomics.

The ubiquitin precursor UBA52 is identified as the MLKL-binding protein by proteomics

To screen candidate proteins associated with MLKL, we constructed two plasmids containing human MLKL cDNA fused with distinct tags, MLKL-FLAG or MLKL-3×HA. Subsequently, the constructs were separately transfected into HEK293T cells. After cell lysis, control immunoglobulin G (IgG) and equal amounts of anti-FLAG or anti-HA antibody were incubated with the cell lysates, respectively. Magnetic protein A/G beads were used to immunoprecipitate the conjugates for mass spectrometry analysis (Figure 3A). In addition to MLKL, 16 candidates were pulled down in both FLAG and HA groups (Figure 3B and Table S1). First, we excluded the weak signals, i.e., less than 1/1000 of the intensity of immunoprecipitated (IPed) MLKL itself. Second, as the 3×HA tag is more potent than the FLAG (1×) tag (Figure 3B), a stronger MLKL signal was detected using anti-HA antibody compared to the anti-FLAG antibody (Table S1). Hence, the intensity of IPed candidates in the HA group should be higher than that in the FLAG group. Accordingly, the candidate list was narrowed down to EIF4A3 (eukaryotic translation initiation factor 4A3) and UBA52. EIF4A3 was excluded from further consideration as its subcellular location (nucleus) is different from MLKL (cytoplasm). Furthermore, EIF4A3 deletion significantly enhances autophagosome generation [52], which is contrary to our results after eliminating Mlkl (Figure 1).

Figure 3.

The ubiquitin precursor UBA52 is identified as the MLKL-binding protein by proteomics. (A) Schematic diagram of screening MLKL binding partners using proteomics. (B) immunoblotting of IPed MLKL using anti-FLAG or anti-HA antibody after transfecting MLKL-FLAG or MLKL-3×HA cDNA into HEK293T cells (upper panel). List of eluted candidates by mass spectrometry (lower panel). See also table S1. (C) Co-IP analysis of synthesized MLKL-3×FLAG and UBA52-3×HA fusion proteins in vitro. (D-E) Co-IP analysis of synthesized MLKL fragments (MLKL[1-180]-3×FLAG, MLKL[181-471]-3×FLAG) and UBA52-3×HA fusion protein in vitro. (F) Input of co-IP samples (C-E) analyzed by immunoblotting. Asterisks indicate target proteins.

Thus, we next focused on the ubiquitin precursor UBA52. BECN1 [53–57] and ULK1 [58–63] are both modified by ubiquitination after translation, suggesting that aberrant ubiquitin signals may affect BECN1 and ULK1 at protein level in the absence of Mlkl. Here, we confirmed the interaction of MLKL and UBA52 using co-immunoprecipitation (co-IP). Since endogenous UBA52 is rapidly processed after synthesis [64,65], the existence of UBA52 as a precursor is very short. MLKL-3×FLAG and UBA52-3×HA fusion proteins were separately synthesized in vitro (Figure 3F) and incubated in lysis buffer followed by co-IP analysis. Results demonstrated that MLKL (FLAG) was detected in the UBA52 (HA) immune complex but not in the control IgG group (Figure 3C). Likewise, UBA52 was pulled down by anti-FLAG (MLKL) antibody (Figure 3C), indicating that MLKL directly associates with UBA52.

MLKL consists of the 4HB domain plus a brace region (residues 1–180) at the N terminus and a PsKD (residues 181–471) at the C terminus [6,7]. To examine the region through which MLKL binds to UBA52, we synthesized the two MLKL fragments fused with 3×FLAG tag (MLKL [1-180]-3×FLAG and MLKL[181-471]-3×FLAG) in vitro (Figure 3F) and separately performed co-IP analysis with purified UBA52-3×HA fusion protein (Figure 3D-E). Our results revealed that MLKL is associated with UBA52 via the PsKD.

Ubiquitin pools are diminished after Mlkl loss

As the binding protein of MLKL, UBA52 is one of the four ubiquitin precursors (UBA52, RPS27A/UBA80 [ribosomal protein S27A], UBB [ubiquitin B] and UBC [ubiquitin C]) in mammalian cells [66,67]. Thus, we next investigated whether ubiquitin levels were altered in the absence of Mlkl. Results demonstrated that both free ubiquitin (Figure 4A-B) and lysine 48 (K48)-linked (Figure 4C-D) and lysine 63 (K63)-linked polyubiquitination (Figure 4E-F), the two most abundant polyubiquitination species in mammalian brains [68], were markedly decreased in Mlkl KD N2a and NSA-treated HEK293T cells (Figure 4G-L). Consistently, ubiquitin levels in hippocampus from mlkl KO mice were also significantly reduced compared with WT littermates (Figure 4M-N). Notably, ubiquitin levels were strikingly reduced by approximately 30% after Mlkl loss (Figure 4M-N). As free ubiquitin is the dominant species of ubiquitin in both mouse (~60%) and human (~80%) brains [69], loss of Mlkl should have markedly disturbed ubiquitin homeostasis in the brain.

Figure 4.

Ubiquitin pools are diminished after Mlkl loss. (A-F) Representative blots of free ubiquitin (UB) (A-B), K48- (C-D) and K63-linked polyubiquitination (K48- and K63-UB) (E-F) in N2a cells with Mlkl KD. Quantification is based on mean ± SEM and data are normalized to Scr group. P values are calculated using two-tailed t-test, three independent experiments. *: P < 0.05. **: P < 0.01. ***: P < 0.001. (G-L) Representative blots of UB (G-H), K48- (I-J) and K63-UB (K-L) in HEK293T cells treated by NSA or veh. Quantification is based on mean ± SEM and data are normalized to veh group. P values are calculated using two-tailed t-test, three independent experiments. *: P < 0.05. **: P < 0.01. ***: P < 0.001. (M-N) Immunoblotting of UB in the hippocampus from 5-month-old female mlkl KO mice and WT controls. Quantification is based on mean ± SEM. Each data point represents an individual mouse. P values are calculated using two-tailed t-test. **: P < 0.01.

MLKL regulates the processing of UBA52 through the DUB USP7 in the brain

UBA52 contains a single ubiquitin at the N terminus and a ribosomal protein L40 at the C terminus. It has been disclosed that UBA52 undergoes rapid deubiquitination to generate a free ubiquitin and L40, soon after its synthesis in ribosomes [64,65]. To elucidate the mechanism by which Mlkl loss disrupts ubiquitin homeostasis, we examined whether MLKL is involved in UBA52 processing. We established a construct containing 3×HA-UBA52-3×FLAG (HA-UBA52-FLAG) fusion cDNA (Figure 5A) and transfected it into HEK293T cells with shRNAs targeting human MLKL or Scr control. Direct detection of the full-length HA-UBA52-FLAG fusion protein using western blotting was difficult, so immunoprecipitation was utilized for enrichment followed by immunoblotting (Figure 5A). Results showed that full-length UBA52 was significantly accumulated upon MLKL loss, with the generation of ubiquitin (3×HA-UB) and L40 (L40-3×FLAG) markedly down-regulated (Figure 5B). Similar results were obtained for the NSA-treated HEK293T cells transfected with HA-UBA52-FLAG cDNA (Figure 5C-D). These findings demonstrate that MLKL is required for normal cleavage of UBA52.

Figure 5.

MLKL regulates the processing of UBA52 through the DUB USP7 in brain. (A) Representative blots of IPed full-length (FL) HA-UBA52-FLAG and HA-UB, L40-FLAG fragments in HEK293T cells co-transfected with HA-UBA52-FLAG cDNA and MLKL shRNAs or Scr control. WCL, whole cell lysate. (B) Quantification of normalized data (mean ± SEM) in (A). P values are calculated using two-tailed t-test, at least three independent experiments. *: P < 0.05. **: P < 0.01. ***: P < 0.001. (C) Representative blots of IPed full-length (FL) HA-UBA52-FLAG and HA-UB, L40-FLAG fragments in HEK293T cells transfected with HA-UBA52-FLAG cDNA followed by NSA treatment (2 μM) or veh. (D) Quantification of normalized data (mean ± SEM) in (C). P values are calculated using two-tailed t-test, three independent experiments. *: P < 0.05. **: P < 0.01. ***: P < 0.001. (E) Representative blots of HA in HEK293T cells co-transfected with UBA52-3×HA cDNA and 3×FLAG-tagged full-length WT MLKL, or MLKL fragments (amino acids 1-180 or 181-471), or MLKL^C86S mutant or empty vector. (F) Quantification of normalized data (mean ± SEM) in (E). P values are calculated using two-tailed t-test, at least six independent experiments. **: P < 0.01. ***: P < 0.001. (G) Immunoprecipitation analysis of synthesized HA-UBA52-FLAG fusion protein incubated with mouse whole brain lysate. (H) Immunoprecipitation analysis of synthesized MLKL-3×FLAG and UBA52-3×HA fusion proteins incubated with mouse whole brain lysate. (I) Co-IP analysis of synthesized MLKL-3×FLAG and UBA52-3×HA fusion proteins with DUBs USP7 and USP9X after incubation with mouse whole brain lysate. (J) Co-IP analysis of synthesized MLKL-3×FLAG and USP7-V5 fusion proteins in vitro. (K) Co-IP analysis of synthesized UBA52-3×HA and USP7-V5 fusion proteins in vitro.

Subsequently, we explored the mechanism by which MLKL modulates UBA52 processing. We separately co-transfected 3×FLAG-tagged full-length MLKL, N-terminal fragment (amino acids 1–180), pseudo-kinase fragment (amino acids 181–471) with 3×HA-tagged UBA52 (UBA52-3×HA) into HEK293T cells (Figure 5E). Compared to empty vector, overexpression of MLKL or its N-terminal fragment drastically diminished the level of UBA52 precursor (Figure 5F). MLKL PsKD also moderately down-regulated full-length UBA52 (Figure 5E-F). These results suggest that MLKL-UBA52 association facilitates UBA52 cleavage to produce ubiquitin and L40, with the N-terminal fragment of MLKL to be the primary effector. The MLKL-specific inhibitor NSA has been proven to target cysteine 86 (C86) of human MLKL protein [4]. To understand by which UBA52 processing is inhibited by NSA, we constructed MLKL mutant with C86 mutated to serine (C86S) and transfected it with UBA52-3×HA into HEK293T cells (Figure 5E and S3). The immunoblotting results demonstrated that UBA52 cleavage is not rescued by the MLKL C86S mutant in comparison to WT MLKL (Figure 5E-F), with an observed increasing tendency in some trials (Figure S3). In contrast, NSA treatment (2 μM) significantly enhanced UBA52 accumulation (Figure 5C-F). This effect may be attributed to the important role of the MLKL PsKD (amino acids 181–471) in facilitating UBA52 processing. Thus, the alteration of one amino acid only in the N-terminal domain (amino acids 1–180) is insufficient to completely reverse this process. This could potentially explain why NSA treatment displays similar effects to Mlkl KD (Figures 1, 2 and 4).

To elucidate the exact role of MLKL in UBA52 processing, we initially evaluated the possibility whether MLKL itself acts as a DUB to directly mediate the deubiquitination of UBA52. Given that C86 is crucial for functional MLKL, we speculated that MLKL may be a DUB of the cysteine protease class [4]. In the catalysis mediated by the cysteine protease type of DUB, histidine residue is also required [70]. Accordingly, we constructed a series of MLKL N-terminal fragment mutants, including all the 4 cysteines (C18, C24, C28 and C86) mutated to serines (4C-S), single-point mutants (C18A, C18S, C28S, C86A and C86S), and all the 3 histidines (H6, H15 and H33) mutated to alanines (3H-A) (Figure S3A). These mutants were transfected individually into HEK293T cells with UBA52-3×HA. Results showed that these mutants had no evident impact on UBA52 processing compared to the WT MLKL N-terminal domain (Figure S3B). This suggests that MLKL itself is unlikely to be a DUB, at least not a cysteine protease type of DUB, which implies that MLKL regulates UBA52 cleavage indirectly through other DUBs.

It has been disclosed that the DUBs responsible for UBA52 processing include UCHL3 (ubiquitin C-terminal hydrolase L3), USP9X (ubiquitin specific peptidase 9 X-linked) and USP7 in HeLa cells and mouse liver [65]. However, the specific DUB that functions in the brain is not known. We firstly conducted immunoprecipitation using synthesized HA-UBA52-FLAG fusion protein mixed with mouse brain lysate. Immunoblotting results indicated that USP9X and USP7, but not UCHL3, were detected in both anti-FLAG and anti-HA immune complexes (Figure 5G), suggesting that USP9X and USP7 associates with UBA52. The target DUB should interact with both MLKL and UBA52. We then incubated MLKL-3×FLAG and UBA52-3×HA fusion proteins with mouse brain lysate (Figure 5H-I). Consistently, immunoprecipitation analysis ruled out the candidate UCHL3 (Figure 5H). Co-IP results further identified USP7 as the DUB associating with MLKL and UBA52 (Figure 5I). To validate whether the interaction is direct or indirect, we synthesized USP7 protein tagged with V5 at the C terminus (USP7-V5), and separately performed co-IP with MLKL-3×FLAG or UBA52-3×HA fusion protein (Figure 5J-K). The results demonstrated that USP7 directly binds to MLKL (Figure 5J) and UBA52 (Figure 5K), indicating that USP7 is the DUB in the brain mediating UBA52 deubiquitination regulated by MLKL.

Mlkl loss-induced autophagic suppression originates from ubiquitin deficits

To investigate the underlying mechanism for the down-regulation in BECN1 and ULK1 and consequent suppression of autophagy upon loss of Mlkl, K63-linked polyubiquitination of BECN1 and ULK1 was examined. Because this modification has been manifested to be essential for their protein stability and thus positively regulates autophagy [71]. BECN1-3×FLAG and 3×HA-K63-UB (HA-K63) cDNAs were co-transfected into N2a cells, followed by Mlkl KD, and immunoprecipitation was conducted using anti-FLAG antibody (Figure 6A). Immunoblotting results showed that IPed HA-K63 was significantly reduced in response to Mlkl KD (Figure 6B), indicating that BECN1-conjugated K63-linked polyubiquitination declined after Mlkl KD. Likewise, K63-linked polyubiquitination associated with ULK1 also decreased following Mlkl KD (Figure 6C-D). These findings indicate that the reduction in BECN1 and ULK1 and subsequent autophagic suppression upon Mlkl deficiency is due to the decline in K63-linked polyubiquitination.

Figure 6.

Mlkl loss-induced autophagic suppression originates from ubiquitin deficits. (A) Representative blots of IPed HA-K63 in N2a cells co-transfected with BECN1-3×FLAG, HA-K63 cDNAs and Mlkl shRNA (sh1) or Scr control. (B) Quantification of normalized IPed HA-K63 (mean ± SEM) in (A). P values are calculated using two-tailed t-test, five independent experiments. ***: P < 0.001. (C) Representative blots of IPed HA-K63 in N2a cells co-transfected with ULK1-3×FLAG, HA-K63 cDNAs and Mlkl shRNA (sh1) or Scr control. (D) Quantification of normalized IPed HA-K63 (mean ± SEM) in (C). P values are calculated using two-tailed t-test, four independent experiments. ***: P < 0.001. (E) Examination of the KD efficiency of shRNAs targeting mouse Uba52 in N2a cells using qRT-PCR (left panel) and representative blots of autophagy-related proteins and UB in N2a cells transfected with validated Uba52 shRNAs (right panel). (F) Quantification of examined targets (mean ± SEM) in (E). P values are calculated using two-tailed t-test, three or four independent experiments. *: P < 0.05. **: P < 0.01. ***: P < 0.001. (G) Representative blots of autophagy-related proteins and UB in N2a cells transfected with mouse Usp7 shRnas. (H) Quantification of examined targets (mean ± SEM) in (G). P values are calculated using two-tailed t-test, three independent experiments. *: P < 0.05. **: P < 0.01. ***: P < 0.001.

To investigate the effect of MLKL-USP7-UBA52 signaling cascade on autophagy, we individually knock down Uba52 and Usp7 using shRNAs targeting different regions of Uba52 and Usp7 mRNA in N2a cells (Figure 6E-H). The KD efficiency of Uba52 was evaluated by qRT-PCR (Figure 6E, left). Immunoblotting analysis demonstrated that UB, LC3-II and BECN1 were reduced upon Uba52 KD (Figure 6E-F). Notably, LC3-I was also shown to be down-regulated. Similarly, UB, LC3-II and BECN1 were significantly diminished in Usp7 KD cells (Figure 6G-H). The levels of ULK1 were not significantly altered by Uba52 KD (Figure 6E-F), which may be attributed to less extent of UB decline (~25%) compared to that in Usp7 KD (~50%, Figure 6G-H) or Mlkl KD (~45%, Figures 1, 2 and 4) cells. Additionally, the hippocampal neuronal cell line HT22 stably expressing tfLC3 was employed to evaluate autophagic flux following Uba52 or Usp7 KD. The results demonstrated that both the yellow (autophagosomes) and the red-only puncta (autolysosomes) were markedly diminished in response to Uba52 KD (shUba52, sh1 and sh2) or Usp7 KD (shUsp7, sh1 and sh2) after 2 h of EBSS starvation (Figure S4). Taken together, these findings indicate that suppression of autophagy induced by Mlkl loss is attributed to ubiquitin deficiency.

Mlkl loss-disrupted signaling pathways are closely correlated with neurodegenerative diseases

According to proteomic studies, thousands of proteins have been shown to undergo ubiquitination [72–74]. Thus, perturbation of ubiquitin homeostasis by Mlkl loss likely affects numerous cellular events. Here, we conducted snRNA-seq of hippocampus from mlkl KO mice and WT littermates. After cell-type characterization, we separately examined the proportion of cell types between mlkl KO and WT groups. Uniform manifold approximation and projection (UMAP) cluster analysis demonstrated that the proportions of astrocytes/Astro, endothelia/Endo, excitatory neurons/ExN, inhibitory neurons/InN, oligodendrocytes/Oligo, oligodendrocyte precursor cells (OPCs) and microglia/Micro were comparable between mlkl KO and WT littermates (Figure 7A-B). We then assessed cell-type transcriptomic changes in the absence of Mlkl. In total, 2,606 differentially expressed genes (DEGs) were identified, including 96 in astrocytes, 14 in endothelia, 1,128 in excitatory neurons, 212 in inhibitory neurons, 827 in oligodendrocytes, 39 in OPCs and 290 in microglia (Figure 7C). Among the 2,606 DEGs, few were altered in all cell types (Figure 7D), suggesting that mlkl deletion results in cell type-specific transcriptomic changes.

Figure 7.

Mlkl loss-disrupted signaling pathways are closely correlated with neurodegenerative diseases. (A-B) UMAP plots (A) and bar plot (B) showed the proportion of the seven major cell types found in middle-aged (14-month-old) male mlkl KO (n = 3) and WT littermates (n = 3) hippocampus samples. (C) Numbers of DEGs between mlkl KO and WT within each cell type (adjusted P < 0.1, log₂ fold change ≥ 0.1 or ≤ −0.1). Down: down-regulated; Up: up-regulated. (D) Venn diagram showed shared DEGs among all seven cell types. The numbers of DEGs specific to each cell type were also shown. (E) KEGG terms enriched among DEGs after Mlkl loss in log (gene expression) > 2 and FDR-adjusted P value < 0.05 between mlkl KO and WT. Only top 10 KEGG terms with P value < 0.05 were listed for each major cell type.

We then annotated the DEGs of different cell types based on Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis (Figure 7E). Results showed that the DEGs in all cell types, except OPCs, were associated with multiple neurodegenerative diseases, including Parkinson disease/PD, AD, Huntington disease/HD, amyotrophic lateral sclerosis/ALS and prion disease. These findings are in accordance with the observation that loss of MLKL increases neurodegenerative vulnerability in human [2,3]. Interestingly, the DEGs in astrocytes, inhibitory neurons and oligodendrocytes were associated with COVID-19 infected coronavirus disease (Figure 7E), which is in line with current findings that COVID-19 infection is strongly correlated with neurodegenerative diseases [75,76].

Notably, the DEGs in excitatory neurons were involved in proteasome signaling pathway (Figure 7E). This is consistent with the observation that K48-linked polyubiquitination is significantly decreased by Mlkl KD or inhibition (Figure 4), as K48-linked polyubiquitination is the well-established signal for proteasomal degradation. This may also provide a clue for LC3-I reduction in response to Mlkl KD/KO (Figures 1 and 2), as LC3-I is degraded via the UPS [43,44]. Furthermore, transcriptomic changes in astrocytes, inhibitory neurons and oligodendrocytes were associated with ribosome signaling (Figure 7E). This is in accordance with our finding that UBA52 is identified as the MLKL physiological binding protein, as UBA52 contains a ribosomal protein fragment L40 and has been shown to be an important regulator of the ribosomal protein complex [77]. Moreover, a recent study reports that usp7 KO-disturbed pathways in HEK293A cells, including the pathways associated with ribosome, oxidative phosphorylation and multiple neurodegenerative diseases (AD included) [78], are similar to those observed in mlkl KO mice (Figure 7E). This consistency provides more evidence supporting the proposed MLKL-USP7-UBA52 signaling cascade. Taken together, the snRNA-seq results demonstrate that mlkl deletion influences multiple neurodegenerative disease-associated molecular pathways and are in accordance with our experimental findings.

Loss of Mlkl impairs cognition and aggravates AD pathology in an AD mouse model

To determine the impact of Mlkl loss on cognitive function, Y-maze test was administered to evaluate spatial working memory of 7-month-old (mature adult) and 10–14-month-old (middle-aged) WT and mlkl KO mice. Results demonstrated that middle-aged mlkl KO mice exhibited a significant decline in spontaneous alternation rate (Figure 8A), whereas no differences were observed between the WT and mlkl KO mature adult mice (Figure S5). Notably, the number of total arm entries was increased in 10–14-month-old mlkl KO mice (Figure 8B), indicating that middle-aged mice exhibit locomotor hyperactivity in the absence of Mlkl. Interestingly, this phenotype has also been observed in multiple AD mouse models harboring APP (amyloid beta precursor protein) mutants [79,80], suggesting that the impact of mlkl deletion may be prominent.

Figure 8.

Loss of Mlkl impairs cognition and aggravates AD pathology in an AD mouse model. (A) Y-maze test in middle-aged (10–14-months old) male mlkl KO and WT control mice. Quantification of spontaneous alternation rate is based on mean ± SEM and each data point represents an individual mouse. P values are calculated using two-tailed t-test. **: P < 0.01. (B) Quantification of total arm entries of Y-maze test in (A). Each data point represents an individual mouse. P values are calculated using two-tailed t-test. *: P < 0.05. (C-D) Immunoblotting of synaptic markers (DLG4 and SYP) in hippocampus of 5-month-old female mlkl KO and WT control mice. Quantification is based on mean ± SEM. Each data point represents an individual mouse. P values are calculated using two-tailed t-test. ***: P < 0.001. (E) Schematic diagram of AAV-mediated Uba52 KD in the hippocampus of WT mice. (F) Representative blots of related proteins in hippocampus injected with AAV expressing Uba52 shRNAs (shUba52) or Scr control. (G) Quantification of target proteins (mean ± SEM) in (F). Each data point represents an individual mouse. P values are calculated using two-tailed t-test. *: P < 0.05. **: P < 0.01. ***: P < 0.001. (H) Representative images of Aβ plaque burden in the brains of 10-month-old male APP/PS1;Mlkl^+/+ and APP/PS1;mlkl^-/- mice. Scale bar: 100 µm (left), 10 µm (enlarged images on the right). (I) Quantification of Aβ plaque area in the cortex and hippocampus (mean ± SEM) in (H). Each data point represents an individual mouse. P values are calculated using two-tailed t-test. **: P < 0.01. (J) Quantification of normalized Aβ40 concentration in the brains of 5-month-old male APP/PS1;Mlkl^+/+ and APP/PS1;mlkl^-/- mice (mean ± SEM). Each data point represents an individual mouse. P values are calculated using two-tailed t-test. (K) Quantification of normalized Aβ42 concentration in the brains of 5-month-old male APP/PS1;Mlkl^+/+ and APP/PS1;mlkl^-/- mice (mean ± SEM). Each data point represents an individual mouse. P values are calculated using two-tailed t-test. ***: P < 0.001. (L) Quantification of Aβ42:Aβ40 ratio in the brains of 5-month-old male APP/PS1;Mlkl^+/+ and APP/PS1;mlkl^-/- mice. Each data point represents an individual mouse. **: P < 0.01. (M) A model proposed for the function of MLKL in the brain.

To elucidate the mechanisms by which Mlkl loss induces cognitive impairment, we assessed synaptic functions via evaluating the levels of pre- and post-synaptic markers, SYP (synaptophysin) and DLG4/PSD-95 (discs large MAGUK scaffold protein 4) (Figure 8C). The immunoblotting data showed that the post-synaptic scaffold protein DLG4 was significantly reduced in the hippocampus of 5-month-old mlkl KO mice, while SYP was unchanged (Figure 8C-D). DLG4 is a key regulator of synaptic transmission and plasticity [81–85] and is significantly reduced in the brains of AD patients and animal models [86,87]. Moreover, dlg4 KO mice display defective cognitive function [88]. Collectively, these results suggest that Mlkl loss-induced cognitive impairment may be owing to synaptic dysfunction. Additionally, we knock down Uba52 in vivo using shRNAs (Figure 6E) packaged into AAV (serotype 9, AAV9) under the U6 promoter (AAV9-U6-shUba52). Purified virus was bilaterally injected into the hippocampus of 3-month-old WT mice, and hippocampal lysate was collected for immunoblotting analysis three months post-injection (Figure 8E). Consistently, DLG4 was markedly down-regulated in Uba52 KD mice (Figure 8F-G). Besides, LC3-II and BECN1 levels were also markedly diminished upon Uba52 KD (Figure 8F-G), which is in accordance with that in cell models (Figure 6E-F). These findings provide more evidence indicating that the absence of Uba52 agitates similar downstream signaling as mlkl KO.

Next, we assessed the effects of mlkl deletion on AD pathology by crossing mlkl KO mice with an AD mouse model, APPswe/PSEN1ΔE9 (APP/PS1), to generate APP/PS1;Mlkl^+/+ and APP/PS1;mlkl^-/- mice. Amyloid plaques were examined in 10-month-old mouse brains by immunostaining (Figure 8H). The immunofluorescence data showed that amyloid burden was markedly up-regulated in APP/PS1;mlkl^-/- mice compared to littermate controls (Figure 8I). We further examined microgliosis and astrogliosis by assessing AIF1/Iba1 (allograft inflammatory factor 1; marker of microglia) and GFAP (glial fibrillary acidic protein; marker of astrocyte) (Figure 8H and Figure S6A). Imaging data revealed a notable reduction in the number of AIF1-positive cells surrounding amyloid plaques in the absence of Mlkl (Figure 8H). Furthermore, neuroinflammation was evaluated in 5-month-old APP/PS1;Mlkl^+/+ and APP/PS1;mlkl^-/- mouse brains using immunoblotting (Figure S6B-C). The results indicated that AIF1 was significantly decreased after mlkl deletion, aligning with previous observations of age-related neuroinflammation in mlkl KO mice [89]. Additionally, we examined the expression of GFAP and pro-inflammatory cytokines IL6 (interleukin 6) and IL1B (interleukin 1 beta). The data demonstrated that GFAP and IL6 were unchanged in response to Mlkl deficiency (Figure S6B-C). Intriguingly, the levels of IL1B were significantly elevated in APP/PS1;mlkl^-/- mouse brains (Figure S6B-C). The elevation could be attributed to higher amyloid burden stimulating astrocytes to produce more IL1B, potentially compensating for the reduction of IL1B caused by microglial loss.

In addition to insoluble amyloid plaques, we measured the production of soluble Aβ40 and Aβ42 in 5-month-old mouse brains using enzyme-linked immunosorbent assay (ELISA). The results demonstrated that the concentration of the aggregation-prone species Aβ42 was strikingly increased in APP/PS1;mlkl^-/- mice, with Aβ40 unchanged (Figure 8J-K). Accordingly, Aβ42:Aβ40 ratio was significantly elevated in the absence of Mlkl (Figure 8L). These results indicate that loss of Mlkl exacerbates AD pathologies.

Discussion

While loss of MLKL, either at the mRNA or protein level, is an important risk factor for neurodegeneration [2,3], the underlying mechanisms remain poorly elucidated. MLKL is a well-recognized executioner in necroptosis induced by various stimuli such as TNF (tumor necrosis factor). Although some non-necroptotic functions of MLKL have been reported [50], the precise mode of action of MLKL under physiological conditions is still elusive. It is partially owing to the limited identification of MLKL binding partners [51]. RIPK3 [4], CDC37 (cell division cycle 37, HSP90 cochaperone) [90], RBM6 (RNA binding motif protein 6) [91], BECN1 [92], ESCRT (endosomal sorting complexes required for transport) proteins and the flotillins [20,21,93] have been reported to interact with MLKL in the presence of pro-necroptotic stimuli, and CAMK2/CaMKII (calcium/calmodulin dependent protein kinase II) is recently reported to associate with MLKL in response to starvation [94]. However, reports of MLKL-associated proteins under physiological conditions remain scarce.

In the current study, we identified a novel protein, UBA52, showing physiological interaction with MLKL (Figure 8M). As one of the four ubiquitin precursors, UBA52 is deubiquitinated to release ubiquitin soon after synthesis [65,95]. MLKL was indispensable for UBA52 processing, as the absence of Mlkl lead to decreased levels of free ubiquitin and increased accumulation of full-length UBA52. As MLKL itself did not cleave UBA52, USP7 was identified as the DUB directly mediating UBA52 deubiquitination via formation of a complex with MLKL and UBA52 in the brain (Figure 8M). Loss of Mlkl induced free ubiquitin shortage and subsequent deficient polyubiquitination, including K48-linked and K63-linked polyubiquitination. Deficits in K63-linked polyubiquitination suppress autophagy by destabilizing BECN1 and ULK1 [96], and lower levels of K48-linked polyubiquitination inhibit UPS-mediated protein degradation [97]. Disruption of the two main pathways responsible for protein degradation in eukaryotic cells can lead to the accumulation of toxic protein aggregates and damaged organelles [98]. Additionally, neurons are post-mitotic cells unable to dilute overloaded protein aggregates and dysfunctional organelles via cell division, so functional autophagy and UPS are more important to neurons than to other cell types [99]. Simultaneous inhibition of the two catabolic pathways promotes the development of neurodegeneration (Figure 8M).

The relationship between aberrant autophagy and neurodegenerative diseases, especially for AD, is well established [23,24,45,100]. Enhanced autophagy alleviates neurodegeneration [26,37,39,101,102], while defective autophagy facilitates neurodegenerative progression [33–35,42,100]. Our results demonstrated that deletion of Mlkl significantly inhibited autophagy in the brain (Figures 1 and 2). Consistently, middle-aged mlkl KO mice displayed short-term memory loss (Figure 8A-B), and AD pathologies were exacerbated in Mlkl-deficient AD model mice (Figure 8H-L). Neuroinflammation is a common characteristic of neurodegenerative diseases [103]. In mice with age-induced neuroinflammation, mlkl deletion or short-term necroptotic inhibition has been reported to reduce AIF1- and GFAP-positive cells, along with down-regulation of proinflammatory cytokines, IL6 and IL1B in the hippocampus [89]. In the present study, mlkl KO accelerated Aβ deposition in an AD mouse model (Figure 8), which is supposed to enhance neuroinflammation. The immunofluorescence and immunoblotting data revealed a reduction in AIF1, an elevation in IL1B, and no change in GFAP and IL6 in APP/PS1;mlkl^-/- mouse brains (Figures 8 and S6). The observed Mlkl loss-induced amyloid accumulation is presumed to be linked to compromised microglia-mediated Aβ clearance. The changes in other neuroinflammation makers (GFAP, IL6 and IL1B) are likely to be a combined result of facilitated AD progression and Mlkl deficiency.

Our study demonstrated that MLKL played a key role in maintenance of ubiquitin levels in the brain. Around 30% of free ubiquitin was reduced in the brain upon mlkl deletion (Figure 4). As free ubiquitin species in mouse brain and human frontal cortex accounts for about 60% and 80% of total ubiquitin reservoirs [68,69], the brain should have greater dependence on free ubiquitin levels. It has been observed that diminished free ubiquitin by deletion of the Ubb [104], Uchl1 (ubiquitin C-terminal hydrolase L1) [105] or Usp14 (ubiquitin specific peptidase 14) [106,107] genes induces neurodegeneration in mice.

Ubiquitin signaling is tightly involved in neurodegeneration through, for instance, regulating autophagy and UPS [108,109]. Our findings indicated that loss of Mlkl led to ubiquitin deficits, the non-degradative K63-linked polyubiquitination of BECN1 and ULK1 was thus decreased (Figure 6), and their protein stability was reduced [71], resulting in ultimate autophagic suppression. In addition, reduction of the degradative K48-linked polyubiquitination inhibits UPS-mediated protein degradation. Dysregulation of the two degradative pathways leads to disrupted protein homeostasis which is the hallmark of neurodegenerative diseases [103,110].

Notably, we observed a significant decrease in LC3-I levels following Mlkl KD/KO or Uba52 KD (Figures 1, 2 , 6 , and 8). Recent study demonstrates that, upon amino acid starvation, proteasomes are translocated from the nucleus to the cytosol, thereby stimulating protein degradation [111]. In our study, LC3-I was markedly down-regulated by EBSS-induced amino acid starvation (1 and 2 h) in N2a cells stably expressing either the Scr or shRNAs targeting Mlkl (Figure 1B-C). After 3 h of starvation, LC3-I levels in the MLKL KD group returned to comparable to Scr control levels (Figure 1B-C). Given that LC3-I is degraded by the UPS instead of lysosomes, we speculate that Mlkl KD/KO may exhibit similar effects as amino acid starvation on LC3-I levels, recruiting more proteasomes to the cytosol where LC3-I is degraded. However, the ubiquitin deficiency caused by the absence of Mlkl restricts the degradation of LC3-I facilitated by more proteosomes in the cytosol. When cells are deprived of amino acids for a longer period, the cellular system reaches equilibrium based on available ubiquitin and proteasomes. This may explain why LC3-I is down-regulated by Mlkl KD/KO under nutrient-rich conditions, but returns to similar levels as the Scr group after 3 h of amino acid starvation (Figures 1 and 2). A similar scenario may apply to Uba52 KD (Figures 6 and 8). Consistently, the snRNA-seq analysis indicated that loss of Mlkl-disturbed molecular pathways are related with proteasome (Figure 7), thereby supporting the proposed hypothesis.

Ubiquitination plays an essential role in neuronal functions, such as synapse physiology, mitochondrial metabolism, and membrane protein sorting, endosomal trafficking and exocytosis [108,109,112,113]. The levels of the postsynaptic protein DLG4 have been found to be regulated by ubiquitination [114]. The observed down-regulation of DLG4 following mlkl deletion and Uba52 KD (Figure 8) may be owing to its abnormal ubiquitination. Additionally, previous studies observe that loss of MLKL alters endosomal trafficking of EGFR (epidermal growth factor receptor) as well as generation of extracellular vesicles [20]. This may also be attributed to ubiquitin deficiency induced by MLKL deletion.

We identified USP7 as the DUB responsible for MLKL-regulated UBA52 cleavage in the brain (Figure 5). Initially, USP7 was not recognized in the screening of MLKL-interacting proteins (Figure 3B and Table S1). The omission may be attributed to the low cytosolic distribution of USP7 [115,116], while MLKL and UBA52 are primarily cytosolic proteins. So, the MLKL-USP7-UBA52 complex may exist in relatively low amounts, leading to the non-detection of signals in the proteomic screening.

Global ubiquitination and autophagic activity have been observed to decline with age [109,117,118]. Thus, MLKL, USP7 and UBA52 should be considered as potential therapeutic targets, not only for neurodegenerative diseases treatment but also for healthy aging.

Materials and methods

Mice

mlkl KO (mlkl^-/-) mouse strain was generated in C57BL/6J genetic background using CRISPR-Cas9 technology (GemPharmatech, T005339). Genotyping was performed using following PCR primers: mlkl^-/- forward primer: 5’-GTCCATAGCAGAAATAGTGGCAAC-3’; mlkl^-/- reverse primer: 5’-TATCACTGGTCTCTGGATCACAGC-3’. Mlkl^+/+ forward primer: 5’-AACACTCCACAGAAACATCAGCAG-3’; Mlkl^+/+ reverse primer: 5’-TATGTGCAAAGCCACTATCTCCTG-3’. APP/PS1 mice were purchased from Jackson Laboratory (034829-JAX) and then backcrossed to C57BL/6J for at least 10 generations. mlkl KO mice were crossed with APP/PS1 mice to generate APP/PS1;Mlkl^+/+ and APP/PS1;mlkl^-/- mice. The mice were housed in ventilated cages of a SPF facility with a 12/12 h light-dark cycle, temperature of 22 ± 2°C, and relative humidity of 50%–60%. All animal experiments were conducted in accordance with the relevant guidelines and regulations of Institutional Animal Care and Use Committee (IACUC, SIAT-IACUC-200303-NS-ZZG-A1074) at Shenzhen Institute of Advanced Technology (SIAT), Chinese Academy of Sciences (CAS).

DNA constructs

cDNAs were amplified by PCR and inserted into pAAV-CMV-MCS-WPRE adeno-associated viral vector except for the tfLC3 reporter. tfLC3 was separately cloned into the pLenti-CMV-Puro-DEST (Addgene 17,452; deposited by Eric Campeau and Paul Kaufman) lentiviral vector and the pAAV-CAG-MCS-WPRE AAV vector. cDNA mutants were generated using site-directed mutagenesis kit (Agilent Technologies 200,523). For shRNAs, the sequences of optimal 21-mers targeting specific genes were obtained from Sigma Aldrich and one scrambled 21-mer without targeting any known genes was selected as the control [119]. The annealed DNA oligos were cloned into pLKO.1-Puro lentiviral vector (Addgene, 8453; deposited by Bob Weinberg) and/or pAAV-U6-WPRE AAV vector. The information of oligonucleotides was listed in Table S2.

Cell culture, transfection, and starvation

HEK293T (ATCC, CRL-3216), HT22 (Thermo Fisher Scientific, SCC129) and N2a (ATCC, CCL-131) cells were maintained in high-glucose Dulbecco’s Modified Eagle Medium (DMEM; Cytiva, SH3002201) supplemented with 10% fetal bovine serum (FBS; Thermo Fisher Scientific 10,091,148) and 1% penicillin-streptomycin (Thermo Fisher Scientific 15,140,122). All the cells were cultured at 37°C in incubators with 5% CO₂, and 95% humidity. Cells were passaged every 2–3 days.

cDNAs and shRNAs were transfected into cells using Lipofectamine 2000 (Thermo Fisher Scientific 11,668,500) and cell lysates were obtained 24 h after transfection. Culture medium was replenished once with fresh medium the night post-transfection and replenished again 3 h before cell lysis. For starvation treatment, the time point after the 3 h incubation with the second medium replenishment was termed as nutrient rich, while cells in duplicate were starved by replacing the medium with EBSS buffer (1.8 mM CaCl₂, 5.3 mM KCl, 0.8 mM MgSO₄, 117 mM NaCl, 26 mM NaHCO₃, 1 mM NaH₂PO₄, and 5.6 mM D-Glucose, pH 7.2 ~ 7.4) for indicated times.

XTT cell viability assay

Thecell viability was examined using a commercial kit (Roche 11,465,015,001) according to the manufacturer’s instructions. Briefly, N2a cells with Mlkl KD or HEK293T cells treated with NSA (MedChemExpress, HY-100573) were seeded on 96-well plates at the cell density of 10,000 cells per well. Four replicates were used for each sample to minimize the variation. After 24-h incubation, 75 μl XTT labeling mixture was added into each well followed by another 4-h incubation. The absorbance was measured using a plate reader at the wavelength of 480 nm.

Stable cell line generation

Stable cell lines were established using lentivirus. Briefly, shRNAs or tfLC3 in lentiviral vector were co-transfected into HEK293T cells with helper plasmids psPAX2 (Addgene 12,260; deposited by Didier Trono) and pMD2.G (Addgene 12,259; deposited by Didier Trono). The culture medium containing lentiviral particles was collected at 24 and 48 h post-transfection. After centrifugation at ~ 700 × g for 5 min, collected virus was applied to infect WT N2a or HT22 cells in the presence of 8 µg/ml polybrene (MedChemExpress, HY-112735). Stable puromycin-resistant cells were selected for a week by 2 µg/ml Puromycin (Thermo Fisher Scientific, A1113803) 48 h after transduction.

Immunoblotting

Cells and mouse brain tissues were lysed with ice-cold lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% NP-40 [Aladdin Scientific, N105507]) supplemented with protease (Beyotime, P1006) and phosphatase (TransGen Biotech, DI201–02) inhibitors. After centrifugation at 15,000 × g for 15 min at 4°C, the supernatant was collected as the extracted protein samples. Protein concentration was determined by BCA protein assay kit (Thermo Fisher Scientific 23,225). Equal amounts of total protein were denatured in 1 × Laemmli buffer (62.5 mM Tris-HCl, pH 6.8, 2% SDS, 10% glycerol, 5% 2-mercaptoethanol, 0.002% bromophenol blue) by boiling for 5 min before subjected to SDS-polyacrylamide gel electrophoresis (PAGE). Total proteins were transferred onto 0.2-μm PVDF membranes (Bio-Rad 1,620,177). The membranes were then incubated with primary antibodies overnight at 4°C after blocking for 1 h in 5% bovine serum albumin (BSA; Sangon Biotech, A600332) in TBST (Sangon Biotech, C520002) at room temperature. The corresponding horseradish peroxidase (HRP)-conjugated secondary antibodies were applied and the images were taken by Gel Doc™ XR⁺ Gel Documentation System (Bio-Rad) using ECL detection reagents (Tanon, 180–5001).

The following primary antibodies were used: rabbit anti-MLKL (Sigma, SAB1302339), mouse anti-ACTB (Sigma, A5316), rabbit anti-LC3B (Novus, NB100–2220), rabbit anti-GAPDH (Cell Signaling Technology, 5174), rabbit anti-BECN1 (Proteintech 11,306–1-AP), rabbit anti-ULK1 (Proteintech 20,986–1-AP), rabbit anti-SQSTM1 (Proteintech 18,420–1-AP), rabbit anti-PIK3C3 (Proteintech 12,452–1-AP), mouse anti-MTOR (Proteintech 66,888–1-Ig), rabbit anti-ATG5 (Proteintech 10,181–2-AP), rabbit anti-ATG14 (Proteintech 19,491–1-AP), rabbit anti-VPS15 (Proteintech 17,894–1-AP), rabbit anti-phospho-MTOR (Ser2448) (Cell Signaling Technology, 2971), rabbit anti-HA tag (Proteintech 51,064–2-AP), rabbit anti-DYKDDDDK tag (for immunoprecipitation; Proteintech 20,543–1-AP), rabbit anti-FLAG (Sigma, F7425), rabbit anti-V5 (Proteintech 14,440–1-AP), mouse anti-USP7 (Proteintech 66,514–1-Ig), rabbit anti-USP9X (Proteintech 55,054–1-AP), rabbit anti-UCHL3 (Proteintech 12,384–1-AP), rabbit IgG Isotype Control (for immunoprecipitation; Thermo Fisher Scientific, 02–6102), rabbit anti-UBA52 (Abcam, ab109227), rabbit anti-ubiquitin (Abcam, ab134953), rabbit anti-ubiquitin (linkage-specific K48; Abcam, ab140601), rabbit anti-ubiquitin (linkage-specific K63; Abcam, ab179434), mouse anti-SYP (Abcam, ab8049), mouse anti-DLG4 (Thermo Fisher Scientific, MA1–045), rabbit anti-GFAP (Abcam, ab7260), rabbit anti-AIF1 (Abcam, ab178847), rabbit anti-APP (Proteintech 25,524–1-AP), rabbit anti-IL6 (Proteintech 21,865–1-AP), rabbit anti-IL1B (Cell Signaling Technology 12,426). The following secondary antibodies were used: horse anti-mouse IgG, HRP (Cell Signaling Technology, 7076), goat anti-rabbit IgG, HRP (Jackson ImmunoResearch, 111-035-003), goat anti-rabbit IgG, HRP (heavy chain specific; Abbkine, A25222), mouse anti-rabbit IgG, HRP (light chain specific; Abbkine, A25022).

QRT-PCR

Total RNA was extracted using the TransZol up plus RNA kit (TransGen Biotech, ER501–01) according to the manufacturer’s instructions. The extracted RNA was immediately transcribed to cDNA using the commercial kit (TOYOBO, FSQ-101). Real-time PCR was performed using SYBR mix (TransGen Biotech, AQ101–03) on LightCycler® 480 System (Roche). Gapdh was used as the reference transcript and the relative expression levels of target genes were calculated using 2^−ΔΔCt method. The information of oligonucleotides was listed in Table S2.

Immunofluorescence

Mice were perfused with phosphate-buffered saline (PBS; 137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, 1.8 mM KH₂PO₄, pH 7.2 ~ 7.4) and fixed with 4% paraformaldehyde (PFA). The brains were dissected and immersed in 4% PFA for 2 days at 4°C, followed by incubation in 30% sucrose (Aladdin Scientific, S112228) for another 2 days. The brains were then coronally sectioned at 40-μm thick, and at least 10 sections for each mouse were permeabilized in 0.3% Triton X-100 (Sigma, T8787) in PBS. The brain sections were blocked in 3% BSA in 0.3% Triton X-100 in PBS buffer for 1 h, then incubated overnight at 4°C with indicated primary antibodies. Corresponding secondary antibodies were applied for 1 h at room temperature. Imaging of immunofluorescence samples was performed using a Zeiss LSM880 confocal microscope, and the captured images were analyzed and processed by Zeiss ZEN software. The following primary antibodies were used: mouse anti-β-amyloid, 6E10 (BioLegend 803,004), rabbit anti-GFAP (Abcam, ab7260), rabbit anti-AIF1 (Abcam, ab178847). The following secondary antibodies were used: goat anti-mouse IgG, Alexa Fluor 594 (Thermo Fisher Scientific, A11005), goat anti-rabbit IgG, Alexa Fluor 488 (Abcam, ab150077).

For immunofluorescence on cells, HT22 cells stably expressing tfLC3 were seeded on glass-bottom 6-well-plates (NEST Biotech, 801,004), followed by transfection of shRNAs targeting specific genes or Scr control (4 μg plasmids/well). After 2 h starvation in EBSS, the cells were fixed with 4% PFA for 20 min at room temperature. Imaging was performed using FV3000 confocal laser scanning microscope (Olympus).

ELISA

Each mouse hemisphere was homogenized in 1 ml lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% NP-40) supplemented with protease and phosphatase inhibitors, and then centrifuged at 15,000 × g for 15 min at 4°C. The supernatant was saved and diluted 10 times for soluble Aβ measurement. ELISA was performed using the Aβ42 (Wako, 290–62601) and Aβ40 (Wako, 294–64701) ELISA kits according to the manufacturer’s instructions.

Immunoprecipitation and mass spectrometry analysis

HEK293T cells were transfected with plasmids containing the indicated cDNAs, and total protein samples were extracted by abovementioned lysis buffer. Immunoprecipitation was performed using Protein A/G Magnetic Beads (Thermo Fisher Scientific 88,802) according to the manufacturer’s instructions. Equal amount of total protein, antibodies or control rabbit IgG was applied, and samples were eluted in 3 × Laemmli buffer by boiling for 10 min at 100°C. Immunoblotting was then performed using indicated antibodies.

For mass spectrometry analysis, the eluted samples were separated by SDS-PAGE and visualized by Coomassie Brilliant Blue staining. LC-MS/MS analyses were performed by the Taplin Mass Spectrometry Facility (Harvard Medical School). Briefly, the protein bands were excised and digested by trypsin as described previously [120]. The tryptic peptides were reconstituted in HPLC solvent A (2.5% acetonitrile, 0.1% formic acid), loaded onto a Proxeon EASY-nLC II liquid chromatography pump (Thermo Fisher Scientific), and eluted by a gradient of increasing concentrations of solvent B (97.5% acetonitrile, 0.1% formic acid). The eluates were then subjected to electrospray ionization and analyzed by LTQ Orbitrap Velos Pro ion-trap mass spectrometer (Thermo Fisher Scientific). The sequences of eluted peptides were identified by matching with the protein or translated nucleotide databases using the SEQUEST software program (Thermo Fisher Scientific).

SnRNA-seq

Nuclei were isolated from the hippocampus of 14-month-old male mlkl KO mice and WT littermates. Briefly, tissues were homogenized by Dounce homogenizer (~25 strokes/sample) in ice-old homogenization buffer (0.25 M sucrose, 25 mM KCl, 5 mM MgCl₂, 20 mM tricine-KOH [Sigma, T0377], pH 7.8, 1 mM dithiothreitol, 0.15 mM spermine [Sigma, S2876], 0.5 mM spermidine [Sigma, S2626], protease inhibitors, 5 μg/mL actinomycin [Sigma, SBR00013], 0.32% NP-40, and 0.04% BSA). The lysate was mixed with an equal volume of OptiPrep medium [Sigma, D1556], centrifuged at 10,000 × g for 20 min at 4°C, and the nuclei were collected as pellets. The pellets were washed and resuspended in DMEM/F12 medium (Thermo Fisher Scientific 11,320,033) with 10% FBS to obtain a final concentration of 400 nuclei/μl. The purity of the single-nucleus suspensions was evaluated by flow cytometry, and high purity samples (DAPI⁺ nuclei > 95%) were used for analysis. RNAsin (Promega, N2111) was included in all buffers or solutions at 60 U/ml in nuclei extraction.

snRNA-seq libraries were generated using Chromium Single Cell 3’ Library Kit v3 (10 × Genomics 1,000,078) according to the manufacturer’s instructions. The concentration of generated libraries was measured by Qubit (Thermo Fisher Scientific), and the length of fragments was analyzed by Fragment Analyzer (Advanced Analytical Technologies). The libraries were sequenced using the NovaSeq 6000 system (Novogene).

Raw reads were mapped to the mm10 mouse reference genome using CellRanger pipeline with include-intron mode (10 × Genomics). After performing the alignment, we used SoupX (version 1.6.2) to quantify and remove ambient mRNA contamination. The corrected count matrix was then used as input for the further quality control and downstream analysis in Seurat (version 4.3). Briefly, we calculated and retained nuclei within the 2.5th percentile and 97.5th percentile of counts, features and mitochondrial genes. After filtering 21,792 nuclei were retained. Data was logNormalized and high variable features were identified for each sample using FindVariableFeatures function with the parameters selection.method = vst, nfeatures = 3000. To integrate all samples, we used merge function of Seurat (version 4.3), scaled the integrated matrix and performed linear dimensional reduction using the RunPCA function with the parameter npcs = 100. We subsequently used Harmony (version 0.1.1) to eliminate the batch effect of different samples with the parameters dims.use = NULL, group.by.vars = orig.ident. We visualized the P value distribution of each principal component using JackStraPlot function and chose the first 20 principal components for graph-based clustering.

We performed K-nearest neighbor clustering using the FindClusters function and set the parameter resolution from 0.1–0.6 and selected 0.2 for downstream clustering. UMAP and t-SNE clustering were performed using the RunUMAP and RunTSNE function with the parameter reduction = harmony, dims = 1:20. Putative doublets were removed using DoubletFinder (version 2.0.3) and the multiplet rate was selected according to the manufacturer’s instructions. Next, differential expression of genes were generated using the FindMarkers function of Seurat (version 4.3) with the parameters logfc.threshold = 0 and test.use = MAST. The bulk fold changes were calculated as described previously [121]. The level of statistical significance for cell type-specific transcriptomic changes was set at an adjusted P < 0.05 and bulk fold change > 1.5 or < −1.5. DEGs were functionally annotated by clusterProfiler (version 4.4.4).

Protein synthesis in vitro and co-IP

The C-terminal tagged MLKL-3×FLAG, MLKL [1-180]-3×FLAG, MLKL[181-471]-3×FLAG, UBA52-3×HA, HA-UBA52-FLAG and USP7-V5 proteins were synthesized using the PURExpress In vitro Protein Synthesis kit (New England BioLabs, E6800S) according to the manufacturer’s instructions. Briefly, the indicated cDNAs were amplified by PCR, and the concentration of obtained PCR product was adjusted to 250 ng/μl. 1 μl of PCR template was added for each 25 μl reaction. The proteins were synthesized after 4-h incubation at 37°C and validated by immunoblotting using anti-FLAG, anti-HA or anti-V5 antibodies.

In co-IP analysis, for synthesized proteins, two relevant proteins were incubated at a volume ratio of 1:1 in abovementioned lysis buffer. For DUBs screening in mouse brain, relevant synthesized protein(s) was incubated with mouse brain lysate. Immunoprecipitation was performed using Protein A/G Magnetic Beads (Thermo Fisher Scientific 88,802) according to the manufacturer’s instructions. Anti-FLAG, -HA, -V5, -USP7, USP9X antibodies and control rabbit IgG were employed. Samples were eluted in 3 × Laemmli buffer by boiling for 10 min at 100°C, and then examined by immunoblotting. To prevent interference from light or heavy chain, heavy chain (Abbkine, A25222) or light chain specific HRP-conjugated secondary antibody (Abbkine, A25022) was used.

AAV generation and injection

AAV9 serotype virus containing mouse Uba52 sh1, AAV.CAP-B10 serotype virus expressing tfLC3 and mouse Mlkl shRNAs (sh1 and sh2), and virus of Scr control shRNA were separately generated, following a previously described protocol [122]. Purified AAV9, titers of 1 × 10¹⁴ viral genomes per ml (vg/ml), were stereotactically injected into the hippocampus of 3-month-old male WT C57BL/6J mice at the coordinates A/P: −2 mm, M/L: −1.5 mm, D/V: −1.5 mm, with a rate of 400 nl/min and a total volume of 1.5 μl unilaterally. After 3 months, the mice were sacrificed, and hippocampus were collected for immunoblotting analysis. For the AAV.CAP-B10 virus, tfLC3 (3.5 × 10¹¹ vg/mouse) and Mlkl shRNAs (sh1 and sh2, 5 × 10¹¹ vg/mouse) or Scr control (5 × 10¹¹ vg/mouse) were intravenously injected into 3-month-old male WT C57BL/6J mice via the tail veins. After 1 month, the mouse brains were harvested for immunostaining, with the mice starved for 24 h before sacrifice.

Y-maze test

Mice were acclimated to the testing room for 1 h before conducting the Y-maze test. Each mouse was placed at the center of Y-maze and allowed to explore for 10 min while being recorded by a video camera. The arm entries were traced and spontaneous alternation rate was calculated as previously described [80].

Quantification and statistical analysis

Immunofluorescence images and immunoblots were analyzed using ImageJ (https://imagej.nih.gov/ij/index.html). Statistical analysis was performed with the Prism 9.0 software (GraphPad) and Microsoft Excel using unpaired two-tailed t-tests. p values < 0.05 were considered significant. All quantitative graphs were presented as mean ± SEM (standard error of the mean).

Funding Statement

This work was funded by the Key-Area Research and Development Program of Guangdong Province [2023B0303040004 to J.T.], the National Natural Science Foundation of China [NSFC, 32371070 to J.T.], the NSFC-RGC Joint Research Scheme [32061160472 to Y.C.], the Guangdong Provincial Fund for Basic and Applied Basic Research [2019B1515130004 to Y.C.], the Hong Kong Innovation and Technology Fund [MRP/056/21 to Y.S.], the Science and Technology Program of Guangzhou [202007030001 to J.T.], the Shenzhen Key Laboratory of Neuroimmunomodulation for Neurological Diseases [ZDSYS20220304163558001 to J.T.], the Key Basic Research Program of Shenzhen Science and Technology Innovation Commission [JCYJ20210324115811031 to J.T.; JCYJ20220818101618040 to Z.G.Z.; JCYJ20220818101615033 to S.R.J.; JCYJ20210324101813035 to Z.G.Z.; JCYJ20200109150717745 to X.F.Y.; JCYJ20200109144418639 to L.Y. Z].

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data supporting the findings of this study are available within the paper. All snRNA-seq data described in the paper have been deposited in the NCBI Gene Expression Omnibus (GEO) database and are accessible through the GEO accession number GSE270139 (private until publish). All other data or codes supporting the findings of this study are available from the corresponding author upon reasonable request.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15548627.2024.2395727