Cytosolic acetyl-coenzyme A is a signalling metabolite to control mitophagy

Abstract

Acetyl-coenzyme A (AcCoA) sits at the nexus of nutrient metabolism and shuttles between the canonical and non-canonical tricarboxylic acid cycle^1,2, which is dynamically regulated by nutritional status, such as fasting³. Here we find that mitophagy is triggered after a reduction in cytosolic AcCoA levels through short-term fasting and through inhibition of ATP-citrate lyase (encoded by ACLY), mitochondrial citrate/malate antiporter (encoded by SLC25A1) or acyl-CoA synthetase short chain family member 2 (encoded by ACSS2), and the mitophagy can be counteracted by acetate supplementation. Notably, NOD-like receptor (NLR) family member X1 (NLRX1) mediates this effect. Disrupting NLRX1 abolishes cytosolic AcCoA reduction-induced mitophagy both in vitro and in vivo. Mechanically, the mitochondria outer-membrane-localized NLRX1 directly binds to cytosolic AcCoA within a conserved pocket on its leucine-rich repeat (LRR) domain. Moreover, AcCoA binds to the LRR domain and enhances its interaction with the nucleotide-binding and oligomerization (NACHT) domain, which helps to maintain NLRX1 in an autoinhibited state and prevents the association between NLRX1 and light chain 3 (LC3). Furthermore, we find that the AcCoA–NLRX1 axis underlies the KRAS-inhibitor-induced mitophagy response and promotes drug resistance, providing a metabolic mechanism of KRAS inhibitor resistance. Thus, cytosolic AcCoA is a signalling metabolite that connects metabolism to mitophagy through its receptor NLRX1.

Acetyl-coenzyme A functions as a non-canonical signal to trigger mitophagy, and the acetyl-coenzyme A–NLRX1 axis underlies the KRAS-inhibitor-induced mitophagy response and promotes drug resistance, providing a metabolic mechanism of KRAS inhibitor resistance.

Main

Cells selectively degrade mitochondria through two pathways: the PINK1–Parkin pathway^4,5, and the mitophagy receptor-mediated pathway^4,5. NLRX1 is the only mitochondria-localized NLR and functions as a mitophagy receptor that directly binds to LC3 to mediate mitophagy in response to mitochondrial damage^6,7. Fasting offers benefits to human health through improving mitochondrial quality⁸. The mitophagic structure has been observed in muscles after 24 h of fasting⁹ and remains intact after Pink1 deficiency in mice¹⁰. Although several mitophagy receptors could be transcriptionally upregulated by fasting¹¹, it remains unclear whether a specific mitophagy receptor mediates fasting-induced mitophagy in a selective manner. Other open questions include the identification of novel mechanisms of mitophagy receptor-mediated selective mitophagy and delineation of mitophagy-regulated tumour cell death in cancer biology.

In the non-canonical tricarboxylic acid (TCA) cycle, mitochondrial citrate is exported to the cytosol via SLC25A1 and is broken down into oxaloacetate and AcCoA via ACLY to produce malate and fatty acids, respectively². AcCoA-reduction-induced autophagy depends mainly on its canonical function as the substrate of protein acetylation^3,12,13. Whether AcCoA mediates nutritional signals of mitophagy in a non-canonical manner, such as through acetylation and AMPK independently, is of great interest.

Cytosolic AcCoA links to mitophagy

To investigate fasting-induced mitophagy response, we first analysed changes of metabolites in mice serum after overnight starvation (the fasting effect was reflected by weight loss shown in Extended Data Fig. 1a). Among them, the glucose concentration dropped significantly from about 11–13 mM to 5–7 mM (Extended Data Fig. 1b). The glutamine concentration also decreased by about 30% (Extended Data Fig. 1c). As glucose and glutamine are the primary carbon sources for cell growth, we tailored a mild starvation medium (SM): 5 mM glucose and 2 mM glutamine. Mitophagy response was induced in various cells cultured with SM as indicated by decreased levels of mitochondrial proteins TIM23, MT-CO2 and HSP60, and increased LC3 recruitment on mitochondria (Extended Data Fig. 1d), which was blocked by bafilomycin A1 (BafA1), suggesting that this reduction in mitochondrial mass was due to autophagy (Extended Data Fig. 1d). mt-Keima is a highly sensitive reporter for mitophagy measurement¹⁴. SM-induced mitophagy was determined by an increased acidic mitophagy reporter mt-Keima signal and a reduction in the mitochondrial DNA (mtDNA)/nuclear DNA (nDNA) ratio (Extended Data Fig. 1e,f), which could be reversed by resupplementation with normal medium (NM) (Extended Data Fig. 1g–i). To determine whether this mitochondrial reduction was due to decreased mitochondrial biogenesis, we assessed the expression levels of both genes encoding regulators of mitochondrial biogenesis (PPARGC1A (encoding PGC1α), NRF1 and NRF2) and nuclear-encoded mitochondrial housekeeping genes (TFAM, TIM23 and HSP60). Although there was a slight decrease in the NRF1 expression level in A549 cells after SM culture, possibly due to an altered metabolic state (Extended Data Fig. 1j), other genes remained unaffected after SM culture or resupplemention with NM (Extended Data Fig. 1j,k). Notably, U-2 OS cells did not respond to SM-induced mitophagy compared with other cells (Extended Data Fig. 1d,e).

Extended Data Fig. 1. Mild starvation induces mitophagy in an AMPK- and mTOR-independent manner.

a-c, The characterization of fasting mice. 6–8-week-old male mice were ad libitum-fed or fasted overnight for 16 h with free access to water. The mice body weight (a), n = 6 mice per group; the glucose level (b), n = 5 mice per group; the serum level of amino acids (c), the indicated amino acids measured by mass-spectrometry while the cysteine level was too low to be detected (valine, leucine and isoleucine were analysed by GC-MS while other amino acids analysed by LC-MS). n = 4 mice per group. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test. d-f. SM induces mitophagy in a cell-line-dependent manner. Total cell or Mito lysates as indicated for immunoblotting (d), cells expressing mt-Keima reporter for flow cytometry analysis (e) or qPCR analysis of mtDNA/nDNA (f). n = 3 biological replicates. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test for all groups except U-2 OS group in (e) (Mann-Whitney U-test) and MCF7 mtATP6 group in (f) (Welch’s t-test). g-i, Re-feed with NM could rescue SM-induced mitophagy. Total cell or Mito lysates from the indicated cells treated with SM for 16 h, followed by the NM for another 16 h for immunoblotting (g), cells expressing mt-Keima reporter for flow cytometry analysis (h) or qPCR analysis of mtDNA/nDNA (i). n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (h,i). j,k, RT-qPCR analysis of Mito biogenesis-related gene and nuclear-encoded Mito gene expression levels. Cells treated as indicated. n = 3 biological replicates. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test (j) and Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (k) except MCF7 HSP60 in (k) (Kruskal-Wallis test and Dunn’s multiple comparisons test). l,m, SM does not affect AMPK or mTOR pathway. The indicated cells were treated with SM, Torin 1 (100 nM) for 16 h or oligomycin (Oligo, 5 μM) for 5 min for immunoblotting. n = 3 biological replicates. n, Heat map of key intracellular metabolites level. Upper panel: the flow chart that the indicated cells were treated with complete medium (Mock) or SM for 16 h for Mass Spectrometry. Bottom panel: data calculated as relative metabolite ratios of SM/Mock and shown as mean values from 4 biological replicates for all metabolite except AcCoA (3 biological replicates). Data normalized to Actin. o, Metabolic enzyme protein expression level in different cell lines by immunoblotting. n = 3 biological replicates. p, SM reduces Cyto AcCoA level in a cell-line-dependent manner. Cells cultured with SM for 16 h for the relative abundance of Mito or Cyto AcCoA level by LC-MS. Data normalized to Tubulin (Cyto) or HSP60 (Mito) levels. n = 3 biological replicates. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test.

Source Data

AMPK is a master regulator of autophagy in response to nutrient deficiency¹⁵. Compared with the markedly increased level of phosphorylated AMPK (p-AMPK) and p-ULK1-S555 triggered by AMPK activator oligomycin, SM culture did not induce apparent AMPK activation (Extended Data Fig. 1l), consistent with the previous report that low-glucose (5 mM) medium did not induce AMPK activation¹⁶. Autophagy is also inhibited by mTORC1 activation¹⁷. Compared with the mTORC1 inhibitor Torin-1, SM culture did not affect mTORC1 activation status (Extended Data Fig. 1m). Thus, SM-induced mitophagy occurs in an AMPK- and mTOR-independent manner.

We next hypothesized that the various mitophagic responses after SM culture might be due to different intrinsically metabolic traits of cells. Mass spectrometry (MS) analysis showed that the levels of AcCoA, malate and certain fatty acids were decreased in HeLa, A549 and MCF7 cells, but not in U-2 OS cells, after SM culture (Extended Data Fig. 1n). The levels of several amino acids were elevated after SM treatment of HeLa cells compared with other mitophagy-responsive cells, such as A549 and MCF7 cells. It might be a cell-type-specific response to increased nutrient uptake or autophagic turnover of intracellular proteins, rather than a general response correlating with SM-induced mitophagy in our tested cell lines. Notably, these metabolites are linked to a non-canonical TCA cycle. Moreover, compared with U-2 OS cells, SM-responsive cells had significantly high protein levels of ACLY, fatty acid synthase (FASN), isocitrate dehydrogenase 1 (IDH1), SLC25A1 and AcCoA carboxylase 1 (ACC1), all linked to cytosolic AcCoA metabolism (Extended Data Fig. 1o). We further confirmed that cytosolic but not mitochondrial AcCoA levels decreased after SM culture of responsive cells (Extended Data Fig. 1p).

We therefore investigated whether cytosolic AcCoA was implicated in SM-induced mitophagy. Inhibiting ACLY or SLC25A1 via their inhibitors—potassium hydroxycitrate tribasic monohydrate (HC) and SB204990 (SB) (ACLY inhibitors) and 1,2,3-benzenetricarboxylic acid hydrate (BTC) (SLC25A1 inhibitor) (Fig. 1a)—reduced the cytosolic AcCoA levels (Extended Data Fig. 2a) and enhanced mitophagy, as shown by elevated acidic mt-Keima signals and decreased mitochondrial mass (Fig. 1b,c and Extended Data Fig. 2b–d). Congruously, knockdown of ACLY or SLC25A1 induced mitophagy in tested cells cultured with NM (Extended Data Fig. 2e–g). Acetate contributes to cytosolic AcCoA pools¹⁸. Accordingly, ACSS2 knockdown also induced mitophagy in NM (Extended Data Fig. 2g–i). By contrast, mitochondrial content remained intact after knocking down ACC1 or FASN (Extended Data Fig. 2j,k). Notably, adding acetate abolished SM-induced or ACLY-knockdown-induced mitophagy (Extended Data Fig. 2l–q). We found that ACLY or SLC25A1 inhibition mildly upregulated PPARGC1A and NRF2 expression levels in HeLa and A549 cells (Extended Data Fig. 3a), which may contribute to mitophagy-stimulated mitochondrial turnover. However, the expression levels of nuclear-encoded mitochondrial genes, including TIM23 and HSP60, showed no decrease after a reduction in cytosolic AcCoA (Extended Data Fig. 3a–d), suggesting that cytosolic AcCoA reduction does not substantially alter mitochondrial biogenesis. Moreover, knockdown of the autophagy upstream gene FIP200 or the core machinery gene ATG7 (ref. ¹⁹) blocked SM-induced or SLC25A1- or ACLY-inhibition-induced mitophagy (Extended Data Fig. 3e,f). Collectively, these results show that a reduction in cytosolic AcCoA induces mitophagy in vitro.

Fig. 1. NLRX1 is required for cytosolic AcCoA-reduction-induced mitophagy.

a, Schematic of AcCoA metabolism in the cytosol. b, Flow cytometry analysis of HeLa-tet-on-mt-Keima cells treated with HC (20 mM), SB (100 μM) or BTC (5 mM) for 16 h. n = 3 biological replicates. Data are mean ± s.e.m. Statistical analysis was performed using ordinary one-way analysis of variance (ANOVA) and Dunnett’s multiple-comparison test. c, Immunoblotting images of HeLa cells treated as in Fig. 1b with or without BafA1 (200 nM). n = 3 biological replicates. d, Schematic of genome-wide CRISPR screening of HC-induced mitophagy. e, Mitochondrial (mito)-related genes depleted in mitophagic cells treated with HC. Previously reported mitophagy receptors (MRs) are highlighted in red. NLRX1 was the top-ranked mitophagy receptor. P values were calculated using MAGeCK software. f, Flow cytometry quantification of control (sgNC) or NLRX1-knockout (sgNLRX1) HeLa-tet-on-mt-Keima cells with the indicated treatment for 16 h. n = 3 biological replicates. Data are mean ± s.e.m. Statistical analysis was performed using two-way ANOVA with Bonferroni’s multiple-comparison test. g, Immunoblot images of HeLa cells treated as indicated. n = 3 biological replicates. h, Immunoblot images of liver tissues of WT or Nlrx1^−/− mice given intraperitoneal injection of PBS or HC (100 mg per kg) for 4 h. n = 3 mice per group. i,j, Confocal microscopy analysis of mt-Keima signals in liver tissues of WT or Nlrx1^−/− mice after HC treatment. i, Representative images of liver tissue sections. Scale bar, 5 μm. j, Quantification of the mt-Keima signal in Fig. 1i. Values are normalized to the red/green signal in the WT + PBS group. n = 24 images from 4 mice per group. Data are mean ± s.e.m. Statistical analysis was performed using two-way ANOVA with Bonferroni’s multiple-comparison test.

Source Data

Extended Data Fig. 2. Cytosolic AcCoA level controls mitophagy in vitro.

a-d, SLC25A1 or ACLY inhibition decreases Cyto AcCoA level (a) and induces mitophagy (b-d). Indicated cells treated with HC (20 mM), SB (100 μM), BTC (5 mM) for 16 h for Mito or Cyto AcCoA analysis by LC-MS (a), immunoblotting (b), cells expressing mt-Keima reporter for flow cytometry analysis (c) or qPCR analysis of mtDNA/nDNA (d). n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (a,c,d). e-i, SLC25A1, ACLY or ACSS2 knockdown induces mitophagy. The indicated cells transfected with indicated siRNAs were cultured with or without Baf A1 (200 nM). Total cell or Mito lysates for immunoblotting (e,h), cells expressing mt-Keima reporter for flow cytometry analysis (f,i) or qPCR analysis of mtDNA/nDNA (g). n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (f,g) or unpaired two-tailed Student’s t-test (i). j,k, ACC1 or FASN knockdown does not affect mitophagy. The indicated cells were transfected with indicated siRNAs. Total cell or Mito lysates for immunoblotting (j), cells expressing mt-Keima reporter for flow cytometry analysis (k). n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (k). l-n, Ace blocks SM-induced mitophagy. The indicated cells cultured with SM for 16 h with or without 10 mM Ace. Total cell or Mito lysates for immunoblotting (l), cells expressing mt-Keima reporter for flow cytometry analysis (m) or qPCR analysis of mtDNA/nDNA (n). n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (m,n). o-q, Ace blocks ACLY knockdown-induced mitophagy. The indicated cells transfected with indicated siRNAs cultured with or without 10 mM Ace. Total cell or Mito lysates for immunoblotting (o), cells expressing mt-Keima reporter for flow cytometry analysis (p) or qPCR analysis of mtDNA/nDNA (q). n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (p,q).

Source Data

Extended Data Fig. 3. Mitochondria reduction induced by decreased cytosolic AcCoA is dependent on autophagy.

a-d, RT-qPCR analysis of Mito biogenesis-related gene and nuclear-encoded Mito gene expression levels in indicated cells. Cells treated as indicated for 16 h. n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test for all groups except HeLa HSP60 in (b), A549 PPARGC1A and all HSP60 in (c), A549 (NRF1, HSP60) and MCF7 HSP60 in (d) (Kruskal-Wallis test and Dunn’s multiple comparisons test). e,f, ATG7 or FIP200 knockdown blocks SM- and SLC25A1 or ACLY inhibition-induced mitophagy. HeLa cells were transfected with siRNA targeting control (siNC), ATG7 (siATG7), or FIP200 (siFIP200), respectively, followed by indicated treatments for 16 h. Mitophagy was determined by indicated mitochondria markers (e) or qPCR analysis of mtDNA/nDNA (f). n = 3 biological replicates in (e and bottom panel of (f)) and n = 6 biological replicates in (upper panel of (f)). Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test (f).

Source Data

Fasting led to a significant reduction in cytosolic AcCoA levels in the brain and gastrocnemius but not in liver tissue²⁰ (Extended Data Fig. 4a). Meanwhile, mitochondrial AcCoA levels in the brain and gastrocnemius tissues were not affected (Extended Data Fig. 4a). Accordingly, the mitochondrial mass decreased in starved gastrocnemius and brain but not in liver tissues, as determined by the levels of mitochondrial proteins TIM23 and CYTB, mitochondrial LC3 recruitment and the mtDNA/nDNA ratio (Extended Data Fig. 4b,c), corroborating the linkage between mitophagy response and cytosolic AcCoA levels in vivo. Consistent with that, fasting-induced mitophagy was demonstrated by elevation of acidic mt-Keima signals in gastrocnemius and brain but not liver tissues (Extended Data Fig. 4d), while fasting-upregulated Nrf2 levels in gastrocnemius tissue probably contributed to mitochondrial turnover (Extended Data Fig. 4e). Refeeding restored levels of cytosolic AcCoA, mitophagy (Extended Data Fig. 4f–j) and Nrf2 expression without affecting other mitochondrial-biogenesis-related gene expression (Extended Data Fig. 4k). Furthermore, intraperitoneal administration of sodium acetate increased cytosolic AcCoA levels and blocked fasting-induced mitophagy in gastrocnemius tissue without affecting mitochondrial biogenesis (Extended Data Fig. 4l–p). Intraperitoneal injection with HC³ to inhibit ACLY in vivo decreased the cytosolic but not mitochondrial AcCoA levels (Extended Data Fig. 4q) and induced mitophagy without affecting mitochondrial biogenesis (Extended Data Fig. 4r–u). Thus, mitophagy could be regulated by cytosolic AcCoA levels both in vitro and in vivo.

Extended Data Fig. 4. Cytosolic AcCoA controls mitophagy in vivo.

a-d, 24 h-fasting reduces Cyto AcCoA level (a) and induces mitophagy (b-d) in gastrocnemius (Gastro) and brain but not liver tissues. C57BL/6 male mice fasted for 24 h, and then Cyto and Mito fractions from Gastro, brain, and liver tissues or total lysates were harvested for LC-MS (a) or immunoblotting analyses (b). n = 6 mice for all groups except for the Brain samples (n = 5 mice) in (a) and n = 3 mice per group in (b). Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test for all groups except Cyto of Brain (Mann-Whitney U-test) and Cyto of Gastro (Welch’s t-test). c, qPCR analysis of mtDNA/nDNA (n = 4 mice per group). Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test except Gastro mtAtp6 and Brain mtCytb (Mann-Whitney U-test). d, Relative mt-Keima signal in each group. 10-week-old male mice expressing AAV-delivered mt-Keima reporter treated as indicated. Values normalized to ad libitum level of mitophagy. n = 42 (Gastro, ad libitum, from 5 mice) and n = 58 (Gastro, fasted, from 6 mice), n = 40 (Brain, from 4 mice) and n = 49 (Liver, from 4 mice) images. Data shown as the mean ± s.e.m. Mann-Whitney U-test. e, Mito biogenesis-related gene and nuclear-encoded Mito gene expression levels (n = 6 mice per group). Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test except liver Ppargc1a and Nrf1 (Welch’s t-test). f-k, Refeeding could rescue fasting-induced reduction of Cyto AcCoA level (f) and mitophagy (g-k). 6-week-old male mice fasted for 24 h and ad libitum-fed for another 24 h. f, Cyto Gastro tissues harvested for LC-MS. n = 5 mice per group. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test. g, Relative mt-Keima signal in each group. 10-week-old male mice expressing AAV-delivered mt-Keima reporter were treated as indicated, and Gastro tissues were harvested for confocal microscopy. Values normalized to the ad libitum level of mitophagy. n = 24 images from 4 mice per group. Data shown as the mean ± s.e.m. Welch one-way ANOVA and Dunnett’s T3 multiple comparisons. h,i, Total or Mito Gastro lysates for immunoblotting. n = 3 mice for all groups except for the fasted groups (n = 4 mice) in (h) and n = 3 mice for all groups in (i). j, qPCR analysis of mtDNA/nDNA (n = 4 mice per group). k, Mito biogenesis-related gene and nuclear-encoded Mito gene expression levels (n = 6 mice per group). Data in (j, k) shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test for all groups except Nrf2 in (k) (Welch one-way ANOVA and Dunnett’s T3 multiple comparisons). l-p, Ace blocks fasting-induced decreased Cyto AcCoA level (l) and mitophagy (m-p). l,m, C57BL/6 male mice fasted for 24 h, and PBS or Ace (1 g/kg) was intraperitoneally injected 10 h and 1 h before sacrifice. Then Cyto and Mito fractions from Gastro tissues or total lysates were harvested for LC-MS (l) or immunoblotting (m). n = 4 mice per group in (l) and n = 3 mice per group in (m). Data shown as the mean ± s.e.m. Welch one-way ANOVA and Dunnett’s T3 multiple comparisons. n, Relative mt-Keima signal in each group. 10-week-old male mice expressing AAV-delivered mt-Keima reporter were treated as indicated. Values normalized to the ad libitum level of mitophagy. n = 24 images from 4 mice per group. Data shown as the mean ± s.e.m. Welch one-way ANOVA and Dunnett’s T3 multiple comparisons. o, qPCR analysis of mtDNA/nDNA (n = 4 mice per group). p, Mito biogenesis-related gene and nuclear-encoded Mito gene expression level (n = 6 mice per group). Data in (o, p) shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test except mtCo1 in (o) (Welch one-way ANOVA and Dunnett’s T3 multiple comparisons) and Nrf1 in (p) (Kruskal-Wallis test and Dunn’s multiple comparisons test). q-u, Acly inhibition decreases Cyto AcCoA level (q) and induces mitophagy (r-u) by intraperitoneally HC (100 mg per kg) injection for 4 h. q,r, Cyto and Mito fractions from liver tissues or total lysates harvested for LC-MS (q) or immunoblotting (r). n = 6 mice per group in (q) and n = 3 mice per group in (r). Data shown as the mean ± s.e.m. Welch’s t-test. s, 10-week-old male mice expressing AAV-delivered mt-Keima reporter treated as indicated. Left panel: representative images of liver tissue sections shown. Scale bar, 10 μm. Right panel: relative mt-Keima signal values in each group normalized to PBS level of mitophagy. n = 54 (PBS, from 4 mice) and n = 93 (HC, from 5 mice) images. Data shown as the mean ± s.e.m. Mann-Whitney U-test. t, qPCR analysis of mtDNA/nDNA (n = 4 mice per group). u, Mito biogenesis-related gene and nuclear-encoded Mito gene expression level (n = 6 mice per group). Data in (t,u) shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test.

Source Data

NLRX1 mediates cytosolic AcCoA mitophagy

We next investigated how cytosolic AcCoA reduction triggers mitophagy. SM culture or inhibition of SLC25A1 or ACLY had little effect on mitochondrial functions, including mitochondrial membrane potential (Extended Data Fig. 5a), mitochondrial reactive oxygen species (ROS) (Extended Data Fig. 5b), ATP production (Extended Data Fig. 5c), mitochondrial protein import (Extended Data Fig. 5d), permeability transition (Extended Data Fig. 5e) and oxidative phosphorylation rate (Extended Data Fig. 5f). Cells remained intact in different experimental conditions (Extended Data Fig. 5g). Moreover, neither PINK1 stabilization (Extended Data Fig. 5h) nor the mitochondrial recruitment or the E3 activity of Parkin was observed after SM culture or inhibition of SLC25A1 or ACLY compared with activating effects of carbonyl cyanide m-chlorophenyl hydrazone (CCCP)¹⁹ (Extended Data Fig. 5i,j). Alternatively, we hypothesized about whether the mitophagy receptor is involved in this mitophagy response. To this end, a genome-wide CRISPR screening (19,050 genes, 6 single guide RNAs (sgRNAs) per gene) assay was performed to systemically and unbiasedly investigate mitophagy receptors involved in HC-induced mitophagy response in HeLa-tet-on-mt-Keima reporter cells (Fig. 1d). ULK1, ATG4A, WIPI2 and LAMP1 were identified among the top candidates lost in the top mitophagic cells after HC treatment (Extended Data Fig. 5k), demonstrating the robustness of the screening assay. The complete list of the screening was shown in Supplementary Table 1. Although transient knockdown of ACLY, SLC25A1 or ACSS2 induced mitophagy in multiple cell lines (Extended Data Fig. 2e–i), these genes were not identified as significantly enriched genes during the genome-wide CRISPR screening assay. There might be a compensatory effect during generating the knockout cells²¹. To explore which mitophagy receptor mediates this pathway, we listed all mitochondrial candidates using the MitoCarta 3.0 database²². Gene Ontology (GO) pathway enrichment analysis of the top 100 genes revealed the carboxylic acid metabolic process as the top candidate pathway (Extended Data Fig. 5l), highlighting the role of metabolism in mitophagy. NLRX1 ranked first among the screen-hit mitophagy receptors^23,24 and was ranked tenth of all mitochondrial proteins (ranked by negative log-transformed fold change and P < 0.05) (Fig. 1e and Supplementary Table 2). Although the function of NLRX1 in mitophagy has been reported^6,7, the upstream signal triggering NLRX1 activation remains unclear. Indeed, NLRX1 deficiency significantly impaired SM-induced, SLC25A1- or ACLY-inhibition-induced, and ACLY-, SLC25A1- or ACSS2-knockdown-induced mitophagy response, on the basis of the analysis results of mitochondrial proteins, LC3 recruitment on mitochondria, mtDNA/nDNA levels and acidic mt-Keima signal levels (Fig. 1f,g and Extended Data Fig. 6a–o).

Extended Data Fig. 5. Effects of Cytosolic AcCoA reduction on mitochondrial functions and PINK1-Parkin pathway.

HeLa cells were cultured with SM, HC (20 mM), SB (100 μM), BTC (5 mM), or CCCP (10 μM) for 16 h in (a-h), CCCP (10 μM) for 1 h in (i, j), MitoBloCK-6 (MB-6) (100 μM) for 6 h in (d), or actinomycin D (10 μM) for 16 h in (e). a, SM and SLC25A1 or ACLY inhibition do not decrease Mito membrane potential according to flow cytometry analysis of TMRM staining. Left panel: histogram plot of the TMRM fluorescence. Right panel: relative quantitative analysis of the fluorescence level of TMRM. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test. n = 3 biological replicates. b, Effects of SM and SLC25A1 or ACLY inhibition on Mito ROS production according to flow cytometry analysis of Mito ROS production by MitoSOX staining. Left panel: histogram plot of the MitoSOX fluorescence. Right panel: relative quantitative analysis of the fluorescence level of MitoSOX. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test. n = 3 biological replicates. c, Effects of SM and SLC25A1 or ACLY inhibition on intracellular ATP level according to luminescence determination. Data shown as the mean ± s.e.m. Kruskal-Wallis test and Dunn’s multiple comparisons test. n = 4 biological replicates. d, Effects of SM and SLC25A1 or ACLY inhibition on Mito protein import. HeLa-rtTA cells transfected with the vector expressing doxycycline (dox)-inducible MTS-EGFP and treated with dox (0.25 μg per ml) for 6 h before stimulation. Total Mito were stained using Mitotracker (purple). Left panel: representative cell images of MTS-EGFP (green) localization compared with mitochondria (purple) shown. Scale bar, 10 μm. Right panel: quantitative analysis of the proportion of cells with improperly localized MTS-EGFP (localized in the cytosol or nucleus). n = 26 (Mock), n = 32 (SM), n = 27 (HC), n = 28 (SB), n = 27 (BTC) and n = 29 (MB-6) images. Data shown as the mean ± s.e.m. Kruskal-Wallis test and Dunn’s multiple comparisons test. e, SM and SLC25A1 or ACLY inhibition do not induce mitochondria permeability transition determined by Cytochrome c release. Cyto fractions were isolated after the indicated treatment, and Cytochrome c level was determined by immunoblotting. n = 3 biological replicates. f,g, Effects of SM and SLC25A1 or ACLY inhibition on oxygen consumption rate (OCR) at routine rates (f), cell death according to analysis of LDH release (g). n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test. h-j, SM and SLC25A1 or ACLY inhibition do not induce PINK1 protein stabilization (h), Parkin’s E3 activity (i) and localization on mitochondria (j). n = 3 biological replicates (h). i,j, HeLa cells stably expressing GFP-Parkin were generated, and after treatments, immunoblotting analyses of ubiquitinated GFP-Parkin were determined (i) and Mito marker TOM20 was stained (j). n = 3 biological replicates (i). j, Left panel: representative cell images of colocalization of GFP-Parkin with TOM20 are shown Scale bar, 10 μm. Right panel: the statistical analysis of Parkin colocalization with mitochondria. n = 10 images for all groups except for the Mock and CCCP groups (n = 11 images). Data shown as the mean ± s.e.m. Kruskal-Wallis test and Dunn’s multiple comparisons test. k, Volcano plot of the whole genome depleted genes in mitophagic cells treated with HC. Autophagy-related genes were highlighted in red. l, Gene ontology analysis of Fig. 1e. The P values were calculated using MAGeCK software (k) or Metascape database (l).

Source Data

Extended Data Fig. 6. Cytosolic AcCoA reduction induces NLRX1-dependent mitophagic response.

a-c, NLRX1 is required for SM-induced mitophagic response. The indicated control (sgNC) or NLRX1 knockout (sgNLRX1) cells cultured with SM for 16 h. d-i, NLRX1 is required for SLC25A1 or ACLY inhibition-induced mitophagic response. d-f, The indicated control (sgNC) or NLRX1 knockout (sgNLRX1) cells were treated with HC (20 mM) and BTC (5 mM) for 16 h. g-i, The indicated control (sgNC) or NLRX1 knockout (sgNLRX1) cells were treated with SB (100 μM) for 16 h. j-o, NLRX1 is required for ACLY/SLC25A1/ACSS2 knockdown-induced mitophagic response. The indicated control (sgNC) or NLRX1 knockout (sgNLRX1) cells were transfected with the indicated siRNAs. a, d, g, j, m, Total cell or Mito lysates for immunoblotting. n = 3 biological replicates. b, e, h, k, n, Cells expressing mt-Keima reporter treated as indicated for flow cytometry analysis. c, f, i, l, o, qPCR analysis of mtDNA/nDNA. b, c, e, f, h, i, k, l, n, o, n = 3 biological replicates. Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test.

Source Data

Cytosolic AcCoA reduction initiates general autophagy through EP300-mediated protein acetylation³. Notably, NLRX1 depletion had little effect on p62 degradation (Extended Data Fig. 7a,b), overall protein acetylation (Extended Data Fig. 7c) and LC3 lipidation (Extended Data Fig. 7d) after cytosolic AcCoA reduction, and LC3 recruitment to mitochondria was largely blunted in NLRX1-deficient cells (Fig. 1g and Extended Data Fig. 6a,d). NLRX1 deficiency did not affect cytosolic AcCoA levels in the tested cells (Extended Data Fig. 7e). Furthermore, NLRX1 deficiency did not alter general-starvation-induced (Earle’s balanced salt solution, EBSS) autophagosome formation, on the basis of LC3 staining, transmission electron microscopy (TEM) analysis (Extended Data Fig. 7f,g) or p62 degradation (Extended Data Fig. 7h). However, EBSS-induced mitophagy was significantly abolished by BafA1 or NLRX1 deficiency (Extended Data Fig. 7h,i). Thus, cytosolic AcCoA-reduction-induced mitophagy is highly selective and requires the mitophagy receptor NLRX1.

Extended Data Fig. 7. NLRX1 depletion specifically inhibits mitophagy but not general autophagy.

a-c, NLRX1 does not affect p62 degradation (a,b) or pan-protein acetylation (c) upon treatments of SM and SLC25A1 or ACLY inhibition. Control (sgNC) or NLRX1 knockout (sgNLRX1) HeLa cells cultured with SM, HC (20 mM), or BTC (5 mM) for 16 h for total lysates immunoblotting. n = 3 biological replicates. d, NLRX1 deficiency showed little effect on LC3 lipidation. Control (sgNC) or NLRX1 knockout (sgNLRX1) HeLa cells cultured with SM or HC (20 mM) for 16 h with or without Baf A1 (100 nM) for immunoblotting. The ratio represents LC3 II/LC I. n = 3 biological replicates. e, NLRX1 knockout does not affect the level of Cyto AcCoA. The level of AcCoA in the indicated control (sgNC) or NLRX1 knockout (sgNLRX1) cells was detected by LC-MS. n = 3 biological replicates. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test. f-i, Control (sgNC) or NLRX1 knockout (sgNLRX1) HeLa cells cultured with EBSS for 2 h with or without Baf A1 (100 nM) for confocal microscopy (f), electron microscopy (EM) (g), immunoblotting (h), or cells expressing mt-Keima reporter for flow cytometry analysis (i). f, Left panel: representative cell images are shown. Scale bar, 10 μm. Right panel: the statistical analysis of the numbers of LC3 puncta per cell. n = 47 cells. Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test. g, Asterisk indicates autophagosome/autolysosome, scale bar, 1 μm. h,i, n = 3 biological replicates. Data in (f,i) shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test (i). j, Blunted mitophagy after HC treatment in the liver of Nlrx1^−/− mice. Wild type (WT) or Nlrx1^−/− mice intraperitoneally injected with PBS or HC (100 mg per kg) for 4 h, and qPCR analysis of total liver tissue mtDNA/nDNA (n = 3 mice per group). Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test. k-m, NLRX11 is required for fasting-induced mitophagy in Gastro tissue. WT or Nlrx1^−/− mice fasted for 24 h. k, Total or Mito Gastro lysates for immunoblotting. n = 3 mice per group. l, qPCR analysis of mtDNA/nDNA. WT Ad libitum or WT fasted group (n = 4 mice), Nlrx1^−/− Ad libitum or Nlrx1^−/− fasted group (n = 3 mice). Two-way ANOVA and Bonferroni’s multiple comparisons test. m, Analysis of fasting-induced mitophagy in Nlrx1^−/− mice using AAV-delivered mt-Keima reporter. Left panel: representative Gastro images shown. Scale bar, 10 μm. Right panel: relative mt-Keima signal values in each group normalized to the WT Ad libitum group. n = 24 images from 4 mice per group. Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test. n,o, NLRX1 is required for ACLY inhibition-induced mitophagy in liver tissue. Nlrx1 was knocked down by AAV-carried shRNA in liver tissue. Control (shNC) or Nlrx1 knockdown (shNlrx1) C57BL/6 male mice intraperitoneally injected with HC for 4 h, then liver tissues harvested for immunoblotting (n) or qPCR analysis of mtDNA/nDNA (o). n, Left panel: representative immunoblotting. Right panel: relative quantification of TIM23/Tubulin or CYTB/Tubulin level. n = 5 mice in (n) and n = 6 mice in (o). Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test. p,q, NLRX1 is required for fasting-induced mitophagy in Gastro tissue. Nlrx1 was knocked down by AAV-carried shRNA in Gastro tissue. shNC or shNlrx1 C57BL/6 male mice fasted for 24 h, then Gastro tissues were harvested for immunoblotting (p) or qPCR analysis of mtDNA/nDNA (q). p, Left panel: representative immunoblotting. Right panel: relative quantification of TIM23/Tubulin or CYTB/Tubulin level. n = 6 mice in (p) and n = 3 mice in (q). Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test. r,s, Nlrx1^−/− or Nlrx1 knockdown does not affect mice weight. WT and Nlrx1^−/− mice or shNC or shNlrx1 C57BL/6 mice were ad libitum-fed or fasted for 24 h with free access to water and mice weight were determined. n = 6 mice per group. Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test.

Source Data

HC- or fasting-induced mitochondrial degradation and LC3 recruitment on mitochondria were profoundly negated in both the liver and gastrocnemius tissues of Nlrx1^−/− mice (Fig. 1h and Extended Data Fig. 7j–l). To further explore NLRX1-mediated mitophagy in vivo, we established an AAV-delivered mt-Keima reporter system in Nlrx1^−/− mice. The basal acidic mt-Keima signals were decreased in both the liver and gastrocnemius tissues of Nlrx1^−/− mice (Fig. 1i,j and Extended Data Fig. 7m). Consistent with that, HC treatment or fasting-induced acidic mt-Keima signals were abolished in liver and gastrocnemius tissues of Nlrx1^−/− mice (Fig. 1i,j and Extended Data Fig. 7m). Moreover, Nlrx1 knockdown using AAV delivery system abolished HC- or fasting-induced mitochondria reduction in vivo (Extended Data Fig. 7n–q). NLRX1 deficiency or knockdown did not affect mouse weight during ad libitum feeding or after fasting (Extended Data Fig. 7r,s). Collectively, these results show that NLRX1 controls cytosolic-AcCoA-reduction-induced mitophagy both in vitro and in vivo.

NLRX1 contains an N-terminal mitochondrial targeting sequence (MTS), which could be cleaved after importing into the mitochondria²⁵. The uncleaved NLRX1 is retained in the cytosol to promote LC3 lipidation and recruitment to mitochondria after mitochondria protein import stress (MPIS), including treatment of CCCP or MitoBloCK-6 (MB-6), the inhibitor of the mitochondria protein import pathway MIA40–ERV1 (ref. ⁷). Indeed, uncleaved NLRX1 in the cytosol was found in CCCP-treated cells, but was not detected when reducing AcCoA in cells with endogenous or exogenous NLRX1 (Extended Data Fig. 8a,b), consistent with the result that SM culture or inhibition of ACLY or SLC25A1 induced little protein import stress (Extended Data Fig. 5d). Furthermore, restoration of full-length but not the cytosol-retained NLRX1(∆N-ter) (NLRX1 lacking MTS signal)²⁶ rescued the mitophagy induced by SM culture or inhibition of ACLY or SLC25A1 (Extended Data Fig. 8c).

Extended Data Fig. 8. Cytosol-facing NLRX1 mediates Cytosolic AcCoA reduction-induced cargo degradation through autophagy.

a,b, CCCP but not SM and SLC25A1 or ACLY inhibition induces endogenous NLRX1 (a) or exogenous NLRX1-HA (b) retention in the cytosol. Indicated cells cultured with SM, HC (20 mM), BTC (5 mM) for 16 h, or CCCP (30 µM) for 20 h. Cyto and Mito fractions for immunoblotting. n = 3 biological replicates. c, NLRX1 lacking MTS signal (∆N-ter) loses mitophagic response upon treatments of SM and SLC25A1 or ACLY inhibition. HeLa NLRX1-knockout cells stably expressed NLRX1 WT or NLRX1(∆N-ter) cultured with SM, HC (20 mM), BTC (5 mM) or SB (100 µM) for 16 h for total lysates immunoblotting. n = 3 biological replicates. d, A schematic of the split green fluorescent protein (GFP) system used to report NLRX1 localization on different compartments of mitochondria. NLRX1 fused with a C-terminal GFP11 expressed in cells stably expressing GFP_1–10 in the cytosol (cytoGFP_1–10) or Mito matrix (matrixGFP_1–10) along with the red fluorescent protein (RFP) as the translation normalization marker. NLRX1 localization in either compartment led to the complementation of GFP fluorescence. e,f, NLRX1 is simultaneously present in the Mito matrix and outer membrane, and the distribution is not affected by SM. e, Representative cell images shown. HSP60_GFP11 and _GFP11TOM20 serve as the Mito matrix and outer membrane localization controls, respectively. Images processed by a deconvolution algorithm through the Cellsens Dimension Desktop from Olympus company. Scale bar, 10 μm. f, GFP and RFP fluorescence analysed by flow cytometry, and the percentages of GFP⁺-RFP⁺/RFP⁺ calculated to reflect the distributions of NLRX1. n = 3 biological replicates. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test. g, A schematic of NLRX1(Cyto) or NLRX1(ER) chimeric protein generation. Substitution of the MTS of NLRX1 with the ER-localization sequence of FAM134b (NLRX1(ER)) leads to NLRX1 localization on ER with a topology facing the cytosol while NLRX1 lacking MTS (NLRX1(Cyto)) localizes in the cytosol. h, Validation of NLRX1(ER) localization on ER. HeLa cells stably expressing NLRX1(ER) or NLRX1(Cyto) with Flag tag were fixed and stained for Flag and the ER marker CLIMP63. Quantification of Pearson’s colocalization coefficient between Flag and CLIMP63. n = 38 (NLRX1(Cyto)), n = 19 (NLRX1(ER)) cells. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test. i,j, NLRX1(ER) but not NLRX1(Cyto) promotes SLC25A1 or ACLY inhibition-induced CLIMP63 reduction, while Baf A1 could block NLRX1(ER)-mediated CLIM63 reduction. Cells treated with HC (20 mM) or BTC (5 mM) for 16 h with or without Baf A1 (100 nM). Total lysates for immunoblotting. n = 3 biological replicates. k, Mito protein import stress (MPIS) decreases Cyto AcCoA level. HeLa cells treated with MPIS inducer MitoBloCK-6 (MB-6, 100 μM) for 16 h for Cyto AcCoA level analysis by LC-MS and normalized to Actin level. n = 3 biological replicates. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test. l, CCCP decreases Cyto AcCoA level. HeLa NLRX1-knockout cells stably expressed NLRX1 WT treated with 20 μM CCCP for 6 h for Cyto AcCoA level analysis by LC-MS and normalized to Actin level. n = 3 biological replicates. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test. m,n, Ace blocks MPIS- and CCCP-induced mitophagy. Indicated cells were treated with MB-6 (100 μM) for 16 h or 20 μM CCCP for 6 h with or without 10 mM Ace for total lysates immunoblotting. n = 3 biological replicates. o, MB-6 inhibits PDH complex importation into mitochondria. Data were collected from a published data resource²⁹ and shown as log₂ of the mean MB-6/DMSO ratio and P value. p, A schematic of a working model illustrating that MB-6 inhibits the import of PDH-E1 into mitochondria, thereby blocking pyruvate-derived AcCoA production. PDH-E1, the complex of PDHA1 and PDHB. q, PDH-E1 knockdown decreases Cyto AcCoA level. HeLa cells were transfected with the indicated siRNAs for Cyto AcCoA analysis by LC-MS and normalized to Actin. n = 4 biological replicates. Data shown as the mean ± s.e.m. Unpaired two-tailed Student’s t-test. r, Ace could rescue PDH-E1 knockdown-induced mitophagy. HeLa cells were transfected with the indicated siRNAs with or without 10 mM Ace for immunoblotting. n = 3 biological replicates. s, MB-6 induces weak mitophagy after PDH-E1 deficiency. HeLa cells were transfected with the indicated siRNAs with or without MB-6 (100 μM) treatment for 16 h for immunoblotting. n = 3 biological replicates. t, CCCP inhibits multiple protein importation into mitochondria. Data collected from a published data resource²⁹ and shown as log₂ of the mean CCCP/DMSO ratio and P value.

Source Data

Next, we established split green fluorescent protein (GFP) reporters localizing in the cytosol (cytoGFP) or mitochondrial matrix (matrixGFP)²⁷ and found the presence of NLRX1 in both the mitochondrial matrix and outer membrane (Extended Data Fig. 8d–f). About 45% of cells had outer-mitochondrial-membrane-localized NLRX1, while 70% of cells had matrix-localized NLRX1, and the distribution was unaffected after SM culture (Extended Data Fig. 8f). Importantly, the intact mitochondrial membrane before and after AcCoA reduction (Extended Data Fig. 5e) suggested that only outer-membrane-localized NLRX1 could gain access to the cytosolic LC3-decorated autophagosome to mediate mitophagy. To demonstrate that the cytosol-facing NLRX1 is sufficient to induce cargo degradation, we substituted the endoplasmic reticulum (ER)-localization sequence of FAM134b with the MTS of NLRX1 to generate a chimeric ER-localized NLRX1, which was exposed to the cytosolic face^6,28 (Extended Data Fig. 8g,h). ACLY or SLC25A1 inhibition induced ER degradation, evidenced by decreased ER protein CLIMP63 in NLRX1(ER) but not NLRX1(Cyto) cells (Extended Data Fig. 8i), which were blocked by BafA1 (Extended Data Fig. 8j). Together, the mitochondrial-outer-membrane-localized NLRX1 mediates cytosolic AcCoA reduction-induced mitophagy.

MPIS-led mitochondrial dysfunction and mitophagy depend on NLRX1 (ref. ⁷). Indeed, both the MPIS inducer MB-6 and CCCP caused cytosolic AcCoA reduction in HeLa cells (Extended Data Fig. 8k,l), and acetate supplementation blocked MB-6 or CCCP-induced mitophagy (Extended Data Fig. 8m,n). To examine how MPIS decreased AcCoA, we surveyed the literature and found that among the metabolic proteins that contribute to cytosolic AcCoA production, the pyruvate dehydrogenase (PDH) complex that converts pyruvate into AcCoA³ was significantly decreased in the mitochondria after MB-6 treatment, in a proteomics analysis²⁹ (Extended Data Fig. 8o). We speculated that MB-6 might reduce cytosolic AcCoA through downregulating PDH complex importation in mitochondria (Extended Data Fig. 8p). Indeed, knockdown of the PDH complex (PDH-E1, encoded by PDHA1 and PDHB) significantly decreased cytosolic AcCoA levels (Extended Data Fig. 8q), while acetate supplementation restored the mitochondrial content reduced by silencing of PDH-E1 (Extended Data Fig. 8r). Notably, PDH-E1 knockdown in combination with MB-6 slightly decreased MT-CO2 levels compared with MB-6 alone (Extended Data Fig. 8s), indicating that the mitophagic effect of MB-6 might partially depend on PDH-E1. Besides the PDH complex, CCCP induced more-potent protein import suppression by affecting citrate synthase, SLC25A1 and mitochondrial pyruvate carrier (Extended Data Fig. 8t), leading to cytosolic AcCoA reduction (Extended Data Fig. 8l). Thus, cytosolic AcCoA reduction acts as the unified signalling metabolite to control NLRX1-mediated mitophagy response.

NLRX1 senses cytosolic AcCoA

AcCoA levels are commonly linked to protein acetylation³⁰. However, SM culture with or without acetate did not alter NLRX1 acetylation levels in HEK293T cells stably expressing wild-type NLRX1–Flag (Extended Data Fig. 9a). To define whether NLRX1 directly senses cytosolic AcCoA, we incubated synthesized biotin–AcCoA (biotin was conjugated to the amino group of AcCoA) with cell lysates, followed by streptavidin pull-down. Notably, the interaction of both exogenous and endogenous NLRX1 and biotin–AcCoA was strongly enhanced after SM culture compared with under normal conditions, while supplementation with acetate reduced the association between biotin–AcCoA and NLRX1 (Fig. 2a,b). By contrast, biotin–AcCoA hardly pulled-down NLRP3, another NLR protein (Extended Data Fig. 9b). Notably, supplementing excess AcCoA in mitochondrial lysates competed for the interaction between NLRX1 and biotin–AcCoA, with competition saturated above 0.5 mM (Fig. 2c). To determine whether NLRX1 directly binds to AcCoA, we obtained the purified recombinant NLRX1 (87–975 amino acids) with a N-terminal MBP (maltose-binding protein) tag using insect cells (Extended Data Fig. 9c). In an in vitro pull-down assay, recombinant MBP–NLRX1 bound to biotin–AcCoA, and increasing concentrations of AcCoA competed with biotin–AcCoA for NLRX1 binding. The saturated competition was above 25 µM (Extended Data Fig. 9c), indicating the direct association.

Extended Data Fig. 9. NLRX1 directly binds to AcCoA at conserved sites within the LRR domain.

a, NLRX1 acetylation is not affected by AcCoA level. HEK293T cells stably expressing NLRX1 with Flag tag (NLRX1-Flag) cultured with SM for 16 h with or without 10 mM Ace for immunoprecipitation. n = 3 biological replicates. b, AcCoA binds with NLRX1 but not NLRP3. Constructs of NLRP3 with HA tag (HA-NLRP3) or NLRX1 with HA tag (NLRX1-HA) were transfected in HEK293T cells, followed by culture with SM for 6 h and biotin-AcCoA pull-down to enrich HA-tagged proteins for immunoblotting. n = 3 biological replicates. c, AcCoA competitively decreases the interaction of biotin-AcCoA with recombinant NLRX1. MBP-NLRX1 (87-975aa) purified from insect cells. AcCoA were added as indicated concentrations and biotin-AcCoA was used to pulldown NLRX1, followed by immunoblotting. Left panel: Coomassie staining of gel-filtrated NLRX1 recombinant with MBP tag (MBP-NLRX1) from insect cells. Upper panel of right panel: representative immunoblotting image. Bottom panel of right panel: quantification of competitive binding between NLRX1 and biotin-AcCoA. n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Tukey’s multiple comparisons test. d, 729/754/758/958 sites of NLRX1 are conserved across species. e, Purification of NLRX1-LRR WT and NLRX1-LRR(4A). Left panel: Gel filtration of recombinant MBP, MBP-NLRX1-LRR WT, or MBP-NLRX1(4A) proteins on a Superose 6 column (UV A280). Right panel: Coomassie staining of gel-filtrated recombinant proteins from E. coli. n = 3 biological replicates. f, Affinity of recombinant NLRX1 with acyl-CoAs. MBP-NLRX1 purified from insect cells. Biotin-conjungated acyl-CoAs were used to pull down NLRX1, followed by immunoblotting analysis. n = 3 biological replicates. g, AcCoA exhibits higher affinity with recombinant NLRX1 than CoASH. MBP-NLRX1 (87-975aa) was purified from insect cells. Biotin-AcCoA or CoASH was used to pull down NLRX1 in the presence of indicated metabolites, followed by immunoblotting. n = 3 biological replicates. h, Molecular docking shows that CoASH binds to NLRX1-LRR. Left panel, illustrates three critical residues (Glu729, Lys754, Arg958) at NLRX1-LRR involved in binding to AcCoA. Right panel, the binding affinity calculated by docking analyses. i, Cellular thermal shift assay (CETSA) confirms endogenous NLRX1 associates with AcCoA. HeLa cells cultured with SM for 6 h, lysed and supernatants were collected. Lysates incubated with AcCoA (500 μM), CoASH (500 μM) or control (PBS), followed by heating at different temperatures (range 44.6 to 65 °C). Soluble fractions for immunoblotting. j, pH affects the binding between NLRX1 and AcCoA. HEK293T cells stably expressing NLRX1-HA cultured with SM for 6 h. Isolated Mito lysate in buffers with different pH values were subjected to biotin-AcCoA pull-down assay, followed by immunoblotting. n = 3 biological replicates.

Source Data

Fig. 2. NLRX1 directly senses AcCoA.

a,b, Acetate (Ace) blocks SM-induced binding between NLRX1 and biotin–AcCoA. HEK293T NLRX1–HA cells (a) and A549 cells (b) were treated as indicated for 6 h with or without acetate (10 mM), and biotin–AcCoA was used to pull down NLRX1. n = 3 biological replicates. c, AcCoA competitively decreases the interaction between biotin–AcCoA and NLRX1. Top, representative immunoblotting image. Bottom, quantification of competitive binding between NLRX1 and biotin–AcCoA. n = 4 biological replicates. Data are mean ± s.e.m. Statistical analysis was performed using ordinary one-way ANOVA with Tukey’s multiple-comparison test. d, Biotin–AcCoA binds to the LRR domain of NLRX1. n = 3 biological replicates. e,f, Molecular docking analysis of the binding of AcCoA and NLRX1. e, AcCoA binds to the classical pocket of NLRX1–LRR. f, Four critical residues (Glu729, Lys754, Gln758, Arg958) at NLRX1–LRR are critical for binding to AcCoA. g, NLRX1(4A) blocks the interaction between NLRX1 and AcCoA. n = 3 biological replicates. h, MBP–RR(WT) binds to AcCoA with a K_d of about 6.6 µM. Purified recombinant MBP–LRR(WT) or MBP was incubated with 2 µM [³H]AcCoA and the indicated concentrations of AcCoA for scintillation counting, presented as disintegrations per minute (DPM). n = 5 (MBP–LRR(WT)) and n = 3 (MBP). Data are mean ± s.d. i, MBP–LRR(4A) impairs the AcCoA affinity of the LRR domain. The indicated recombinant protein was incubated with 2 µM [³H]AcCoA for scintillation counting. n = 3 biological replicates. Data are mean ± s.e.m. Statistical analysis was performed using ordinary one-way ANOVA with Dunnett’s multiple-comparison test.

Source Data

To explore the binding domain of AcCoA on NLRX1, we performed a domain mapping assay and found that the LRR domain was crucial for NLRX1–AcCoA binding (Fig. 2d). In silico molecular docking analysis using the experimental structure of LRR³¹ illustrated that the LRR domain of NLRX1 contains four conserved sites (Glu729, Lys754, Gln758 and Arg958) that form a pocket and are potentially crucial for AcCoA binding (Fig. 2e,f and Extended Data Fig. 9d). Among these, the Lys754 and Arg958 sites bind to the negative phosphate groups, and the Gln758 site binds to the acetyl-group of AcCoA, with an AcCoA-interaction mode similar to that of bacterial hybrid malic enzymes that have the phosphotransacetylase domain³². Indeed, mutation of all four sites to alanine (E729A, K754A, Q758A and R958A; hereafter 4A) significantly blocked the biotin–AcCoA pull-down (Fig. 2g).

We next purified MBP–LRR(WT) or MBP–LRR(4A) protein from a bacterial expression system using gel filtration (Extended Data Fig. 9e). Competition binding with increasing concentrations of unlabelled AcCoA to radiolabelled AcCoA showed that recombinant MBP–LRR directly bound to AcCoA, and the dissociation constant (K_d) value was approximately 6.6 µM (Fig. 2h), which is within the reported range of cytosolic AcCoA concentrations³³. Moreover, the AcCoA affinity of MBP–LRR(4A) was significantly impaired (Fig. 2i). Collectively, these results show that NLRX1 senses AcCoA by directly binding at the conserved pocket on the LRR domain.

Besides AcCoA, we also tested the association between NLRX1 and other related CoA metabolites, such as CoASH, malonyl-CoA and succinyl-CoA. According to the results of pull-down experiments, no interaction between recombinant NLRX1 and malonyl-CoA or succinyl-CoA was detected, while CoASH was associated with recombinant NLRX1 to a lesser extent than AcCoA was (Extended Data Fig. 9f). Furthermore, the result of competition assay of CoASH to biotin–CoASH-bound NLRX1 demonstrated that their association is specific (Extended Data Fig. 9g). The in silico docking analysis revealed a more-favourable binding energy for AcCoA to NLRX1 (ΔG = −7.804 kcal mol⁻¹) than CoASH (ΔG = −5.56 kcal mol⁻¹) (Extended Data Fig. 9h). CoASH was predicted to bind to NLRX1 through three major sites (Glu729, Lys754 and Arg958) (Extended Data Fig. 9h), compared with the four AcCoA-bound sites (Glu729, Lys754, Arg958 and Gln758) (Fig. 2f). We further confirmed that AcCoA has stronger NLRX1-binding affinity than CoASH using the competition assays. Increasing concentrations of AcCoA effectively reduced the interaction between biotin–CoASH and NLRX1, but not vice versa (Extended Data Fig. 9g), corroborating the in silico prediction result that NLRX1 has a higher binding affinity for AcCoA than for CoASH (Extended Data Fig. 9h). We next tested the intracellular association between NLRX1 and CoASH or AcCoA using a cellular thermal shift assay. AcCoA addition substantially increased the stability of NLRX1 protein level compared with CoASH (Extended Data Fig. 9i), indicating that AcCoA has stronger binding to NLRX1 than CoASH within cells. Notably, the CoASH level remained steady or slightly increased after nutrient starvation³³. Thus, NLRX1 binds to AcCoA rather than to other related CoA metabolites, under physiological conditions.

Mitochondrial and cytosolic AcCoA are different pools because AcCoA cannot diffuse freely across cellular membranes¹. ACLY or SLC25A1 inhibition decreased AcCoA levels in the cytosol but not in the mitochondria (Extended Data Fig. 2a), indicating that the cytosolic AcCoA pool is sufficient to induce NLRX1 activation. Compared with the neutral pH (around 7.2) in the cytosol, the mitochondrial matrix has a highly alkaline environment (pH of around 8.0)³⁴. The isoelectric points of NLRX1 protein are about 6.8–7.4 according to different scales³⁵, while two positive charge residues (Lys754, Arg958) among the four major AcCoA-binding amino acids are predicted to associate with the negative phosphate group of AcCoA (Fig. 2f). Thus, to define whether the pH affects the association between NLRX1 and AcCoA, the mitochondria were isolated from HEK293T NLRX1–HA cells and lysed using buffers with different pH values (6.8, 7.4 or 8.0). The amount of NLRX1 obtained through biotin–AcCoA pulldown was substantially decreased in pH 8.0 buffer compared with in the neutral buffers (Extended Data Fig. 9j), suggesting that alkaline condition could disrupt the association between NLRX1 and AcCoA. In summary, NLRX1 mainly senses cytosolic AcCoA.

AcCoA controls NLRX1 oligomerization

NLRX1 has a conserved LC3-binding site (LIR) on the NACHT domain, which recruits LC3-decorated autophagosome for mitochondrial degradation after activation⁶. Notably, SM and CCCP promoted the colocalization of LC3 with NLRX1 at both the endogenous and exogenous levels, which was repressed by addition of acetate (Fig. 3a and Extended Data Fig. 10a–c). Meanwhile, ACLY or SLC25A1 inhibition also increased colocalization of NLRX1 with LC3 (Extended Data Fig. 10d,e). Moreover, we reintroduced WT or LIR-deficient (∆LIR) NLRX1 into NLRX1-knockout cells to determine the induction of mitophagy. Notably, SM culture or ACLY or SLC25A1 inhibition did not trigger mitophagy in NLRX1-deficient (empty vector) cells (Fig. 3b,c and Extended Data Fig. 10f–h). Reconstitution of cells with NLRX1 WT successfully restored mitophagy after the stimulation, but reconstitution with NLRX1(∆LIR) did not (Fig. 3b,c and Extended Data Fig. 10f–h). Collectively, these results show that cytosolic AcCoA reduction-induced mitophagy requires the association of NLRX1 with LC3.

Fig. 3. AcCoA triggers NLRX1 autoinhibition.

a, Acetate blocks SM-triggered colocalization of endogenous NLRX1 and LC3. HeLa cells with HA-tagged knock-in (NLRX1–HA KI) were treated with SM for 12 h with or without acetate (10 mM), followed by confocal microscopy analysis. Left, representative microscopy images. Scale bars, 10 μm. Right, quantification of Pearson’s colocalization coefficient. n = 21 (mock), n = 88 (SM) and n = 34 (SM + acetate) cells. Data are mean ± s.e.m. Statistical analysis was performed using Kruskal–Wallis test with Dunn’s multiple-comparison test. b, NLRX1(ΔLIR) does not induce mitophagy. Immunoblot (IB) images of total HeLa cell or mitochondrial lysates as indicated. n = 3 biological replicates. c, NLRX1(ΔLIR) induces mitophagy, on the basis of the mt-Keima reporter analysis. n = 3 biological replicates. Data are mean ± s.e.m. Statistical analysis was performed using two-way ANOVA with Bonferroni’s multiple-comparison test. d, Schematic of NLRX1 oligomerization and association with LC3 in response to AcCoA signalling. e, Addition of AcCoA but not CoASH promotes the association of the NACHT domain with the WT LRR domain. n = 3 biological replicates. f, NLRX1 oligomerization is controlled by cytosolic AcCoA. Total lysates from HEK293T cells expressing HA-tagged WT NLRX1 were treated as indicated for SDS–agarose electrophoresis. n = 3 biological replicates. g,h, SM promotes NLRX1 WT but not NLRX1(4A) oligomerization (g) or binding to GST–LC3 (h). Cells were treated as indicated for SDS–agarose electrophoresis (g) or GST pull-down (h). n = 3 biological replicates. i, AAV-mediated NLRX1(4A) constitutively induces mitophagy in the liver tissue of C57BL/6 mice. Liver tissue or mitochondrial lysates were used for immunoblotting. The ratio represents CYTB/tubulin. j, Quantification of relative CYTB/tubulin levels in Fig. 3i. n = 4 mice per group. Data are mean ± s.e.m. Statistical analysis was performed using ordinary one-way ANOVA with Dunnett’s multiple-comparison test. EV, empty vector.

Source Data

Extended Data Fig. 10. AcCoA reduction promotes NLRX1 oligomerization and binding to LC3.

a, Ace blocks the colocalization of NLRX1-HA and GFP-LC3 triggered by SM. NLRX1-knockout HeLa cells stably expressing NLRX1 WT were transfected with the vector expressing GFP-LC3 and treated with SM for 10 h with or without 10 mM Ace, followed by confocal microscopy. Left panel: representative cell images are shown. Scale bar, 10 μm. Upper panel of the right panel: validation of the expression of HA-tagged NLRX1 by immunoblotting. Bottom panel of right panel: quantification of Pearson’s colocalization coefficient. n = 19 (Mock), n = 39 (SM) and n = 32 (SM+Ace) cells. Data shown as the mean ± s.e.m. Kruskal-Wallis test and Dunn’s multiple comparisons test. b, Ace blocks the colocalization of endogenous NLRX1 and LC3 triggered by CCCP. HeLa cells with HA-tagged knock-in (NLRX1-HA KI) were generated. Cells treated with 20 μM CCCP for 6 h with or without 10 mM Ace, followed by confocal microscopy. Left panel: representative cell images shown. Scale bar, 10 μm. Upper panel of the right panel: Validation of NLRX1-HA KI HeLa cells by immunoblotting. Bottom panel of right panel: quantification of Pearson’s colocalization coefficient. n = 39 (Mock), n = 90 (CCCP) and n = 34 (CCCP + Ace) cells. Data shown as the mean ± s.e.m. Kruskal-Wallis test and Dunn’s multiple comparisons test. c, Ace blocks the colocalization of NLRX1-HA and GFP-LC3 triggered by CCCP. HeLa-NLRX1^WT cells treated with 20 μM CCCP for 1.5 h with or without 10 mM Ace, followed by confocal microscopy. Left panel: representative cell images are shown. Scale bar, 10 μm. Right panel: quantification of Pearson’s colocalization coefficient between NLRX1-HA and GFP-LC3. n = 23 (Mock), n = 46 (CCCP) and n = 33 (CCCP + Ace) cells. Data shown as the mean ± s.e.m. Kruskal-Wallis test and Dunn’s multiple comparisons test. d, SLC25A1 or ACLY inhibition promotes the colocalization of endogenous NLRX1 with LC3. HeLa-NLRX1(HA KI) cells were treated with HC (20 mM), BTC (5 mM), or SB (100 μM) for 12 h, followed by confocal microscopy. Left panel: representative cell images shown. Scale bar, 10 μm. Right panel: quantification of Pearson’s colocalization coefficient. n = 27 (Mock), n = 30 (HC), n = 50 (BTC) and n = 36 (SB) cells. Data shown as the mean ± s.e.m. Kruskal-Wallis test and Dunn’s multiple comparisons test. e, SLC25A1 or ACLY inhibition promotes the colocalization of NLRX1-HA with GFP-LC3. HeLa-NLRX1^WT cells were treated with HC (20 mM), BTC (5 mM), or SB (100 μM) for 10 h, followed by confocal microscopy. Right panel: representative cell images shown. Scale bar, 10 μm. Left panel: quantification of Pearson’s colocalization coefficient between NLRX1-HA and GFP-LC3. n = 33 (Mock), n = 46 (HC), n = 29 (BTC) and n = 48 (SB) cells. Data shown as the mean ± s.e.m. Kruskal-Wallis test and Dunn’s multiple comparisons test. f-h, NLRX1 LIR deletion mutant (NLRX1^ΔLIR) fails to induce mitophagy after SLC25A1 or ACLY inhibition. NLRX1 knockout-HeLa cells stably expressing NLRX1WT or NLRX1(ΔLIR) with HA tag were treated with HC (20 mM), BTC (5 mM), or SB (100 μM) for 16 h. Total cell or Mito lysates for immunoblotting (f), cells expressing mt-Keima reporter for flow cytometry analysis (g) or qPCR analysis of mtDNA/nDNA (h). n = 3 biological replicates. Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test. i, LRR(4A) decreases the binding with NACHT in the complete medium. NACHT-HA and Flag-LRR WT or Flag-LRR(4A) were co-expressed in HEK293T cells for Flag-LRR immunoprecipitation to determine the binding. n = 3 biological replicates. j, Ace blocks CCCP-induced NLRX1 oligomerization. HEK293T cells stably expressing NLRX1 with HA tag (NLRX1-HA) cultured with 20 μM CCCP for 3 h with or without 10 mM Ace for SDS–agarose electrophoresis. n = 3 biological replicates. k, NLRX1(4A) mutant increases the binding with GST-LC3 in the complete medium. HEK293T cells transfected with constructs expressing NLRX1 WT or NLRX1(4A) for GST pull-down assay. n = 3 biological replicates. l,m, Baf A1 blocks NLRX1(4A) constitutively induced-mitophagy. NLRX1^WT, NLRX1^4A, or NLRX1^∆LIR were transfected into NLRX1-knockout HeLa cells, respectively. Baf A1 (200 nM) was added 3 h before harvesting. NLRX1-knockout cells expressing mt-Keima for flow cytometry analysis (l). qPCR analysis of mtDNA/nDNA (m). n = 3 biological replicates. Data shown as mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test. n,o, AAV-mediated NLRX1(4A) overexpression constitutively induces mitophagy. Total or Mito Gastro tissue lysates harvested for immunoblotting (n). Left panel: representative immunoblotting, the ratio represents CYTB/Tubulin. Right panel: quantification of relative CYTB/Tubulin level. n = 3 mice per group. Data shown as mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test. qPCR analysis of mtDNA/nDNA (o) in the liver tissue (left panel) or Gastro tissue (right panel). n = 4 mice per group except EV group in Gastro tissue (n = 3 mice). Data shown as mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test for all groups except the liver mtCO1 group (Welch one-way ANOVA and Dunnett’s T3 multiple comparisons).

Source Data

Previous studies showed the mode of NLRX1 autoinhibition: under a quiescent state, the LRR domain associates with the NACHT domain to inhibit the association of NLRX1 with LC3 at the LIR site. After activation, LRR disassociates with NACHT through an unknown mechanism to release the autoinhibition, promoting oligomerization of NLRX1 and its binding to LC3 (ref. ⁶). On the basis of these premises, we hypothesized that AcCoA could stabilize the autoinhibition state of NLRX1 by enhancing the interaction between LRR and NACHT (Fig. 3d). Indeed, biotin–AcCoA did not bind to the NACHT domain but, when co-transfected with the LRR domain, biotin–AcCoA readily pulled-down NACHT, indicating the formation of the NACHT–AcCoA–LRR complex (Fig. 2d). Moreover, adding AcCoA but not CoASH in cell lysates co-expressing NACHT and LRR domains strengthened LRR binding to the NACHT domain (Fig. 3e). By contrast, LRR(4A) decreased binding to NACHT under NM (Extended Data Fig. 10i) and did not respond to AcCoA after SM culture (Fig. 3e). Furthermore, ACLY or SLC25A1 inhibition reduced the intensity of the monomer band while enhanced the oligomeric band of NLRX1 WT under indicated cross-link conditions (Fig. 3f). Adding acetate blocked SM- or CCCP-promoted oligomerization and restored the monomer band of NLRX1 WT (Fig. 3f and Extended Data Fig. 10j). Consistent with that, the oligomerization level of NLRX1(4A) was higher than that of NLRX1 WT and could not be upregulated by SM culture (Fig. 3g). Moreover, SM culture strengthened the association of NLRX1 WT with LC3 (Fig. 3h). Compared with NLRX1 WT, NLRX1(4A) constitutively increased binding to LC3 (Extended Data Fig. 10k) under NM and did not respond to SM culture (Fig. 3h).

On the basis of the results above, we first found that reintroduction of NLRX1^4A into NLRX1-deficient HeLa cells promoted acidic mt-Keima signal levels under basal conditions, but reintroduction of NLRX1^ΔLIR did not (Extended Data Fig. 10l). BafA1 abolished the difference among NLRX1(4A) and other groups (Extended Data Fig. 10l), demonstrating the suppressive effect of AcCoA binding on NLRX1-mediated mitophagy. Furthermore, NLRX1^4A reintroduction into NLRX1-deficient cells significantly decreased the mtDNA/nDNA level compared with empty vector, NLRX1 WT or NLRX1(ΔLIR), which could be blocked by BafA1 (Extended Data Fig. 10m). Moreover, the introduction of Nlrx1^4A into liver or gastrocnemius tissues using AAVs constitutively induced mitophagy in the ad libitum-fed state, while Nlrx1^WT overexpression did not significantly alter mitophagy (Fig. 3i,j and Extended Data Fig. 10n,o). In summary, NLRX1-mediated mitophagy is controlled by cytosolic AcCoA levels through direct binding both in vitro and in vivo.

Mitophagy drives KRASi resistance

KRAS mutations drive up to 30% of human cancers; specifically, more than 90% of patients with human pancreatic ductal adenocarcinoma (PDAC) have KRAS mutations. Thus, KRAS inhibitors (KRASi) have broad therapeutic potential for cancer therapy³⁶. However, acquired drug resistance also emerges and limits the clinical benefit³⁷. Whether metabolic rewiring participates in KRASi-induced resistance remains unclear. Mitophagy has been implicated in drug resistance, such as chemoresistance³⁸, we therefore wondered whether mitophagy participates in KRASi-induced resistance. Notably, the KRAS(G12D) inhibitor MRTX1133 (ref. ³⁹) and the pan-RAS inhibitor RMC-6236 (ref. ³⁶) decreased ACLY expression and cytosolic AcCoA levels in mouse KPC (Kras^G12D/+Trp53^R172H/+) cells and human PDAC AsPC-1 cells with KRAS^G12D (Fig. 4a,b and Extended Data Fig. 11a,b). Consistent with that, both inhibitors could induce mitophagy analysed by mitochondrial protein, acidic mt-Keima signal and mtDNA/nDNA level (Fig. 4c,d and Extended Data Fig. 11c–f), which could be blocked by acetate (Extended Data Fig. 11g–l). Importantly, KRASi-induced mitophagy was almost completely abolished in NLRX1-deficient cells (Fig. 4e,f and Extended Data Fig. 11m–p). Collectively, these results show that KRASi induces NLRX1-dependent mitophagy through downregulating ACLY–AcCoA axis metabolism.

Fig. 4. KRAS inhibitors enhance drug resistance by NLRX1-dependent mitophagy.

a–d, KRASi decreases Acly mRNA (a) and cytosolic AcCoA levels (b) and induces mitophagy (c,d). KPC cells were treated for 24 h and analysed using quantitative PCR with reverse transcription (RT–qPCR) (a); liquid chromatography coupled with MS (LC–MS) (b), normalized to cell numbers; immunoblotting (c); or flow cytometry for cells expressing mt-Keima reporter (d). n = 3 biological replicates. Data are mean ± s.e.m. Statistical analysis was performed using ordinary one-way ANOVA with Dunnett’s multiple-comparison test (a and d) and unpaired two-tailed Student’s t-tests (b). e–h, Nlrx1 knockout abolishes the KRASi-induced mitophagic response (e,f), and elevates ROS levels (g) and NADP⁺/NADPH ratio (h). KPC cells were used for immunoblot analyses (e), and cells expressing mt-Keima reporter were used for flow cytometry (f), CM-H2DCFDA staining (g) and the NADP⁺/NADPH colorimetric assay (h). n = 3 biological replicates. Data are mean ± s.e.m. Statistical analysis was performed using two-way ANOVA with Bonferroni’s multiple-comparison test. i, Nlrx1 depletion decreases KPC cell viability after MRTX1133 treatment. Dose–response curves for MRTX1133 treatment, based on 5-day CellTiter-Glo assays. Half-maximal inhibitory concentration (IC₅₀) values are displayed at the top. n = 6 biological replicates. j, Nlrx1 knockout exacerbates the suppressive effect of MRTX1133 on tumour cell growth in vivo. NSG mice given subcutaneous injection of KPC cells were treated with vehicle or MRTX1133 (30 mg per kg, twice a day) when the tumour volume reached around 300 mm³. n = 5 mice per group. Data are mean ± s.e.m. Statistical analysis was performed using two-way ANOVA with Bonferroni’s multiple-comparison test. k,l, Impaired mitophagy and increased ROS production in Nlrx1-deficient tumours with MRTX1133 treatment. The mice were treated as in Fig. 4j, and after 6 d of treatment, tumours were collected for immunoblotting (k; n = 3 mice per group) or CM-H2DCFDA staining with counting by Image J (l; n = 30 images from 5 mice per group). Data are mean ± s.e.m. Statistical analysis was performed using two-way ANOVA with Bonferroni’s multiple-comparison test.

Source Data

Extended Data Fig. 11. KRAS inhibitors downregulate ACLY for drug resistance through mitophagy.

a, KRASi decrease ACLY mRNA level in a dose-dependent manner. AsPC-1 cells were treated with MRTX1133 or RMC-6236 over a two-point dose response for 24 h for RT-qPCR analysis. n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test. b, KRASi decrease Cyto AcCoA levels. AsPC-1 cells were treated with MRTX1133 (10 nM), or RMC-6236 (10 nM) for 24 h and the Cyto AcCoA level was measured by LC-MS and normalized by cell number. n = 3 biological replicates. Data shown as the mean ± s.e.m. Welch’s t-test for MRTX1133 analysis and unpaired two-tailed Student’s t-test for RMC-6236 analysis. c-f, KRASi decrease ACLY protein level and induce mitophagy in a dose-dependent manner. KPC (e) or AsPC-1 (c,d,f) cells treated with MRTX1133 or RMC-6236 over a three-point dose (c) or a two-point dose (d-f) for 24 h with or without Baf A1 (200 nM). Total cell or Mito lysates for immunoblotting (c), cells expressing mt-Keima reporter for flow cytometry analysis (d), or qPCR analysis of mtDNA/nDNA (e,f). n = 3 biological replicates. Data shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (d-f). g-l, Ace blocks KRASi-induced mitophagy. KPC (g,h,k) or AsPC-1 (i,j,l) cells treated with MRTX1133 (10 nM), or RMC-6236 (10 nM) for 24 h with or without 10 mM Ace. Total cell or Mito lysates for immunoblotting (g,i), cells expressing mt-Keima reporter for flow cytometry analysis (h,j), or qPCR analysis of mtDNA/nDNA (k,l). n = 3 biological replicates. Data are shown as the mean ± s.e.m. Ordinary one-way ANOVA and Dunnett’s multiple comparisons test (h,j,k,l). m-p, NLRX1 is required for KRASi-induced mitophagic response. Control (sgNC) or NLRX1 knockout (sgNLRX1) KPC (o) or AsPC-1 (m,n,p) cells treated with MRTX1133 (10 nM), or RMC-6236 (10 nM) for 24 h. Total cell or Mito lysates for immunoblotting (m), cells expressing mt-Keima reporter for flow cytometry analysis (n), or qPCR analysis of mtDNA/nDNA (o,p). n = 3 biological replicates. Data are shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test (n-p). q, Elevated ROS level in Nlrx1-depleted AsPC-1 cells after KRASi treatment. Control (sgNC) and NLRX1-depleting (sgNLRX1) AsPC-1 cells were treated with MRTX1133 (10 nM), or RMC-6236 (10 nM) for 24 h for CM-H2DCFDA staining by flow cytometry analysis. n = 3 biological replicates. Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test. r, NLRX1 depletion and mitophagy inhibitor Mdivi-1 decrease cell viability upon KRASi treatment. IC₅₀ values calculated based on the dose-response of MRTX1133 or RMC-6236 in control (sgNC) or NLRX1-depleting (sgNLRX1) cells by 3-day (AsPC-1 and A549) or 5-day (KPC) cell-titer-glo assays. n = 3 biological replicates for all analyses, except for KPC cells (MRTX1133, n = 6 biological replicates), A549 cells (Mdivi-1, n = 6 biological replicates), and AsPC-1 cells (sgNLRX1, n = 4 biological replicates). s, NAC rescues low-dose MRTX1133-induced cell death in Nlrx1-deficient cells. Control (sgNC) or Nlrx1-depleting (sgNlrx1) KPC cells were treated with or without 1 nM MRTX1133, and NAC (4 mM) was added on the second day. Four days later, cell viability was determined by cell-titer-glo assays. n = 3 biological replicates. Two-way ANOVA and Bonferroni’s multiple comparisons test. t, Mitophagy inhibitor Mdivi-1 decreases cell viability in KPC cells. Dose-response curves for MRTX1133 in combination with mitophagy inhibitor Mdivi-1 (20 μM) in KPC cells based on 5-day cell-titer-glo assays. IC₅₀ values are displayed on the top panel. n = 6 biological replicates. u,v, Nlrx1 knockout exacerbates the suppressive effect of MRTX1133 on tumour cell growth in vivo. sgNC and sgNlrx1 KPC cells were subcutaneously injected in NSG mice for vehicle or MRTX1133 (30 mg per kg, twice a day) treatment when tumour volume reached around 300 mm³. Tumour weight (u), images (v) of KPC-derived allograft in NSG mice. n = 5 mice per group. Data shown as the mean ± s.e.m. Two-way ANOVA and Bonferroni’s multiple comparisons test. w, MRTX1133 upregulates ROS production in Nlrx1-deficient tumour. Tumours were harvested after 6 days of indicated treatments. ROS level in tumour tissue sections determined by CM-H2DCFDA staining. n = 5 mice per group. Scale bar, 10 μm.

Source Data

Given that cellular redox management is a hallmark of cancers with KRAS mutations^40,41 and mitochondria are the centre of ROS production, we speculated that mitophagy is a stress response to KRASi therapy in tumour cells to sustain cell survival. Indeed, both ROS levels and the cellular NADP⁺/NADPH ratio were significantly elevated in NLRX1-deficient cells after KRASi treatment (Fig. 4g,h and Extended Data Fig. 11q), indicating that mitophagy inhibition increased oxidative response. Notably, KRASi exhibited a better tumour-killing effect on NLRX1-deficient PDAC and lung cancer cells with KRAS mutations (Fig. 4i and Extended Data Fig. 11r). Moreover, adding the antioxidant N-acetyl-L-cysteine (NAC) blocked low-dose-MRTX1133-induced cell death in Nlrx1-deficient KPC cells, highlighting the critical role of oxidative stress in mitophagy inhibition-induced cell death (Extended Data Fig. 11s). Mitophagy inhibitor Mdivi-1 combined with KRASi exhibited a synergized suppressive effect on cancer cells compared with KRASi alone (Extended Data Fig. 11r,t), suggesting the potential of targeting ACLY–AcCoA–NLRX1-axis-mediated mitophagy in anti-tumour therapy. Moreover, MRTX1133 administration decreased tumour volume, and Nlrx1 deficiency could synergistically enhance the suppressive effect of MRTX113 on tumour growth (Fig. 4j and Extended Data Fig. 11u,v). Notably, MRTX1133 suppressed Acly expression to a similar level in both control and Nlrx1-deficient tumours, while mitochondrial protein reduction was observed only in control tumours, not in Nlrx1-deficient tumours (Fig. 4k). Consistent with the in vitro results, the basal ROS levels were the same in both control and Nlrx1-deficient tumours. However, MRTX1133 significantly decreased ROS level in the tumour sections of the control group but elevated it in Nlrx1-deficient tumours (Fig. 4l and Extended Data Fig. 11w). Collectively, these results show that KRAS inhibition induces an Nlrx1-dependent mitophagy response to elicit drug resistance.

Discussion

Our data demonstrate that NLRX1 could function as the key mitophagy receptor to selectively mediate reduced cytosolic AcCoA-induced mitophagy both in vivo and in vitro without affecting general autophagy receptor p62 degradation. NLRX1 has previously been shown to mediate mitophagy after mitochondrial damage such as bacterial infection and mitochondria protein import stress^6,7. Here we show that cytosolic AcCoA is the bona fide ligand of NLRX1 in an acetyaltion- and AMPK-independent manner that underlies nutrient stress and mitochondria damage to control its activation.

The NLR family could not only detect pathogen-associated molecular patterns but also sense ‘altered-self’ signals to activate subsequent cascades, such as inflammasomes or autophagy for surveillance^42,43. A common working mode for this family is sensing ligands directly through the LRR domain, getting oligomerized and then initiating downstream signalling pathways⁴². Notably, we found that MPIS-induced cytosolic AcCoA reduction and the mitophagy could be countered by the addition of acetate, proving that cytosolic AcCoA functions as the altered-self signal to control NLRX1’s activation under various conditions. Thus, our study provides a unified model of NLRX1-mediated mitophagy.

Although mitophagy has been implicated in drug resistance³⁸, here we identify that rewired metabolism-driven mitophagy could be used by KRAS inhibitors to decrease oxidative stress and sustain tumour cell survival. Thus, mitophagy inhibition could be a sensitization strategy for KRAS inhibitors to enhance their anti-tumour efficacy. It will be interesting to explore whether this AcCoA–NLRX1 axis is also present in other physiological or pathological situations.

Methods

Mouse studies

Wild-type C57BL/6 male mice (6–8 weeks of age) were purchased from BIKAI. Nlrx1^−/− mice were purchased from Cyagen. NSG mice were purchased from Shanghai Model Organisms Center. All mice were housed in the specific-pathogen-free animal facility of Fudan University with the following environmental parameters: temperature maintained at 21–25 °C, relative humidity at 45–65% and a 12 h–12 h light–dark cycle.

All mice were randomly separated into each experiment group. Mice were fasted from 10:00 for 24 h with free access to water without food. PBS or HC (59847, Sigma-Aldrich; 100 mg per kg) was intraperitoneally injected into mice at 10:00 for 4 h. For acetate administration, mice were fasted for 24 h and PBS or sodium acetate (S5636, Sigma-Aldrich; 1 g per kg) was intraperitoneally injected 10 h and 1 h before mice were euthanized. For serum collection, mice were fasted overnight for 16 h with free access to water. For food reintroduction, mice were fasted for 24 h and then re-fed for another 24 h. Mice were euthanized and the indicated tissues were collected for subsequent analysis.

For the KPC model used in the MRTX1133 therapy experiment, 1 × 10⁶ KPC cells were subcutaneously injected into 6- to 8-week-old NSG mice. The vernier calliper measurements begun when the tumours reached around 200 mm³. Tumour volume measurements were recorded three times per week using the formula 0.5 × length × width². A blinded study design was used in the mouse tumour experiment to prevent bias during data collection and assessment, thus mice were randomized into control and treatment groups, and treated by intraperitoneal injection with vehicle (10% DMSO + 90% (20% SBE-β-CD in saline) or MRTX1133 in vehicle (30 mg per kg, twice a day) when the tumour volume reached around 300 mm³. Tumours were collected after 6 days of treatment. All of the animal experiment procedures, including the maximal tumour volume, were approved by ethics committee of Department of Laboratory Animals, Fudan University.

AAV production and infection in vivo

Plasmids for the AAV2/9 system, including pAAV RC2/9 plasmids, pAAV helper plasmids and transgene plasmids with the CMV or U6 promoter were used for global expression or the knockdown of genes in vivo respectively as previously described⁴⁴. The plasmids were mixed with PEI solution and transfected into HEK293T cells. Then, 60–72 h after transfection, the cells and medium were collected by centrifugation (3,500 rpm, 4 °C, 5 min). 5× polyethylene glycol (40% PEG 8000, 2.5 M NaCl) was added to the supernatant and incubated at 4 °C overnight followed by centrifugation (3,000 rpm, 4 °C, 5 min) to collect the virus pellet. Meanwhile, the cell pellet was resuspended with lysis buffer (150 mM NaCl, 20 mM Tris-Cl, pH 8.0) and lysed by three freeze–thaw cycles between liquid N₂ and a 37 °C water bath followed by centrifugation (5,500 rpm, 4 °C, 10 min) to obtain the supernatant. Then the supernatant was mixed with the virus pellet. The mixture was purified by Optiprep (D1556-250mL, Sigma-Aldrich) gradients (17%, 25%, 40% and 60%) centrifugation (40,000 rpm, 16 °C, 2 h). The viral fraction was collected from the 40% gradient, then washed three times with PBS using 100 kDa columns (3,500 rpm, 4 °C, 30 min).

AAVs were administered to C57BL/6J mice through gastrocnemius injection (5 × 10¹⁰ copies, 25 μl per mouse, three sites) or tail injection (1 × 10¹¹ copies, 150 μl per mouse). All experiments were performed 3–4 weeks after AAV injection. The efficiency of Nlrx1 knockdown or overexpression mediated by AAV delivery was validated by immunoblotting.

Plasmids, reagents and antibodies

Plasmids

WT NLRX1 (HA tag), the NACHT domain (amino acids 160–483, HA tag), LRR domain (amino acids 669–975, Flag tag), ΔLRR (amino acids 1–668, HA tag), 4A (four sites, Glu729, Lys754, Gln758 and Arg958, were mutated to Ala, HA tag) of NLRX1, NLRX1-GFP11 (the C terminus of NLRX1 without stop codon was fused to the linker GGSGGGS and the GFP11 tag RDHMVLHEYVNAAGIT), NLRX1-GFP11-IRES-RFP (the C terminus of NLRX1-GFP11 fused to IRES and RFP), HSP60-GFP11 (the C terminus of HSP60 without stop codon was fused to the linker and the GFP11 tag), GFP11-TOM20 (the GFP11 and the linker fused to the N terminus of TOM20) and HA-NLRP3 were constructed into pcDNA3.1 vector; WT NLRX1-HA, ΔLIR-NLRX1 (amino acid deletion 461–466, HA tag), NLRX1(ΔN-ter) (amino acids 156–975, HA tag), NLRX1(Cyto) (amino acids 87–975, Flag tag), NLRX1(ER) (the N terminus of NLRX1(Cyto) fused to the amino acids 81–250 of FAM134B, Flag tag) were generated as previously described⁶; cytoGFP(1–10), matrixGFP(1–10) (the N terminus of GFP1–10 was fused to the MTS of COX8, residues 1–36) and GFP-Parkin were generated into the pLVX-hygro vector, and GFP-LC3B was generated into pQCXIH. Constructs encoding mt-Keima and MTS-eGFP were generated into the pLVX or pLVX-Tet-On vector.

Guide RNAs targeting human or mouse NLRX1 were designed online (http://www.e-crisp.org/E-CRISP/) and inserted into the pLentiCRISPR v2 vector.

For NLRX1–HA knock-in cell generation, the guide RNA (5′-TCTGGAAGCTGAGACACTGG-3′) was cloned into the pX458 plasmid. The homology arm of NLRX1-800-stop codon-+800 cloned into pcDNA 3.1 with a mutation at the PAM site from CGG to CCG and HA tag (TACCCCTACGACGTCCCCGACTACGCC) sequence was inserted before the stop codon.

Metabolites

AcCoA sodium salt (AcCoA) (A2056), CoASH (C4780), malonyl-CoA (M4263) and sodium acetate (acetate) (S5636) were obtained from Sigma-Aldrich. Succinly-CoA (HY-137808) was obtained from MCE. Biotin was conjugated to the amino groups (-NH2) of AcCoA by EZ-Link sulfo-NHS-LC-biotin (A39257, Thermo Fisher Scientific) according to the manufacturer’s instructions.

Antibodies

Anti-TIM23 (mouse, 611223, BD Biosciences, 1:5,000), anti-MT-CO2 (rabbit, ab79393, Abcam, 1:3,000), anti-CYTB (rabbit, 55090-1-AP, Proteintech, 1:5,000), anti-HSP60 (goat, sc-13115, Santa Cruz Biotechnologies, 1:1,000), anti-LC3 (rabbit, 3868S, CST, 1:1,000), anti-HA (mouse, 901513, BioLegend, 1:1,000), anti-Flag (mouse, F3165, Sigma-Aldrich, 1:3,000), anti-NLRX1 (rabbit, 17215-1-AP, Proteintech, 1:1,000), anti-PINK1 (rabbit, BC100-494SS, Novus Biologicals, 1:1,000), anti-ACLY (rabbit, 15421-1-AP, Proteintech, 1:1,000), anti-ACSS2 (rabbit, 16087-1-AP, Proteintech, 1:1,000), anti-FASN (rabbit, 10624-2-AP, Proteintech, 1:3,000), anti-ACC1 (rabbit, 21923-1-AP, Proteintech, 1:2,000), anti-IDH1 (rabbit, 12332-1-AP, Proteintech, 1:3,000), anti-p62 (rabbit, 18420-1-AP, Proteintech, 1:3,000), anti-acetylated-lysine (rabbit, 9441, CST, 1:1,000), anti-phospho-AMPKα (rabbit, 40H9, CST, 1:1,000), anti-AMPKα (rabbit, 10929-2-AP, Proteintech, 1:3,000), anti-ULK1 (rabbit, 8054, CST, 1:1,000), anti-phosphorylated ULK1 (Ser757) (rabbit, 6888, CST, 1:1,000), anti-phosphorylated ULK1 (Ser555) (D1H4) (rabbit, 5869, CST, 1:1,000), anti-S6 kinase (S6K) (rabbit, 9202, CST, 1:1,000), anti-phospho-S6K (Thr389) (mouse, 9206, CST, 1:1,000), anti-cytochrome c (rabbit, 556432, BD Biosciences, 1:1,000), anti-Parkin (rabbit, Proteintech, 66674-1-Ig, 1:500), anti-CLIMP63 (mouse, ENZ-ABS-669-0100, ENZO, 1:500), anti-FIP200 (rabbit, 17250-1-AP, Proteintech, 1:3,000), anti-ATG7 (rabbit, 10088-2-AP, Proteintech, 1:3,000), anti-tubulin (rabbit, 11224-1-AP, Proteintech, 1:5,000), anti-actin (mouse, 66009-1-Ig, Proteintech, 1:5,000) were used in immunoblotting. Anti-LC3 (rabbit, PM036, MBL, 1:100), anti-HA (mouse, 901513, BioLegend, 1:1,000), anti-TOM20 (mouse, 612278, BD Biosciences, 1:1,000) and anti-HSP60 (goat, sc-13115, Santa Cruz Biotechnologies, 1:1,000) were used in immunofluorescence. The fluorescent secondary antibodies goat anti-mouse Alexa Fluor 594 (A11032, Invitrogen, 1:1,000), donkey anti-rabbit Alexa Fluor 594 (A21207, Invitrogen, 1:1,000), donkey anti-mouse Alexa Fluor 488 (A21202, Invitrogen, 1:1,000) and donkey anti-goat Alexa Fluor 647 (A21447, Invitrogen, 1:1,000) were used in immunofluorescence.

Inhibitors

HC (59847), BTC (51520) and CCCP (C2759) were from Sigma-Aldrich. SB (HY-16450), actinomycin D (HY17559), MRTX1133 (HY-134813), RMC-6236 (HY-148439), Torin-1 (HY-13003) and Mdivi-1 (HY-15886) were from MCE. Bafilomycin A1 (S1413) was from Selleck. Oligomycin (9996) was from CST.

Cell culture and cell line generation

HEK293T, HeLa, A549, MCF7 and U-2 OS cells were purchased from ATCC and AsPC-1 cells were purchased from National Collection of Authenticated Cell Cultures (NCACC), Shanghai. Sf9 cells were purchased from Invitrogen. KPC (Kras^G12D/+Trp53^R172H/+) cells were obtained from Z.-G. Zhang. HEK293T, HeLa, A549, MCF7, U-2 OS, KPC cells and AsPC-1 cells were cultured in DMEM (Invitrogen) or RPMI-1640 (Invitrogen) supplemented with 10% FBS (BI) and 1% penicillin–streptomycin (HyClone). Sf9 cells were cultured in SF900 II SFM (Gibco). The SM is the DMEM formula with 5 mM glucose, 2 mM glutamine, 1 mM pyruvate and supplemented with 10% dialysed serum (BI) and 1% penicillin–streptomycin (HyClone). All cell lines were grown at 37 °C and 5% CO₂ and were tested to be mycoplasma free using the mycoplasma detection kit (40612ES25, YEASEN). A notable exception was the Sf9 cell line, which was maintained under distinct conditions: incubation at 28 °C with shaking on a horizontal shaker at a rotational speed of 100 rpm.

NLRX1-knockout cells were generated using the CRISPR–Cas9 system. pLentiCRISPR v2 vectors carrying sgRNA were mixed in Opti-MEM and transfected into cells with polycation polyethylenimine (PEI) (Sigma-Aldrich) and selected by puromycin for 3 days to get NLRX1-deficient cells. Single cells were seeded into 96-well plates and validated by sequencing and immunoblotting to get NLRX1-knockout cells. To generate NLRX1–HA-tag knock-in cells (HeLa-NLRX1(HA-KI)), the plasmid pX458 together with donor DNA (amplification of plasmid pcDNA3.1 containing the homology arm of NLRX1) with a ratio of 1:1 in Opti-MEM were transfected with Lipo3000 (Invitrogen) into HeLa cells. After 48 h transfection, GFP-positive cells were sorted and seeded into the 96-well plate (single clone per well) by flow cytometry. The knock-in cells were validated by sequencing and immunoblotting.

To generate cells with the inducible expression of mt-Keima, HeLa cells were infected with viruses expressing pLVX-Tet3G-rtTA and selected with G418 (800 μg ml⁻¹) for 1 week to get HeLa-rtTA cells. HeLa-rtTA cells were then infected with viruses expressing mt-Keima, followed by blasticidin (10 μg ml⁻¹) selection for an additional 1 week to generate HeLa-Tet-On-mt-Keima cells. The expression of mt-Keima was induced by doxycycline (1 μg ml⁻¹) for 6 h.

To generate stable cell lines, cells were infected with indicated viruses together with 10 μg ml⁻¹ polybrene. After 48 h, cells were selected with 2 μg ml⁻¹ puromycin, 50 μg ml⁻¹ hygromycin, 10 μg ml⁻¹ blasticidin or 800 μg ml⁻¹ G418 for 1–2 weeks. The overexpression or knockdown efficiency was verified by immunoblotting.

For transient gene overexpression, cells were transfected with indicated plasmids using Lipo3000 (for HeLa cells) or PEI (for HEK293T cells). Gene expression was validated by immunoblotting or immunofluorescence 24–48 h after transfection.

Virus packing

Lentiviral or retroviral vectors carrying the indicated genes, together with packaging plasmids psPAX2 and pMD2.G or VSVG and GAG were transfected into HEK293T cells. After 48–72 h, the supernatants were collected, filtered with a 0.45-μm filter and concentrated with PEG8000.

Gene knockdown by siRNA

siRNAs were transfected by Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer’s instructions and, after 48 h, the transfected cells were treated as indicated and collected for subsequent analysis.

Mitochondria isolation

For mitochondria isolation, cells were washed with cold PBS twice and collected with 1 ml cold mitochondrial lysis buffer as previously described⁴⁵. The buffer containing 210 mM mannitol, 70 mM sucrose, 5 mM Tris–HCl (pH 7.5) and 1 mM EDTA (pH 8.0), was adjusted to pH 7.5 with protease inhibitors. Then the cell suspension was transferred to the Dounce Tissue Grinder (P1110, T2690, Sigma-Aldrich) and lysed by 26 strokes. The homogenate was centrifuged at 1,300g for 10 min at 4 °C and the supernatant was collected in new tubes followed by centrifugation (10,000g, 4 °C, 20 min) to generate the supernatant as the cytosolic fraction and the cell pellet as the mitochondrial fraction.

Immunoblotting, immunoprecipitation and GST-pull-down assay

For immunoblotting, cells were lysed in 1× SDS buffer, boiled at 95 °C for 10 min and analysed by SDS–PAGE. For LC3 analysis, cells were lysed by buffer F (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 10% glycerol, 0.5% NP-40 and protease inhibitors) and centrifuged for 15 min at 4 °C. The supernatants were collected and boiled with 3× SDS and analysed by SDS–PAGE. For biotin–AcCoA pull-down assays, cells were collected and mitochondria were purified as described above. Mitochondria pellets were lysed by buffer C (50 mM HEPES, pH 7.5, 150 mM NaCl, 1% NP-40, 2 mM EDTA and protease inhibitors) and centrifuged for 15 min at 4 °C; the supernatants were then collected for the biotin–AcCoA-binding assay. Streptavidin beads (3419, CST) were incubated with biotin or biotin-labelled AcCoA in PBS for 1 h at room temperature, the beads were washed once with PBS and then incubated with cell lysates overnight at 4 °C with rotation. On the second day, the beads were washed four times with buffer C, boiled with 1× SDS and analysed by SDS–PAGE. For GST–LC3 pull-down assays, GST beads (AGM90049, AOGMA) were incubated with recombinant GST–LC3 for 4 h at 4 °C, then washed with buffer C and incubated with cell lysates at 4 °C for 4 h with rotation. The beads were washed four times with buffer C, boiled with 1× SDS and analysed by SDS–PAGE. For LRR-domain binding with the NACHT domain, cells were lysed with lysis buffer (0.5% Triton X-100, 20 mM HEPES pH 7.6, 150 mM NaCl, 12.5 mM β-glycerophosphate, 1.5 mM MgCl₂, 2 mM EGTA with protease inhibitors). AcCoA was co-added with Flag beads (A2220, Sigma-Aldrich) into the lysates overnight at 4 °C with rotation. The beads were washed and protein samples were processed as described above.

For AMPK activation and mTOR inhibition analysis, HeLa, A549 and MCF7 cells were treated with SM (DMEM containing 5 mM glucose, 2 mM glutamine and 1 mM pyruvate sodium with 10% dialysed serum and 1% penicillin–streptomycin) for 16 h. Oligomycin (5 μM, 5 min) or Torin-1 (100 nM, 16 h) was used as a positive control. Cells were washed with precooled PBS twice and lysed with precooled lysis buffer containing 20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 2.5 mM pyrophosphate, 50 mM NaF, 5 mM β-glycerol phosphate, 50 nM calyculin A, 1 mM Na₃VO₄ and protease inhibitors. The lysates were centrifuged at 17,000g for 10 min at 4 °C, and the supernatant was boiled with 3× SDS and analysed by SDS–PAGE⁴⁶.

For all immunoblotting analyses, proteins were separated by 10–12% SDS–PAGE, transferred to PVDF membranes (Amersham), blocked by 5% non-fat milk in 0.1% Tween-20/PBS buffer for 1 h at room temperature, and immunoblotted by antibodies according to molecular mass. All uncropped raw immunoblotting data are provided in Supplementary Fig. 1.

Purification of recombinant NLRX1 proteins

Human NLRX1 without the N-terminal MTS (amino acids 87–975) was cloned into the pFastBac His vector with an additional N-terminal MBP tag. The vector was then transfected into DH10Bac to get recombinant bacmids, which were further transfected into SF9 insect cells to get amplified baculovirus. SF9 cells were infected with amplified baculovirus for 3 days and cells were collected and lysed in HEPES buffer (20 mM HEPES pH 7.5, 150 mM NaCl) with protease inhibitors and 0.5 mM Tris (2-carboxyethyl) phosphine (TCEP) with sonication. After centrifuging with 10,000 rpm for 1 h, supernatants containing recombinant NLRX1 were purified with Ni-NTA (QIAGEN) and gel-filtration chromatography on the Superdex 200 column (GE Healthcare). The purified recombinant MBP–NLRX1–His was confirmed by immunoblotting using NLRX1 antibody and used for biotin–AcCoA pull-down assay.

NLRX1 LRR domain (629–975) or LRR(4A) were cloned into the pMAL-c5X vector with an N-terminal expressed MBP tag. Constructs were transfected into Escherichia coli BL21 (DE3) cells, which were incubated in LB medium (50 μg ml⁻¹ ampicillin) for 6 h at 37 °C with shaking. Protein expression was induced with 0.2 mM isopropyl-β-D-thiogalactopyranoside overnight at 18 °C. Cells were collected and resuspended in HEPES buffer with protease inhibitors and 0.5 mM TCEP. The proteins were further purified by gel-filtration chromatography on the Superdex 200 column (GE Healthcare Life Sciences) equilibrated with the HEPES buffer with 0.5 mM TCEP. Dextrin beads (SA077025, Smart-Lifesciences) were used to purify recombinant proteins, washed with HEPES buffer and eluted with 5 mM maltose.

[³H]AcCoA binding assay

Recombinant MBP–NLRX1-LRR proteins (50 μg) and equal amounts of MBP or MBP–LRR(4A) proteins were incubated with dextrin beads for 2.5 h at 4 °C with rotation. Beads were washed twice with lysis buffer 2.0 (HEPES buffer with 2 mM MgCl₂, 0.5 mM TCEP and 0.05% Tween-20). The beads were incubated with 2 μM [³H]AcCoA (NET290250UC, Perkin Elmer) and the indicated concentrations of cold AcCoA for 1 h at room temperature. The tubes were flicked every 10 min. The beads were then washed four times with lysis buffer 2.0 and quantified using the TriCarb scintillation counter (PerkinElmer). The binding affinity K_d was calculated as previously described⁴⁷.

Protein oligomerization analysis

Protein oligomerization analysis was conducted as previously described^6,48. Cells were washed twice with PBS and centrifuged at 3,000 rpm for 5 min at 4 °C. After resuspending in PBS, cell pellets were pipetted 24 times with a 22-gauge needle and centrifuged at 13,000 rpm for 1 h at 4 °C followed by gentle sonication in PBS. The samples were divided into two parts—one part was cross-linked with 1 mM glutaraldehyde for 10 min at 16 °C and the other was not cross-linked as inputs. The samples were boiled at 95 °C for 10 min and analysed by immunoblotting after SDS–agarose or SDS–PAGE electrophoresis.

Molecular modelling for AcCoA binding to NLRX1

The NLRX1 structure was from the Protein Data Bank³¹ (PDB: 3UN9). The simulation of AcCoA or CoASH docking to LRR of NLRX1 was performed by Schrödinger Computational Suite, Maestro v.11.5.011, MMshare v.4.1.011, release 2018-1, platform Windows-x64. All structure figures were prepared in Pymol (http://www.pymol.org).

Flow cytometry

Cells were treated as indicated and washed with PBS, collected in DMEM or RPMI1640 and centrifuged at 800g for 5 min. Cells were stained with 100 nM TMRM (T668, Invitrogen), 5 μM MitoSOX (M36008, Invitrogen) or 10 μM CM-H2DCFDA (HY-D0940, MCE) in DMEM or RPMI1640 for 30 min at 37 °C and 5% CO₂. After the incubation, cells were washed twice with PBS and resuspended in DMEM or RPMI1640 followed by flow cytometry.

The split GFP system used for monitoring mitochondrial outer membrane or matrix protein localization has been previously described^27,49,50. In brief, HeLa cells stably expressing cytoGFP(1–10) or matrixGFP(1–10) were transfected with construct encoding NLRX1(G11)-IRES-RFP. After 36–48 h, cells were treated as indicated and flow cytometry was performed by Beckman coulter CytoFLEX S instrument. NLRX1 localization on the mitochondrial outer membrane or matrix was calculated on the basis of the GFP⁺RFP⁺/RFP⁺ ratio. For flow cytometry, 1 × 10⁴ to 2 × 10⁴ cells were collected using the Beckman Courtier instrument, and the data were analysed by FlowJo v.10.8.1 software or CytExpert 2.5.

ATP measurement and cell death assay

For intracellular ATP production measurement, the ATP Assay Kit (S0026, Beyotime) was used according to the manufacturer’s instructions. In brief, cells were washed twice with PBS and lysed with lysis solution for 20 min on ice followed by centrifugation (12,000g, 5 min, 4 °C). ATP assay working solution and the supernatants were co-added into a black 96-well plate. The luminescence was measured using a microplate reader (BioTek).

For the cell death assay, LDH was detected using the CytoTox 96 Non-Radioactive Cytotoxicity Assay kit (G1782, Promega) according to the manufacturer’s instructions. In brief, cells were treated as indicated and the supernatants were transferred to a fresh 96-well plate. An equal volume of LDH detection working solution was added to each well plate and incubated at 37 °C for 30 min. The LDH positive control in the kit was used as the positive control. Finally, the absorbance signal was measured at 490 nm using a microplate reader (BioTek).

Measurement of NADP(H)

For the measurement of NADP(H), the NADP⁺/NADPH Assay Kit (Beyotime, S0180) was used according to the manufacturer’s instructions. In brief, cells were washed twice with PBS, resuspended with extraction buffer and then centrifuged (12,000g, 5 min, 4 °C). The supernatant was divided into two equal parts. One part was used for total NADPH measurement. The other part was incubated for 30 min at 60 °C to decompose NADP⁺. The G6PDH assay working solution was then added to the supernatants in the 96-well plate and the mixture was incubated for 10 min. Finally, the absorbance signal was measured at 450 nm with a microplate reader (BioTek).

2D cell proliferation

For 2D cell proliferation assay, cells were seeded into the black 96-well plate and treated with serial dilutions of MRTX1133 or RMC-6236, Mdivi-1 (20 μM) or DMSO. Then, 50 μl of CellTiter-Glo reagent was added to 50 μl of medium-containing cells and the contents were mixed for 2 min. The plate was next incubated at room temperature for 10 min. The luminescence was measured using a microplate reader (BioTek). Luminescence signal was normalized to DMSO treated cells (percentage DMSO = (lum_treated/mean(lum_DMSO)) × 100). A log[inhibitor] versus response-variable slope (four parameters) model was used to calculate the IC₅₀.

Metabolite extraction and GC–MS

For intracellular metabolite measurements in Extended Data Fig. 1n, cells were cultured in 10 cm dishes and treated with SM for 16 h. When cell confluency was about 80–100%, the medium was removed, cells were washed with cold PBS twice, collected in extraction buffer (acetonitrile:isopropanol:water, 3:3:2, v/v/v). The resuspended cells were placed in liquid N₂ for 5 min and thawed on ice for 5 min and the freeze–thaw cycle was repeated four times and the samples were then centrifuged (12,000 rpm, 4 °C, 10 min) to collect the supernatants. For serum valine, leucine and isoleucine measurements in Extended Data Fig. 1c, blood was collected from the eyes and clotted at room temperature for 1 h and then centrifuged (3,000 rpm, 10 min). Next, 20 μl serum was diluted with 80 μl precooled methanol and vortexed, then centrifuged at 12,000g for 15 min to remove proteins. The cell or serum supernatants were then evaporated by freeze-vacuum and analysed by gas chromatography coupled with MS (GC–MS). The pellets were resuspended in 75 μl acetonitrile at 60 °C for 15 min, then oximated by MTBSTFA (N-tert-butyldimethylsilyl-N-methyltrifluoroacetamide) in 50 μl pyridine and further incubated in 60 °C for 1 h. The samples were centrifuged and the supernatants were transferred to glass vials. Then, 1 µl of each sample was injected and analysed on the Agilent 7890B-5977B GC–MS system with DB-5MS (0.25 mm internal diameter, 0.25 μm film, with 30 m empty column, Agilent J&W). Metabolite m/z ratios were compared with those of previous studies⁵¹. Each metabolite was quantified by the retention time and peak area by MassHunter Workstation (Agilent). The final intracellular metabolites and amino acid level was normalized to actin level of immunoblotting or volume of serum.

Metabolite extraction and LC–MS

Cytosolic and mitochondrial AcCoA levels were determined as previously described⁴⁵. In brief, approximately 1 × 10⁷ to 2 × 10⁷ cells were lysed with 1 ml cold mitochondrial lysis buffer and the cytosolic and mitochondria fractions were then isolated as described above. The cytosolic fraction was quenched by 50% (w/v) trichloroacetic acid in water (the final concentration of trichloroacetic acid is 10%) and the mitochondria fraction was resuspended in 1 ml 10% (w/v) trichloroacetic acid. The mitochondria fraction was placed into liquid N₂ for 5 min and thawed on ice for 5 min and the freeze–thaw cycle was repeated four times followed by centrifugation (17,000g, 4 °C, 10 min). The mitochondria supernatants and cytosolic fractions were then purified using Oasis HLB 1cc (30 mg) SPE columns (Waters). Columns were washed with methanol, equilibrated with water, loaded with the cytosolic and mitochondrial fractions, washed with water and eluted with elution buffer (25 mM ammonium acetate in methanol). The elutions were evaporated by freeze-vacuum, resuspended in 20 μl 20% acetonitrile in water and analysed by LC–MS. For in vivo AcCoA measurement, tissues were obtained from C57BL/6J mice, and 50–100 mg tissue was collected in 1 ml precooled 10% trichloroacetic acid. The samples were then homogenized and lysed on a rotating shaker (30 min, 4 °C) followed by centrifugation (12,000g, 10 min, 4 °C). The supernatant was purified, dried and resuspended as described above, and analysed using LC–MS. Then, 5 µl of each sample was injected and analysed using SHIMADZU LC-30AB LC system coupled to the QTRAP7500 Mass Spectrometer (SCIEX). Hydrophilic interaction chromatography (HILIC) with the BEH column (1.7 µm, 2.1 mm × 100 mm; Waters) was used. Mobile phase A was as follows: ammonia with 10 mM ammonium formate and 0.2% ammonia. Mobile phase B was acetonitrile. The flow rate was 0.2 ml min⁻¹ and the column temperature was set at 40 °C. Linear gradient: 0 min, 80% B; 3 min, 50% B; 10 min, 50% B; 10.1 min, 80% B; 15 min, 80% B. Multiple reaction monitoring (MRM) technology using MS/MS was used for specific detection of AcCoA. The LC–MS system was operated in negative ionization mode. The source parameters included curtain gas (CUR) at 40 psi; collision active dissociation (CAD) gas at 6; ion source gas 1 (GS1) at 40 psi, GS2 at 70 psi; ion spray voltage (IS) at 4,500 V; ion source temperature (TEM) at 450 °C. The specific transition was recorded as follows: AcCoA 808.0945 > 407.9000. The final AcCoA level was normalized to the actin immunoblot level, or the tissue weight or cell number, as described in the figure legends.

For serum amino acid measurement (proline, glutamate acid, serine, asparagine, glutamine, arginine, glycine, alanine, aspartic acid, tyrosine, histidine, lysine, methionine, phenylalanine, threonine and tryptophan, while cysteine was too low to be detected), blood was collected from eyes and clotted at room temperature for 1 h followed by centrifugation (3,000 rpm,10 min). Then, 20 μl serum was diluted with 80 μl precooled methanol and vortexed, then centrifuged at 12,000g for 15 min to remove proteins. The supernatants were evaporated by freeze-vacuum, resuspended in 100 μl 80% methanol in water and analysed by LC–MS. Then, 1 µl of each sample was injected and analysed using UPLC-H Class LC system (Waters) coupled to the 6500 QTRAP Mass Spectrograph (SCIES). The ultimate AQ-C18 column (5 µm, 2.1 mm × 250 mm; Welch) was used at room temperature. Mobile phase A was as follows: water (0.1% formic acid, v/v); and mobile phase B was acetonitrile (0.1% formic acid, v/v). Linear gradient: 0–1 min, 0% B; 1–14 min, 0–90% B; 14–16 min, 90% B; 16–16.1 min, 90–0% B; 16.1–20 min, 0% B. The flow rate was 0.2 ml min⁻¹. MRM technology using MS/MS was used for specific detection of various amino acids. The LC–MS system was operated in positive ionization mode. The source parameters included CUR at 40 psi; CAD at medium; GS1 at 40 psi, GS2 at 40 psi; IS at 5,500 V; and TEM at 500 °C. All LC–MS analysis was performed by the Metabolic Platform at the Fudan University. Data were analysed using Skyline (22.2.0.527) software to calculate the peak area values.

Seahorse analysis

The mitochondrial oxygen consumption rate (OCR) in HeLa cells was measured with the Seahorse XFe96 equipment (Agilent) using the Cell Mitochondrial Stress Test kit (103015-100, Agilent) according to the manufacturer’s instructions. In brief, 0.8 × 10⁴ cells were seeded onto an XFe96 cell culture microplate (Agilent) per well and treated with CCCP (10 μM), SB (100 μM), BTC (5 mM), HC (20 mM) and SM for 16 h. Before analysis, cells were washed twice and equilibrated with XF DMEM in a 37 °C incubator without CO₂ for 1 h. Oligomycin (final concentration: 1.5 μM), FCCP (final concentration: 2 μM), and rotenone/antimycin A (final concentration: 0.5 μM) were used in OCR analysis. Data were analysed by Seahorse Wave Desktop Software (Agilent).

CRISPR screening

To generate lentivirus for screening, we used the genome-wide GeCKO v2.0 Human library⁵² in the lentiCRISPR v2 vector (Addgene, 1000000048), which contains six sgRNAs per gene (123,411 sgRNAs targeting 19,050 genes). A total of 1 × 10⁸ HEK293T cells was seeded into six T225 flasks. Each flask was transfected with 20 μg of plasmid library, 10 μg of psPAX2 and 5 μg of pMD2.G using PEI. After 48 h of transfection, the supernatant was collected, centrifuged at 3,000 rpm for 5 min, filtered through a 0.45-μm filter and stored at −80 °C. Virus titres were determined using puromycin selection. The lentiviral library was used to infect HeLa cells stably transfected with doxycycline-inducible mt-Keima reporter at a multiplicity of infection of approximately 0.3 with 8 μg ml⁻¹ polybrene. Then, 48 h after infection, 1 μg ml⁻¹ puromycin was added to the cells and selected for 5 days, followed by an additional 2 days of expansion in puromycin-free medium. During selection, cells were maintained at >500 cells per sgRNA.

To induce mitophagy, cells were first treated with 1 μg ml⁻¹ doxycycline overnight to induce mt-Keima expression, followed by 20 mM HC treatment for 16 h. Treated cells were trypsinized, filtered through a 40-μm cell strainer and resuspended in PBS containing 2% FBS. Cell sorting was performed using a BD FACSAria II instrument with two channels: 405 nm excitation for mt-Keima at pH 7 and 562 nm excitation for mt-Keima at pH 4, with a 610 nm emission bandwidth⁵³. The top 25–30% and bottom 25–30% cells were sorted to represent mitophagy-enhanced and -inhibited cell subsets, respectively. A total of 10⁷ cells was sorted for each group in two biological replicates for subsequent sequencing.

Genomic DNA was extracted from both the mitophagy-enhanced and -inhibited groups. Sequencing libraries were prepared by two rounds of PCR to amplify target DNA fragments, followed by the ligation of index and adapter sequences. The prepared libraries were then subjected to paired-end sequencing (2 × 150 bp) on the Illumina NovaSeq 6000 platform. After sequencing, 20 bp gRNA sequences were extracted and aligned to the GeCKO v2.0 library reference sequence. The alignment results for all library sequences were counted to obtain the number of matched reads. Sequencing depth and coverage were calculated to assess data reliability and accuracy. Two rounds of screening data were analysed using MAGeCK software⁵⁴ with the default settings, and the results were ranked by negative log₂-transformed fold change and P value. Mitochondrial genes were defined using the mitoCarta3.0 database²². Volcano plots for genome-wide and mitochondria-targeted analyses of the CRISPR screening were generated using the R package ggplot2. GO biological processes enriched in the top 100 mitochondrial genes from the screen were analysed using the Metascape database⁵⁵.

qPCR analysis

To quantify the mtDNA/nDNA ratio, genomic DNA was isolated from cells or tissues using the TIANamp Genomic DNA Kit (DP304-02, TIANGEN) according to the the manufacturer’s instructions, and qPCR was conducted to amplify the mitochondria genome (MT-CYTB, MT-CO1, MT-ATP6 in human; mt-Cytb, mt-Co1, mt-Atp6 in mouse) or nuclear genome (RPL13A in human; Rpl13a in mouse) separately as previously described⁶. Total RNAs were extracted using the RNA Easy Fast Tissue/Cell Kit (TIANGEN, DP451) and reverse transcribed using the PrimeScript RT Reagent kit (TaKaRa, RR047A) according to the manufacturer’s instructions. The qPCR was performed on the ABI QuantStudio 7 Flex system using the TB Green Premix ExTaq kit (TaKaRa, RR820A). The relative fold changes were calculated using $2^{-\mathrm{\Delta }\mathrm{\Delta }{C}_{\mathrm{t}}}$ method. qPCR primer sequences are provided in Supplement Table 3.

Immunofluorescence and confocal microscopy

For colocalization analysis of exogenous NLRX1 with exogenous LC3, HeLa cells stably expressing NLRX1 with a C-terminal HA tag (HeLa-NLRX1-WT) were cultured on glass coverslips and transfected with the vector expressing GFP–LC3. For colocalization analysis of endogenous NLRX1 with endogenous LC3, HeLa cells with HA tag knock-in (HeLa-NLRX1(HA-KI)) were used. For colocalization analysis of GFP–Parkin with mitochondria, HeLa cells stably expressing GFP–Parkin were generated (HeLa-GFP–Parkin). Cells were treated as indicated in the figure legends and washed with PBS twice, fixed with 4% PFA for 10 min, permeabilized with 0.1% Triton X-100 in PBS for 10 min, blocked with 5% BSA in PBS for 1 h at room temperature and incubated with primary antibodies (diluted in 5% BSA) at 4 °C overnight. The next day, cells were washed three times with PBS and incubated with fluorescent secondary antibodies for 2 h at room temperature and then washed three times with PBS. After incubation with DAPI for 5 min and mounted with antifade reagent, samples were observed with the ×60 oil objective of confocal microscopy (Leika SP5 and Olympys FV3000).

For analysis of mitochondrial protein import, HeLa-rtTA cells were transfected with vectors expressing Tet-On-MTS-EGFP. After 6 h of transfection, 0.25 μg ml⁻¹ doxycycline was added to induce MTS-EGFP expression. Then cells were stained with 50 nM Mitotracker Deep Red FM (M22426, Thermo Fisher Scientific) for 20 min and washed with PBS twice, and fresh medium was added for living cell imaging using Olympys confocal microscope (FV3000) with a ×60 oil objective in two channels: 488 nm excitation and 520 nm emission for eGFP and 640 nm excitation and 685 nm emission for Mitotracker Deep Red FM. Three replicates with a total of 100 cells per condition were analysed^29,56.

The split GFP system used for monitoring mitochondrial outer membrane or matrix protein localization was described above. In brief, HeLa cells stably expressing cytoGFP(1–10) or matrixGFP(1–10) were transfected with constructs encoding NLRX1-GFP11, HSP60-GFP11 or GFP11-TOM20 vectors. After 24 h, living cells were observed using Olympus confocal microscope (FV3000) with a ×60 oil objective in the channel with 488 nm excitation and 520 nm emission for GFP. Images were processed by deconvolution using OLYMPUS CellSens Dimension Desktop (v.4.1.1) to improve resolution, remove background fluorescence and recover the real distribution.

Mitophagy reporter assay

For measuring in vivo mitophagy, 6–8-week-old mice were given intramuscular injection of AAV-mt-Keima (3 × 10¹¹ copies, 25 μl per mouse, three sites) or intravenously (1 × 10¹¹ copies, 150 μl per mouse). After 4–6 weeks, mice were fasted for 24 h or given intraperitoneal injection of HC for 4 h, and gastrocnemius or liver tissues were collected. The samples were cut into sections with a thickness of 6 μm and observed using Olympus confocal microscope (FV3000) with a ×60 oil objective in two channels: 445 nm excitation for mt-Keima pH 7 and 561 nm excitation for mt-Keima pH4 with a 570–695 nm emission bandwidth. For mitophagy index quantification, total mitochondrial area and mitolysosome area were calculated by green fluorescence area and red-only puncta area gated by a fixed threshold individually using Image J software. Mitophagy level was quantified by the ratio of mitolysosome area/mitochondrial area.

For measuring in vitro mitophagy, cells were treated as indicated, and the flow cytometry was performed by Beckman coulter CytoFLEX S instrument in two channels (BV605 and PE-Texas Red channels) through two lasers (405 nm and 562 nm) and emission at 610 nm. The gating strategy was provided in Supplementary Fig. 2. Data were analysed using CytExpert 2.5. The ratio of the mitophagic percentage was calculated and the quantification was pooled from three independent biological replicates.

Cytochrome c release analysis

HeLa cells were treated as indicated, washed with cold PBS and lysed with digitonin lysis buffer (150 mM NaCl, 50 mM HEPES pH 7.4, 25 μg ml⁻¹ digitonin with protease inhibitors) for 10 min on ice. The lysates were then centrifuged at 2,000g for 10 min at 4 °C. The supernatants were centrifuged at 20,000g at 4 °C twice, and the final supernatant was the cytosolic fraction for the detection of cytosolic cytochrome c^57,58.

Tissue mitochondrial isolation

Cytosolic and mitochondrial fraction isolation was determined as previously described²⁰. In brief, for mitochondria isolation from gastrocnemius tissue, mice were killed and muscles were removed, washed with cold PBS supplemented with 10 mM EDTA and minced. The muscles were incubated with PBS supplemented with 10 mM EDTA and 0.05% trypsin for 30 min followed by centrifugation (200g, 10 min, 4 °C). The pellet was resuspended with IBm1 (67 mM sucrose, 50 mM Tris/HCl, 50 mM KCl, 10 mM EDTA and 0.2% BSA, pH adjusted to 7.4) and transferred to the Dounce Tissue Grinder and lysed by 15 strokes followed by centrifugation (700g, 10 min, 4 °C). The supernatant was then centrifuged at 8,000g for 10 min at 4 °C. The supernatant is the cytosolic fraction. Then pellet was resuspended with IBm2 (0.25 M sucrose, 3 mM EGTA/Tris and 10 mM Tris/HCl, pH adjusted to 7.4) followed by centrifugation (8,000g, 10 min, 4 °C) to obtain the mitochondrial pellet.

For mitochondria isolation from liver, mice were killed and livers were collected, washed with cold IBc (10 mM Tris–MOPS, 1 mM EGTA/Tris, 0.2 M sucrose, pH adjusted to 7.4) and minced. Then the liver suspension was transferred to the Dounce Tissue Grinder and lysed by ten strokes, followed by centrifugation (600g, 10 min, 4 °C). The supernatant was then centrifuged at 7,000g for 10 min at 4 °C. The supernatant was the cytosolic fraction. The pellet was then resuspended with IBc followed by centrifugation (7,000g, 10 min, 4 °C) to obtain the mitochondrial pellet. For LC–MS, both the cytosolic and mitochondrial fractions were performed as described above. For immunoblotting analysis, both the cytosolic and mitochondrial fractions were boiled with 3× SDS and analysed by SDS–PAGE.

Electron microscopy

HeLa cells were treated with indicated conditions, transformed into suspension cells using a cell shovel and centrifuged at 2,000 rpm for 10 min. The pellets were fixed in 2.5% glutaraldehyde for 1 h at room temperature and then overnight at 4 °C. The next day, after washing three times with 0.1 M PBS, the pellets were fixed with 1% osmic acid at room temperature for 1 h, washed three times with double-distilled H₂O, dehydrated in a graded ethanol series, slowly infiltrated with 100% acetone and 50% acetone (acrylic resin: acetone,1:1, v/v) for 2 h and embedded in acrylic resin at 60 °C for 48 h. The embedded samples were cut into sections with a thickness of 70 nm, and the sections were stained with 2% uranyl acetate at room temperature for 10–20 min and lead citrate stain the sections for 5 min. Lastly, samples were observed by electron microscope and images were captured by FEI Tecnai G2 spirit electron microscope.

Thermal shift assay

HeLa cells were lysed using lysis buffer and centrifuged at 12,000 rpm for 10 min at 4 °C. PBS, AcCoA (500 µM) or CoASH (500 µM) were added to cell lysates, heated to graded temperatures (44.6–65 °C, 3 min) and centrifuged at 12,000 rpm for 10 min at 4 °C. Soluble proteins were extracted and analysed by immunoblotting.

Statistical analysis

All data were analysed using GraphPad Prism (v.8.3.0) or Excel, with n ≥ 3 biological replicates unless otherwise specified. Data are presented as mean ± s.e.m. or mean ± s.d. as indicated. For the CRISPR screen data in Fig. 1e and Extended Data Fig. 5k, n = 2 biological replicates. All statistical tests were two-tailed. Statistical parameters, including scale bars and statistical significance, are shown in the figures and the figure legends. Two-group comparisons were analysed using unpaired t-tests. Multiple comparisons among more than two groups were performed using one-way ANOVA. A two-way ANOVA was used when two categorical variables were analysed. Post hoc analysis was conducted to identify specific group differences following ANOVA. P < 0.05 was considered to be statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-025-09745-x.

Author contributions

Y.Z. and X.S. designed the experiments. Y.Z., X.S., Y.S. and C.W. co-performed experiments, co-analysed data and co-wrote the manuscript. J.H. and S.C. performed RT–qPCR analysis. C.Y. performed part of experiments. L.Q. and M.M. performed protein purification. S.H. and W.W. helped with molecular docking and binding sites prediction. M.Y. provided intellectual discussion and co-wrote the manuscript. Q.-Y.L. conceptualized, supervised the study, designed experiments, analysed data and co-wrote the manuscript.

Peer review

Peer review information

Nature thanks Kivanc Birsoy, Luca Scorrano and Haitao Wen for their contribution to the peer review of this work. Peer reviewer reports are available.

Data availability

All data supporting the findings of this study are available in the Article and its Supplementary Information. The CRISPR screen data have been deposited in the Genome Sequence Archive in National Genomics Data Center, China National Center for Bioinformation (Beijing Institute of Genomics) of the Chinese Academy of Sciences (GSA-Human: HRA013003). The metabolomic data are available at the Metabolomics Workbench under study ID ST004160. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41586-025-09745-x.

Supplementary information

The online version contains supplementary material available at 10.1038/s41586-025-09745-x.