Inflammation causes remodeling of mitochondrial cytochrome c oxidase mediated by the bifunctional gene C15orf48

Exchange of mitochondrial subunits C15ORF48 and NDUFA4 occurs during macrophage activation in vitro and in inflammatory disease.


INTRODUCTION
Changes in activation state of immune cells are accompanied by, and dependent on, profound alterations of cellular metabolism (1,2). Mitochondria are versatile organelles that play central roles in energy metabolism and act as signaling hubs. The ability of mitochondria to shape cellular phenotype has become a key concept in the field of immunometabolism (3). An example of mitochondrial plasticity is the reprogramming of myeloid cells in response to activation or polarization cues. Stimulation of murine macrophages and dendritic cells with the pathogen-associated molecular pattern lipopolysaccharide (LPS) results in a Warburg-like metabolic state involving rapid up-regulation of glycolytic metabolism and lactate production. Mitochondria in these cells are described as being repurposed for the generation of reactive oxygen species (ROS), which have both signaling and direct bactericidal roles, as well as for the diversion of tricarboxylic acid (TCA) cycle metabolites into biosynthetic and immunoregulatory pathways (4)(5)(6). However, there is growing evidence that metabolic responses to proinflammatory stimuli differ between human and mouse myeloid cells, suggesting that conclusions based on mouse cells should be extrapolated with caution (7). It has been demonstrated that metabolic pathways and mitochondrial activity contribute to autoimmune and chronic inflammatory pathologies such as rheumatoid arthritis (RA), underpinning the concept that these conditions might be effectively treated by targeting cellular metabolism (8)(9)(10).
Mitochondrial adaptation occurs on several different scales (11)(12)(13)(14), and disruption of these processes is a common hallmark of many inflammatory, neurodegenerative, oncological, and age-related diseases (15). Mitochondrial regulation at a gross scale includes the balancing of biogenesis and mitophagy to control mitochondrial number and quality, the opposing processes of fusion and fission, and architectural organization of mitochondrial membranes and formation of cristae (16,17). The complexes of the mitochondrial electron transport chain associate with each other forming supercomplexes, the assembly and turnover of which may represent an additional level at which mitochondrial activity is regulated (18,19). At a smaller scale, mitochondrial function may be fine-tuned by remodeling of individual electron transport chain complexes (20). This is particularly evident in the case of complex IV [the cytochrome c oxidase (CcO) complex], a multiprotein complex that transfers electrons from cytochrome c to molecular oxygen as the final and rate-limiting step of the electron transport chain (21). This complex is the only one within the electron transport chain that contains tissuespecific isoforms of several subunits (20). The hypoxia-responsive transcription factor hypoxia-inducible factor-1 (HIF-1) mediates interchange of CcO subunits in response to changing availability of oxygen (20,(22)(23)(24).
NDUFA4 [reduced form of nicotinamide adenine dinucleotide (NADH) dehydrogenase (ubiquinone) 1 subcomplex subunit 4] was originally identified as a subunit of electron transport chain complex I. More recent evidence indicates that NDUFA4 is a component of the CcO complex and is essential for maximal CcO enzymatic activity in muscle and HeLa cells (25)(26)(27)(28). Very rare homozygous loss-offunction mutations of the NDUFA4 gene causes a severe neuromuscular condition resembling Leigh syndrome, characterized by impairment of CcO activity in muscle, lactic acidosis, dystonia, ataxia, and spasticity (26). A recent crystallographic structure of human CcO, isolated using the mild detergent digitonin, showed NDUFA4 to be located at the periphery of the complex, as a transmembrane protein, making contacts with the mitochondrial membrane lipid cardiolipin (28). NDUFA4 is thought to be added late in the biogenesis of the CcO complex, its peripheral site accounting for its facile loss under certain conditions of mitochondrial isolation (18,29). Renaming it as COXFA4 (cytochrome c oxidase subunit FA4) has been proposed but not yet formally adopted (27). A recently identified form of electron transport chain subunit exchange may occur between NDUFA4 and the related protein NDUFA4L2. Increased NDUFA4L2 expression and reciprocal down-regulation of NDUFA4 occur under hypoxia and are associated with poor prognosis in patients with cancer (30)(31)(32). NDUFA4L2 expression has been shown to aid cell survival and proliferation by reducing oxidative stress (30,(32)(33)(34). However, it has not yet been formally shown whether NDUFA4L2 replaces NDUFA4 in CcO.
The C15orf48 gene is otherwise known as NMES1 in human and as Nmes1 or AA467197 in mouse. Until relatively recently, the main insight into the role of this gene came from the observations that the gene locus is highly methylated in epithelial cell cancers, resulting in markedly reduced transcript expression (35,36). The gene encodes a short protein product with poorly described function. In addition, the C15orf48/NMES1 gene encodes the microRNA (miRNA) miR-147b (miR-147 in mouse), which is up-regulated in response to Toll-like receptor (TLR) activation of macrophages and has been reported to indirectly down-regulate expression of inflammatory cytokines interleukin-6 (IL-6) and tumor necrosis factor- (TNF) (37). More recently, the C15ORF48 protein has been described to be mitochondrially localized and was found to interact with subunits of complexes I, III, and IV (38).
Although much is known about metabolic reprogramming in activated myeloid cells (2,5), the process of mitochondrial electron transport chain component switching is poorly understood in this context. Here, we describe an elegant and economical mechanism by which a single bifunctional transcript causes substitution of the CcO component NDUFA4 by the related protein C15ORF48. This substitution is a conserved response to many different inflammatory challenges both in vitro and in vivo. Expression of C15orf48 is elevated in peripheral monocytes and synovial macrophages of patients with RA. Its expression forms part of a coherent metabolic reprograming, correlates positively with disease severity, and declines in remission. In severe coronavirus disease 2019 (COVID- 19), C15orf48 is highly expressed by pathogenic populations of macrophages that invade the airways. We also show that macrophages from individuals lacking NDUFA4 display a hyperinflammatory phenotype, characterized by elevated chemokine production, linking this regulatory axis to potentially pathogenic mechanisms during inflammatory disease.

RESULTS
The highly conserved gene C15orf48 and its miRNA product are induced during macrophage stimulation The C15orf48 transcript, transcribed from chromosome 15 in humans, encodes a small protein of 83 amino acids. It is also the source of miRNA-147b-3p (hereafter referred to as miR-147b), which is produced from the 3′ untranslated region (3′UTR) encoded by the fourth exon. Both the open reading frame and miRNA source sequence are very highly conserved (Fig. 1A). For example, the latter is identical in human, cow, hen, and green anole lizard C15orf48 transcripts. In the orthologous mRNAs of coelacanth, a lobe-finned fish, and six randomly selected ray-finned fish, we found a maximum of two nucleotide differences within the miR-147b source sequence, none within the seed sequence that is the principal determinant of target specificity. This suggests that the miRNA has been subject to strong selective pressure for more than 450 million years. Expression of mature C15orf48 mRNA and mature miR-147b was increased by LPS treatment of primary human monocyte-derived macrophages (MDMs) (Fig. 1, B and D). Both the mRNA and its miRNA product were even more strongly induced by LPS in mouse bone marrow-derived macrophages (BMDMs) (Fig. 1, C and E), consistent with a previous report (37). According to publicly available microarray datasets, C15orf48 was among the most highly up-regulated genes in several inflammatory pathologies including sepsis, intracerebral hemorrhage, psoriasis, and RA ( Table 1).
Analysis of publicly available expression data highlighted polyinosinic:polycytidylic acid [poly(I:C)] and type 1 interferons (IFNs) as strong inducers of C15orf48 expression in murine macrophages. However, these responses were notably reduced or absent in human MDMs (table S1). We confirmed that IFN- is a potent inducer of C15orf48 in mouse BMDMs (Fig. 1G). Janus kinase-signal transducer and activator of transcription (JAK-STAT) signaling was necessary for LPS-induced expression in mouse BMDMs as the response could be blocked by the Janus kinase inhibitor ruxolitinib (Rux) (Fig. 1I), suggesting that secondary signaling through the type 1 IFN receptor controls C15orf48 expression in these cells. Consistent with this hypothesis, the expression of C15orf48 in response to lipid A (a purified TLR4 agonist) was strongly impaired by disruption of mouse Ifnar1, Trif, or Irf3 genes, all of which play essential roles in IFN- signaling (C15orf48 is identified by its alternative name AA467197 in this paper) (39). In contrast, human C15orf48 expression was not induced by IFN- (Fig. 1F), and its LPS-induced expression was insensitive to Rux (Fig. 1H). These results indicate that while the sequences of this gene and its miRNA are highly conserved, the signaling pathways controlling their expression are divergent.

miR-147b targets electron transport chain component NDUFA4
Four online algorithms were used to predict mRNA targets of miR-147b: TargetScan7.1, miRTarBase, miRDB, and miRanda. NDUFA4 was the only target predicted by all four algorithms ( Fig. 2A) and was the top scoring target in each case. Interaction between miR-147b and murine Ndufa4 mRNA was conserved, although with some differences of base pairing outside of the seed sequence (Fig. 2B), and was also predicted by the algorithms listed above. LPS treatment of MDMs caused sustained down-regulation of NDUFA4 mRNA (Fig. 2C). A miR-147b mimic was transfected into primary human MDMs, with careful titration to achieve a level similar to that in LPS-treated MDMs to minimize off-target effects (Fig. 2D). The miR-147b mimic significantly reduced the expression of both NDUFA4 mRNA and NDUFA4 protein (Fig. 2, E and F). A luciferase reporter containing the NDUFA4 3′UTR was down-regulated by miR-147b mimic cotransfection but not by a negative control miRNA. Mutation of the putative seed binding sequence rendered the NDUFA4 reporter nonresponsive to the miR-147b mimic (Fig. 2G). These results confirm a conserved, seed-dependent interaction between miR-147b and NDUFA4 mRNA.

C15ORF48 protein replaces NDUFA4 within CcO under inflammatory conditions
In the UniProt database, C15ORF48 protein is annotated as a relative of NDUFA4. Amino acid identity between C15ORF48 and NDUFA4 is 26%, but the two proteins display high conservation of amino acid properties (fig. S1, A and B) and matching predicted -helical transmembrane regions (Fig. 3A). The topology of NDUFA4 within the CcO complex is known from a published cryoelectron microscopy (cryo-EM) structure (28). In this three-dimensional model, NDUFA4 could be substituted by C15ORF48 without any conflicts (Fig. 3B). Predicted interactions between NDUFA4 and cardiolipin within the inner mitochondrial membrane (IMM) were partially conserved. In the hypothetical structure, the regions of sequence divergence between the two proteins projected outward from the CcO complex on the matrix side of the IMM (Fig. 3B). A systematic analysis of protein-protein interactions within mitochondria (38) identified four structural components of the CcO complex (COX5A, COX6A1, COX7C, and COX8A) that interact with both NDUFA4 and C15ORF48 but none that interact uniquely with either NDUFA4 or C15ORF48. No interactions between NDUFA4 and C15ORF48 were detected, suggesting that occupancy of CcO by the two proteins is mutually exclusive.
In human MDMs stimulated with LPS, there was notable inverse regulation of NDUFA4 and C15ORF48 protein levels over a 36-hour time course (Fig. 3C). Several other TLR ligands also increased C15ORF48 and decreased NDUFA4 protein levels (Fig. 3D). Notable exceptions were poly(I:C) and ODN2006, which activate the endosomally localized receptors TLR3 and TLR9, respectively. This result is consistent with the transcriptomic analysis that showed little or no response of C15orf48 or NDUFA4 genes to poly(I:C) in human macrophages (table S1). Imiquimod and single-stranded RNA caused relatively weak up-regulation of C15ORF48 protein and partial loss of NDUFA4 protein. In mouse BMDMs, LPS also caused inverse regulation of NDUFA4 and C15ORF48 protein levels, although the loss of NDUFA4 was less marked than observed in human primary macrophages (Fig. 3E). Inspection of publicly accessible macrophage gene expression datasets revealed that C15orf48 expression was commonly increased in both mouse and human primary macrophages by proinflammatory stimuli such as purified TLR agonists or exposure to pathogens (Fig. 3F). The response of the C15orf48 gene to these different stimuli was consistently ranked above the 90th centile in terms of both magnitude and statistical significance (table S1). In the same datasets, the expression of Ndufa4/ NDUFA4 mRNA was generally decreased, although with smaller fold changes (Fig. 3F). Changes in the expression of C15orf48 and Ndufa4/NDUFA4 mRNA were significantly negatively correlated across these varied experimental conditions (P = 3.42 × 10 −6 ), suggesting a mechanistic link and a conserved response to several pathogen-related challenges.
conditions, Western blotting confirmed the presence of NDUFA4 in all CcO-containing structures (based on well-established migration patterns) (Fig. 3H). Specifically, NDUFA4 was present in CcO (complex IV) monomers, dimers, III 2 + IV supercomplexes, and higher-order I + III 2 + IV supercomplexes. Consistent with the expression data, the overall levels of NDUFA4 protein and amounts within each CcO-containing species were decreased following LPS treatment (Fig. 3H). C15ORF48 was detected in all of the same electron transport chain complexes and supercomplexes but showed the reverse pattern of expression, being increased in response to LPS (Fig. 3H). No band corresponding to the size of "free" complex I could be detected with either anti-NDUFA4 or anti-C15ORF48 antibodies under resting or LPS-stimulated conditions. LPS-activated mouse macrophages enter a pseudo-hypoxic state, in which HIF-1 becomes activated despite the presence of adequate oxygen (40,41). In some solid tumors, hypoxia promotes down-regulation of NDUFA4 and reciprocal up-regulation of NDUFA4L2, at least partly mediated by HIF-1 (30,32,34,42). We therefore asked whether the expression of NDUFA4L2 was influenced by LPS treatment of MDMs. We were unable to detect NDUFA4L2 mRNA in primary human macrophages activated by LPS, subjected to hypoxia or treated with the hypoxia mimetic dimethyloxallylglycine ( fig. S2A), despite the fact that all of these treatments caused robust activation of HIF-1 ( fig. S2B). Hypoxiatreated HeLa cells robustly increased NDUFA4L2 mRNA as previously reported (34), providing a positive control for this experiment ( fig.  S2A). In primary human macrophages, hypoxia partially decreased NDUFA4 protein levels, but this response occurred in the absence of any detectable change in expression of C15ORF48 protein ( fig. S2C). Therefore, the substitution of NDUFA4 by C15ORF48 in macrophages is a specific response to proinflammatory stimuli and not a consequence of the pseudo-hypoxic state.
An additional, miR-147b-independent mechanism of down-regulation of NDUFA4 protein To test whether miR-147b is essential for the down-regulation of NDUFA4, we transfected MDMs with a miR-147b-targeting antagomir or nontargeting control, then treated with LPS, and harvested for both mRNA and protein. The miR-147b antagonist prevented the LPS-induced down-regulation of NDUFA4 mRNA (Fig. 4A) but did not protect NDUFA4 protein (Fig. 4B). We conclude that elevation of miR-147b is sufficient for down-regulation of NDUFA4 protein in transfected MDMs but not necessary for NDUFA4 down-regulation in the context of an LPS stimulus. In addition, note that LPS decreased NDUFA4 mRNA levels by approximately 50% (Fig. 2C) but decreased NDUFA4 protein levels by greater than 90% (Figs. 3, C and D, and 4, C to F). This suggests that at least one other mechanism may regulate NDUFA4 protein expression at a translational or posttranslational level. To investigate this phenomenon, we treated MDMs with vehicle or LPS for 12 hours (allowing LPS-induced processes to initiate) and then chased those with the protein synthesis inhibitor cycloheximide (CHX) or vehicle for a further 12 hours. LPS caused a steady decrease in NDUFA4 protein levels between the 12-and 24-hour time points (Fig. 4C). The lack of impact of CHX indicates that the phenomenon was not dependent on changes of de novo protein synthesis during this interval. LPS therefore causes active destabilization of NDUFA4 protein.
Under hypoxic conditions, HIF-1 induces expression of LONP1 and COX4I2 genes encoding, respectively, the mitochondrial protease LON and the alternative CcO subunit COX4-2 (20,22,43). The protease mediates degradation of COX4-1 and permits its substitution by COX4-2, modulating the affinity of the CcO complex for oxygen and sustaining its activity. LPS treatment of MDMs decreased the expression of the LONP1 gene ( fig. S2D) and did not alter COX4-1 levels (Fig. 4D). Therefore, despite the involvement of HIF-1, the response of macrophages to LPS does not resemble a classical hypoxia response, and LON is unlikely to be involved in the degradation of NDUFA4 protein. Unchanged expression of the mitochondrially encoded CcO component MT-CO1 (mitochondrially encoded cytochrome c oxidase 1) (Fig. 4D) and components of other electron transport chain complexes (Fig. 4E) suggests that the degradation of NDUFA4 is selective.
We then tested the effects of four distinct small interfering RNA (siRNA) molecules designed to target C15orf48 mRNA. Two of the siRNAs effectively blocked the LPS-induced increase in C15ORF48 protein, whereas the other two were nonfunctional (Fig. 4F). Each of the functional siRNAs partially blocked the LPS-induced decrease in NDUFA4 protein. The C15orf48-targeting siRNAs also prevented the replacement of NDUFA4 by C15ORF48 in CcO in LPS-treated MDMs (Fig. 4G). The differential effects of antagonizing C15orf48 . miR-147b (D) or NDUFA4 mRNA (E) detected by RT-qPCR (n = 3; means ± SEM). NDUFA4 protein (F) detected by Western blotting. Representative blot and quantification by densitometry normalized to -tubulin, reported relative to nontransfected condition (n = 3; means ± SD). (G) miRNA luciferase reporter assay performed in human embryonic kidney (HEK) 293 cells using unmodified reporter plasmid (pmirGLO), reporter plasmid containing wild-type (WT) NDUFA4 3′UTR (pmir-GLO NDUFA4 WT) and reporter plasmid containing mutated NDUFA4 3′UTR (pmirGLO NDUFA4 mut) (n = 3; means ± SEM). Schematic shows miRNA seed sequence and corresponding target sequence in WT and mutated plasmids. (C) One-way ANOVA with Dunnett's correction for multiple comparisons relative to time 0. (D to F) One-way ANOVA with Tukey's correction for multiple comparisons. (G) Two-way ANOVA with Sidak correction for multiple comparisons. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. (Fig. 4, F and G) and miR-147b ( Fig. 4B) suggested that expression of C15ORF48 protein itself may contribute to the down-regulation of NDUFA4 protein. To test this hypothesis, we transiently transfected human embryonic kidney (HEK) 293 cells with a vector expressing the C15ORF48 open reading frame but lacking the 3′UTR from which miR-147b is derived. Increasing the quantity of exogenous C15ORF48 protein was sufficient to cause a corresponding downregulation of endogenous NDUFA4 protein (Fig. 4H). C15ORF48 protein therefore promotes loss of NDUFA4 protein, most likely via its destabilization.

Expression of C15orf48 in RA and COVID-19
Bulk RNA sequencing (RNA-seq) was used to analyze differential gene expression in peripheral blood monocytes of patients with RA and healthy controls (K.W. and C.G., manuscript in preparation). Mean C15orf48 mRNA abundance was 30.48-fold higher in patients  with RA than in healthy, age-matched controls (Fig. 5A). This transcript was among the most strongly differentially expressed between patients and controls (P adj = 1.1 × 10 −17 ).
C15orf48 expression was elevated in bulk synovial tissue of patients with RA (Table 1). In situ hybridization showed expression of C15orf48 mRNA almost exclusively in CD68 + macrophages of RA biopsies (Fig. 5B), whereas no expression was detected in synovial biopsies of patients with osteoarthritis (OA). Strongly staining cells tended to be found in discrete clusters, which were present in the lining layer (Fig. 5B, left) and inflammatory infiltrates in the sublining layer (Fig. 5B, right). We then used our existing dataset (44) to investigate the expression of members of this gene family in the RA synovium at single-cell resolution. NDUFA4 was broadly expressed in all cell types, whereas expression of NDUFA4L2 was restricted to myofibroblasts and synovial lining layer fibroblasts, and C15orf48 mRNA was enriched in cells of myeloid origin (Fig. 5C). These patterns of expression were confirmed in a second, independent single-cell RNA-seq (scRNA-seq) dataset ( fig. S3A) (45). In this second cohort, expression of C15orf48 mRNA was significantly elevated in myeloid cells of leukocyte-rich synovial biopsies compared with OA or leukocyte-poor RA biopsies (P = 3 × 10 −5 ) ( fig. S3B). In a third, independent cohort of patients with RA (46), synovial biopsies were grouped into three pathotypes on the basis of immunohistology. F (fibroid or pauci-immune) pathotype displays little infiltration of the synovium by immune cells; M (diffuse myeloid) pathotype is characterized by strong myeloid infiltration with little evident cellular organization; L (lympho-myeloid) pathotype is characterized by infiltration of macrophages, T cells, and B cells demonstrating distinct organization into ectopic lymphoid structures. In bulk RNA-seq of synovial biopsies, C15orf48 mRNA was significantly enriched in biopsies of L and M pathotypes; NDUFA4L2 mRNA was enriched in M and F pathotypes; and NDUFA4 displayed no significant differential expression between L-, M-, and F-type synovial biopsies ( fig. S4, A and B). Abundance of C15orf48 mRNA in synovial tissue was significantly positively correlated with measures of disease activity, whereas NDUFA4 and NDUFA4L2 displayed no significant correlation with disease activity ( fig. S4C).
Our scRNA-seq of synovial biopsies (44) identified phenotypic clusters of synovial tissue macrophages (STMs) that broadly fall into two main populations with distinct roles in the regulation of joint inflammation (Fig. 5D, left). MERTK neg (MER tyrosine kinase) STMs promote inflammation by producing cytokines and activating synovial fibroblasts, whereas MERTK pos STMs have broadly anti-inflammatory or proresolution properties and are expanded in resolution of disease. NDUFA4L2 expression was extremely low in STMs. Expression of NDUFA4 was quite uniform but highest in population 1 (MERTK pos TREM2 high ) macrophages, which are thought to serve a protective, barrier-like function in the healthy joint (Fig. 5D, middle) (44,47). MERTK pos populations 1 to 4 expressed C15orf48 weakly, whereas the levels were higher in MERTK neg populations 6 to 9, particularly populations 8 (CD48 high S100A12 pos ) and 9 (CD48 pos SPP1 pos ) (Fig. 5D, right). The levels of C15orf48 mRNA were also relatively high in STMs of population 5 (MERTK pos FOL2R pos ICAM pos ), which, uniquely among MERTK pos STMs, display proinflammatory properties (44). Increased C15orf48 expression in proinflammatory STM populations was accompanied by increased expression of several genes involved in aerobic glycolysis and the pentose phosphate pathway ( fig. S5) (8), both of which fuel proinflammatory functions of myeloid cells (1-10, 41). STM expression of C15orf48 was elevated in active RA but, in remission, returned almost to the levels seen in healthy control synovium (Fig. 5E). The proportion of NDUFA4 pos / C15orf48 pos macrophages significantly increased in active RA and returned to near healthy levels in remission. Contrarily, the proportion of NDUFA4 pos /C15orf48 neg macrophages significantly declined in active RA and returned to near healthy levels in remission (Fig. 5F).
Although expression of C15orf48 mRNA was clearly elevated in STMs, in two independent studies (44,45), expression was also detected in some synovial fibroblasts ( Fig. 5C and fig. S3A). TNF strongly and consistently elevated the expression of C15orf48 mRNA in synovial fibroblasts (Fig. 5G). Although not conclusive, these observations collectively suggest that increased C15orf48 expression may contribute to the pathogenesis of RA: C15orf48 expression is enriched in myeloid cell populations that demonstrate proinflammatory properties; associated with active, myeloid cell-mediated joint inflammation; positively correlated with clinical measures of disease activity; negatively correlated with remission of disease; increased in peripheral monocytes of patients with RA; and up-regulated by a highly disease-relevant proinflammatory stimulus in stromal cells that contribute to joint destruction (48).
C15orf48 expression was also significantly elevated in lung tissue or bronchoalveolar lavage (BAL) of patients with severe COVID-19 (49,50). We recently reported (51) that S100A12 pos and SPP1 pos STM clusters (clusters 8 and 9 in Fig. 5D), which are abundant in active RA, transcriptionally resemble FCN1 pos and FCN1 pos SPP1 pos BAL macrophage clusters that predominate in severe COVID-19 (fig. S6, A and B) (52). We therefore analyzed the expression of C15orf48 and NDUFA4 in BAL macrophages from healthy controls and patients with COVID-19, using data from Liao and colleagues (52).

Consequences of remodeling of CcO in human macrophages
LPS treatment of murine myeloid cells provokes repurposing of mitochondria for production of ROS, accompanied by impairment of oxidative phosphorylation (40,41). This response is at least partly mediated by nitric oxide as a consequence of up-regulation of Nos2 gene expression (6,53). To meet energetic and biosynthetic needs of the cell, aerobic glycolysis is increased in parallel. We hypothesized that the substitution of NDUFA4 by C15ORF48 contributes to these profound metabolic changes. However, the LPS responses of human macrophages were found to differ from those of murine macrophages. LPS treatment of mouse BMDMs caused the expected impairment of maximum mitochondrial respiratory capacity but no change in basal respiration (Fig. 6A, top). In contrast, LPS increased basal respiration of human macrophages but did not affect maximal respiratory capacity (Fig. 6A, bottom). Glycolytic capacity was increased by LPS in macrophages of both species (Fig. 6B), indicating that Warburg-like metabolism can be uncoupled from impairment of oxidative phosphorylation. LPS treatment of human macrophages did not significantly increase mitochondrial ROS production (Fig. 6C). Because the switch between NDUFA4 and C15ORF48 occurs in LPS-treated mouse and human macrophages but human macrophages do not lose respiratory capacity under these conditions, we suggest that the subunit exchange within CcO is not responsible for impaired oxidative phosphorylation. The fact that repurposing of mitochondria for ROS production does not occur in human MDMs may be explained by the fact that, unlike murine BMDMs, they do not produce nitric oxide in response to LPS, even when IFN- is added (Fig. 6D).
We directly measured CcO activity in macrophages to gain better insight into the potential consequences of the subunit switch. In-gel activity assays did not show prominent changes of CcO activity in response to LPS (Fig. 6E). A more quantitative complex-specific metabolic flux assay revealed a small but statistically significant (16 ± 4%) LPS-induced reduction of CcO activity (Fig. 6F). We attempted to use RNA interference to investigate the metabolic and functional consequences of remodeling of the CcO complex. However, the process of transfecting primary MDMs (using three chemically distinct transfection reagents) altered both metabolic functions and inflammatory outputs. Three different control siRNAs exerted additional donor-and gene-specific effects, rendering conclusions about specific consequences of NDUFA4 or C15ORF48 knockdown impossible. Electroporation also caused sustained mitochondrial ROS production indicating disruption of metabolism. Derangement of cellular metabolism is a hallmark of oncogenic transformation (54); for example, LPS did not increase aerobic glycolysis in the mouse macrophage cell line RAW264.7. We reasoned that knockdown or gene editing approaches in transformed myeloid cell lines were likely to generate misleading results.
We therefore turned our attention to a very rare genetic condition, in which a mutated allele of the NDUFA4 gene produces no detectable NDUFA4 protein (26). Individuals who are homozygous for this loss-of-function mutation display severe neuromuscular symptoms, but inflammatory features have not previously been investigated. Monocytes were isolated from all three known, surviving homozygous individuals, and differentiated into macrophages in vitro. In these macrophages, there was a de facto substitution of NDUFA4 by C15ORF48 under resting conditions (Fig. 7A). Basal expression of C15ORF48 was possibly increased in the absence of NDUFA4, but the limited number of replicates makes this conclusion tentative at best. The levels of the CcO components MT-CO1 and COX4-1 were unaffected by the NDUFA4 mutation. These macrophages represent a model for the specific substitution of NDUFA4 in the absence of a proinflammatory stimulus. Under resting conditions or following stimulation with LPS, there were no significant differences between NDUFA4-null macrophages and those derived from healthy controls in terms of mitochondrial, glycolytic, or total adenosine 5′-triphosphate (ATP) production rates (Fig. 7B). Constitutive loss of NDUFA4 therefore did not seem to impair mitochondrial ATP production or cause a shift to glycolytic metabolism under resting conditions in this cell type; neither did it influence the metabolic response to LPS.
Gene expression was compared between control and NDUFA4null macrophages using a NanoString inflammation panel. We focused on unstimulated macrophages because differences in expression of CcO components were most notable under this condition (Fig. 7A). Principal components analysis clearly separated NDUFA4null from control macrophages, with the first principal component accounting for 98% of experimental variance (Fig. 7C). Controls clustered together well, whereas cases were more dispersed, suggesting that there may be variations in penetrance of the condition caused by loss of NDUFA4 function. Across 249 inflammation-related transcripts represented in the panel, 47 were significantly (P < 0.05) differentially expressed between NDUFA4-null and control macrophages (Fig. 7D). All were overexpressed by NDUFA4-null macrophages with the single exception of HIF1A (fold difference case/ control = 0.652, P = 0.026). One of the most notable differentially expressed genes was CFB, encoding complement factor B, the C3 convertase of the alternative pathway (fold difference case/control = 16.0, P = 0.00018). Expression of the complement genes C1QA, C1QB, and C1R was also elevated in NDUFA4-null macrophages (Fig. 7D), whereas other complement genes were either not expressed or not differentially expressed.
Several genes encoding cytokines and CCL and CXCL chemokines were overexpressed at the mRNA level by unstimulated NDUFA4-null macrophages (Fig. 7D). To confirm this broadly proinflammatory pattern, we used multiplexed bead immunoassays to quantify secreted factors in supernatants of unstimulated control and NDUFA4-null macrophages. Principal components analysis again demonstrated clear separation between control and NDUFA4-null macrophages, 92% of experimental variance being accounted for by the first principal component (Fig. 7E). Again controls clustered well, whereas cases were more dispersed in PC1. Twenty-one of 45 analytes were significantly overexpressed by NDUFA4-null macrophages, whereas none were significantly underexpressed ( Fig. 7F; absolute quantities are plotted in fig. S7). The expression of several cytokines and chemokines was increased by more than 10-fold in NDUFA4-null macrophages ( fig. S7), for example, CCL13 (10.75fold), CXCL1 (12.78-fold), and CXCL2 (13.58-fold). In some cases, fold differences of expression had to be estimated because levels were either below detection limit in control or above detection limit in NDUFA4-null macrophages. For example, IL-6 was overexpressed at least 40-fold by NDUFA4-null macrophages ( fig. S7). Several secreted factors (for example, CCL21 and IL-12p70) were well within linear range of the assay and did not differ significantly between control and NDUFA4-null macrophages, emphasizing that constitutive loss of NDUFA4 has broad but not indiscriminate effects on macrophage biology. Although the NanoString and multiplexed bead panels did not perfectly overlap, where they did overlap there was generally good agreement as to altered expression at mRNA and protein levels. Collectively, these data reveal a notable proinflammatory phenotype of macrophages lacking NDUFA4 protein, suggesting that exchange of CcO components plays an important role in the regulation of macrophage functions.

DISCUSSION
C15ORF48 (NMES1) was originally described as a nuclear protein (36); however, this was based on low-resolution immunohistology using an unvalidated polyclonal antiserum. Although lacking canonical mitochondrial localization signals, both NDUFA4 and C15ORF48 are now annotated as mitochondrial proteins in the expert-curated databases MitoCarta (55) and the Integrated Mitochondrial Protein Index (www.mrc-mbu.cam.ac.uk/impi). This annotation is supported by published evidence (26,38,(55)(56)(57) including confocal microscopy and cell fractionation experiments using overexpressed C15ORF48 (38,56). To this, we have added cell fractionation of primary human macrophages and Western blotting of endogenous proteins with antibodies validated by knockdown, showing predominantly mitochondrial staining of both proteins (Fig. 3G). Under conditions of maximal expression (without and with LPS in the cases of NDUFA4 and C15ORF48, respectively), we could detect very weak staining of each protein in nuclear fractions. This was almost certainly a result of mitochondrial contamination, because similar weak staining of the mitochondrial protein SDHA (succinate dehydrogenase complex subunit A) could be detected in nuclear fractions. We conclude that both NDUFA4 and C15ORF48 are exclusively, or almost exclusively, mitochondrial proteins.
The NDUFA4 and C15orf48 genes share an extraordinarily complex relationship. They have limited sequence similarity but their exon-intron structures are the same. The positions of the first and third exon boundaries are identical, while the second is shifted by only two nucleotides, a very high degree of relatedness that points to an origin in gene duplication. The encoded proteins share less than 30% amino acid identity but are predicted to have similar secondary structures and electrostatic properties (Fig. 3B and fig. S1, A and B) (56,57). The 3′UTR of C15orf48 mRNA is also the source of the miRNA miR-147b. The protein and miRNA products of the C15orf48 transcript converge to bring about down-regulation of NDUFA4 protein levels; miR-147b specifically recognizes a conserved site within the NDUFA4 3′UTR, causing down-regulation of this target mRNA (Fig. 2); and C15ORF48 protein promotes loss of NDUFA4 protein (Fig. 4H), most likely by inducing its degradation (Fig. 4C). While this manuscript was in preparation, other researchers described similar combinatorial effects of miR-147b and C15ORF48 protein on NDUFA4 expression in transfected HEK293 (embryonic kidney) or A549 (airway epithelial) cell lines (56,57). The exceptional evolutionary conservation of both protein-and miRNAencoding portions of the C15orf48 gene (Fig. 1A) suggests that this complex phenomenon has been subject to powerful positive selection for some hundreds of millions of years (56).
Not all of the effects of miR-147b can be attributed to downregulation of NDUFA4 (57). miRNAs commonly target several transcripts within a single biological process or pathway; therefore, we were interested in two putative miR-147b targets that encode mitochondrial proteins; SDHD and ALDH5A1 (aldehyde dehydrogenase 5 family member A1) (58,59). miR-147b-mediated downregulation of SDHD was reported to contribute to TCA cycle dysfunction in lung adenocarcinomas (59). The predicted interactions between miR-147b and SDHD mRNA were not conserved in the mouse. Furthermore, SDHD mRNA levels in MDMs were not significantly reduced by either LPS treatment or transfection of a miR-147b mimic. Seed sequence interactions between miR-147b and ALDH5A1/ Aldh5a1 mRNAs were conserved between human and mouse, but there were few base pairs outside the seed region. ALDH5A1 mRNA was strongly down-regulated following treatment of human MDMs with LPS, but transfection with a miR-147b mimic failed to recapitulate this effect. We conclude that neither ALDH5A1 nor SDHD is likely to be a major target of miR-147b in MDMs and that further relevant targets remain to be identified.
As indicated by its name, NDUFA4 was originally thought to be a component of mitochondrial electron transport chain complex I (the NADH dehydrogenase complex). More recent cryo-EM structural evidence places NDUFA4 at the outer surface of the CcO complex (complex IV) (28), and molecular modeling suggests that C15ORF48 could occupy the same position (Fig. 3B). Consistent with these reports, in primary human macrophages, the pattern shown by NDUFA4 and C15ORF48 proteins was consistently that of CcO complex subunits (Figs. 3H and 4G). Others have reported similar findings (25,26,57,60), lending strong support to a change in systematic name from NDUFA4 to COXFA4 (27). LPS treatment of primary human macrophages caused an increase in levels of C15ORF48 protein and a reciprocal decrease in levels of NDUFA4 (Fig. 3C), resulting in substitution of NDUFA4 by C15ORF48 in all CcO complexes (Fig. 3H). LPS-induced degradation of NDUFA4 protein (Fig. 4C) may be a consequence of its exclusion from CcO (57). Overexpression of C15ORF48 caused a similar protein substitution in CcO complexes of mouse heart mitochondria (57). Efficient substitution of NDUFA4 by C15ORF48 in CcO also occurred during mouse spermatogenesis (60), most likely involving the mechanisms described above. The question is why cells should go to such lengths to replace one mitochondrial protein with another. Put another way, what are the consequences of the replacement of NDUFA4 by C15ORF48?
Depletion of NDUFA4 impaired CcO function in HeLa or muscle cells (25,26). HeLa cells do not express C15orf48 mRNA in either normoxia or hypoxia, and muscle tissue expresses little or no C15ORF48 protein (Human Protein Atlas); therefore, the decrease in CcO activity upon NDUFA4 depletion in these studies likely represents conditions in which neither protein is present. In contrast, we found only very slight impairment of CcO function in LPS-treated MDMs (Fig. 6F), and no deficit in mitochondrially derived ATP production due to either genetically determined or LPSinduced loss of NDUFA4 protein (Fig. 7B). Therefore, NDUFA4 is dispensable for mitochondrial respiration in MDMs, most likely because it can be functionally replaced by C15ORF48, even under resting conditions (Fig. 7, A and B). Under hypoxic conditions, HIF-1 promotes the substitution of NDUFA4 by NDUFA4L2, mitigating mitochondrial ROS production and consequent cell death (30,32,34,42). Others have reported or speculated that substitution of NDUFA4 by C15ORF48 similarly reduces mitochondrial ROS production under hypoxia or other stressful conditions (56,57,60). This appears not to be the case in primary human MDMs, for the following reasons: (i) Although LPS activates HIF-1 and decreases NDUFA4 protein levels in MDMs, these occur in the absence of any changes of C15ORF48 expression ( fig. S2). (ii) Unlike mouse BMDMs, human MDMs do not produce nitric oxide (Fig. 6D), increase mitochondrial ROS production (Fig. 6C), or display impairment of oxidative phosphorylation (Fig. 6A) in response to LPS. (iii) In mouse myeloid cells, LPS-induced mitochondrial ROS production is principally mediated by reverse electron transport at complex I (61). There is no documented mechanism by which alterations in composition of CcO (complex IV) can influence this process. In the absence of new evidence, consequences of the NDUFA4-C15ORF48 substitution must be sought elsewhere.
An alternative hypothesis is that exchange of subunits affects the ability of CcO to participate in higher-order electron transport chain structures. Dimerization of CcO was proposed to exclude the binding of NDUFA4, with effects on respiratory efficiency and sensitivity to negative feedback by high intracellular concentrations of ATP (21). However, both NDUFA4 and C15ORF48 were detected in both CcO monomeric and dimeric structures (Fig. 3H), implying that the switch does not contribute to regulation of dimerization. Both NDUFA4 and C15ORF48 can interact with subunits of complexes I and III, leading to the suggestion that they may regulate the formation of higher-order electron transport chain complexes (38,62,63). Assembly factors by definition do not stably associate with the complexes that they help to form. Both NDUFA4 and C15ORF48 could be consistently detected in supercomplexes (Fig. 3H) (28), and therefore, they do not seem to be supercomplex assembly factors but rather structural components of CcO necessary for its optimal activity (26). We reproducibly observed an increase in abundance of supercomplexes following LPS treatment of MDMs (fold increase 1.90 ± 0.24 SEM; n = 4; P < 0.01) (Fig. 6E). However, we cannot yet conclude whether this increase was dependent on exchange of CcO subunits. We also cannot rule out functions that are not directly related to oxidative phosphorylation. For example, loss of NDUFA4 has been linked to increased sensitivity to stress-induced apoptosis (56). Functions outside of the mitochondria appear unlikely.
Expression of C15orf48 was increased by LPS treatment of primary macrophages from mouse, human ( Fig. 1) (37), and several other mammalian species (64). It was also increased by treatment of synovial fibroblasts with TNF (Fig. 5G); A549 cells (56) or mouse macrophages with type I IFN (Fig. 1G); RAW264.7, A549, HL-60, and L929 cells with type II IFN (56); A549 cells or astrocytes with IL-1 (65,66); microglia with M1 polarizing agonists LPS and/or IFN- (67); and natural killer cells with IL-2 and IL-12 (68). Of course, these potent responses to proinflammatory agonists do not prove a proinflammatory role for C15orf48. Most reports to date, using overexpression in a variety of primary and transformed cell types, agree that the protein and miRNA products of C15orf48 mRNA have anti-inflammatory effects (37,57,66,69,70). However, a proinflammatory function of miR-147b was previously reported (71) and is also suggested by the studies described here.
Expression of C15orf48 was notably increased under several inflammatory conditions of autoimmune, metabolic, cardiovascular, bacterial, or viral nature (Table 1). These associative data still do not prove that C15orf48 has a causative role in inflammatory pathologies; therefore, we investigated in more detail the expression of C15orf48 and its relatives in RA. Expression of C15orf48 was positively correlated with disease activity scores in bulk RNA-seq of RA synovial biopsies ( fig. S4) and decreased in remission of disease (Fig. 5E). By scRNA-seq C15orf48 expression was found to be strongly enriched in STM populations having demonstrable proinflammatory functions (Fig. 5D) (44). In these cell populations, elevated expression of C15orf48 was accompanied by elevated expression of genes belonging to glycolytic and pentose phosphate pathways (fig .  S5) (8,44). NDUFA4 was expressed quite widely but most strongly in TREM2 high STMs (Fig. 5D), which are believed to play an important role in the preservation of synovial joint integrity (44,47). In this population, elevated expression of NDUFA4 was accompanied by elevated expression of genes belonging to the oxidative phosphorylation pathway and by low expression of genes contributing to aerobic glycolysis and the pentose phosphate pathway ( fig. S5) (8, 44).
Analogous to its up-regulation in proinflammatory synovial macrophages during RA, C15orf48 was also found to be elevated in severe COVID-19 and to be expressed by subsets of macrophages that have been implicated in pathogenic processes ( fig. S6) (51,52). In addition, mirroring RA STM subsets, NDUFA4-expressing alveolar macrophages (FABP4 pos ) showed elevated expression of genes related to oxidative phosphorylation and fatty acid metabolism and decreased glycolytic gene expression, while the opposite was true for C15orf48-expressing subsets (FCN1 pos or FCN1 pos SPP1 pos ) (52). Divergent expression of NDUFA4 and C15orf48 in macrophage populations therefore appears to be part of a consistent pattern of metabolic commitment to either homeostatic or proinflammatory, destructive processes across different diseases (72). Mitochondrial dysregulation has been linked to pathogenic processes in RA (8), and in COVID-19, mitochondrial fitness has been posited as a determinant of COVID-19 disease severity (73). Severe acute respiratory syndrome coronavirus 2 viral components have also been suggested to interact with mitochondria to facilitate viral replication and evade host immunity, thereby disrupting mitochondrial function (73)(74)(75). These hypotheses still require full experimental validation, and it will be interesting to ascertain the precise role that the C15orf48/NDUFA4 axis plays in these disease processes.
Mutations of the NDUFA4 gene allowed us to ask what happens in primary human macrophages that already lack NDUFA4 protein in the absence of any stimulus, and in which only C15ORF48 protein is available to occupy the position on the outer surface of CcO. Under resting conditions, these macrophages very strongly overexpressed a large number of secreted factors, including several chemokines of the CCL and CXCL families (Fig. 7 and fig. S7). It remains to be determined whether dysregulated innate immunity contributes to the severe condition caused by homozygous NDUFA4 mutations (26). Collectively, these observations provide circumstantial evidence that the up-regulation of C15orf48 mRNA and the consequent replacement of NDUFA4 by C15ORF48 in CcO promote inflammation, although there are contradictory data derived from in vitro experiments. Conclusive evidence may depend on the development of better experimental tools, such as conditional knockout mouse strains. The notable evolutionary conservation of the mechanism and its recapitulation at many sites of inflammatory pathology suggest that further research is strongly justified.
All mice were maintained at the Biomedical Services Unit of the University of Birmingham. Animal care and experimental procedures were performed according to the Home Office guidelines and approved by the University of Birmingham Local Ethical Review Committee. Wild-type (WT) C57/BL6J mice between 6 and 12 weeks of age were humanely culled, bone marrow-flushed from femurs, and BMDM-obtained by culture for 7 days in RPMI 1640 with l-glutamine (Gibco, Thermo Fisher Scientific, 21875034) supplemented with 10% heat-inactivated fetal bovine serum (Sigma-Aldrich, F0392) and recombinant M-CSF (50 ng/ml; PeproTech, 300-25).

Synovial fibroblasts
Synovial tissue samples were obtained from patients undergoing joint replacement surgery or by ultrasound-guided biopsy from treatmentnaïve patients attending an early arthritis clinic with joint pain and/ or inflammation of duration of less than 12 weeks . Fibroblasts were isolated as previously described (76). Between passages 4 and 6, fibroblasts were left untreated or stimulated with TNF (100 ng/ml) for 24 hours. Differential gene expression was determined by Agilent microarray (A.F. in preparation). Stratification of patients (76) revealed no differences in the response of the C15orf48 gene to TNF according to disease stage or outcome; therefore, all 48 independent synovial fibroblast isolates were grouped together. All human samples were obtained with written, informed consent and approval from the West Midlands Black Country Local Research Ethics Committee (references 07/H1204/191 and 07/H1203/57).

NDUFA4 patients
Three individuals suffering from NDUFA4 mutation-associated Leigh syndrome were recruited at the Queen Square Centre for Neuromuscular Diseases, University College London, under the ethical approval by the Queen Square Research Ethics Committee, London (no. 09/H0716/76). Informed consent was obtained from all participants. Genetic and clinical patient details are described in (26). Monocytes were isolated from peripheral blood and differentiated into macrophages as described above.

Transfections
Macrophage transfections were performed on day 6 of macrophage differentiation in 12-well culture plates using Mirus TransIT-X2 reagent (Geneflow, E7-0106) according to the manufacturer's instructions. miRNA mimic experiments were performed using Dharmacon miRIDIAN human miR-147b or negative control mimics (2 nM; Horizon Discovery). Antagomir experiments were performed using miRCURY LNA (locked nucleic acid) human miR-147b or negative control power inhibitors (25 nM; QIAGEN). siRNA transfections were performed using Dharmacon siGenome siRNAs and nontargeting controls (25 nM; Horizon Discovery). Nontransfected samples were treated with TransIT-X2 transfection reagent alone. Samples were harvested 24 hours after transfection for RNA analysis or 48 hours after transfection for protein analysis.

Western blotting
For whole-cell lysates, cells were harvested in radioimmunoprecipitation assay buffer and samples passed through a QIAshredder column to remove genomic DNA (QIAGEN, 79656). Cell fractionation was performed essentially as described in (77), using 10 × 10 6 to 20 × 10 6 primary MDMs per condition and a 2-ml Dounce tissue grinder with small clearance pestle (30 strokes) (Sigma-Aldrich, D8938). Nuclear fraction was taken as pellet following homogenization and centrifugation at 1200g. Cytoplasmic fraction was taken as supernatant following mitochondrial sedimentation at 17,000g. Protein was quantified by the Pierce BCA Assay (Thermo Fisher Scientific, 23225). Laemmli buffer was added and samples heated to 95°C for 5 min (unless blotting for MT-CO1). Western blotting was performed using Criterion TGX protein gels (Bio-Rad) and tris-glycine SDS buffer (Geneflow, B9-0032) or using XT bis-tris protein gels (Bio-Rad) and XT MES running buffer (Bio-Rad, 1610789). Protein was transferred to Bio-Rad Trans-Blot polyvinylidene difluoride (PVDF) membranes (Bio-Rad, 1704157) using Bio-Rad Trans-Blot Turbo transfer system. The following antibodies and dilutions were used: C15orf48 (1:1000; Antibodies-online.com, ABIN2784037); NDUFA4 (1:5000; Thermo Fisher Scientific, PA5-50068); -tubulin Luciferase 3′UTR reporter assay WT NDUFA4 3′UTR was amplified from human genomic DNA and cloned into the pmirGLO Dual-Luciferase miRNA Target Expression Vector (Promega, E1330) using standard molecular cloning methods. ImiRP computational biology tool (imirp.org) was used to predict tolerated mutation of the miR-147b target site without the generation of new miRNA binding sites. The WT miR-147b seed sequence target site (CCGCACA) was mutated to CCGCGGG to produce pmirGLO NDUFA4mut using the QuikChange II Site-Directed Mutagenesis Kit (Agilent, 200523). Cloning was confirmed by Sanger sequencing (Eurofins Genomics).
HEK293 cells were seeded at 100,000 cells per well in 24-well plates and transfected with 20 nM control mimic or miR-147b mimic (Horizon Discovery) along with pmirGLO plasmid DNA at 0.5 g per well (unmodified/NDUFA4 WT/NDUFA4mut) using TransIT-X2 reagent (Geneflow, E7-0106). Luciferase assays were performed using the Dual Luciferase Assay Kit (Promega, E1910) according to the manufacturer's instructions using cell lysates harvested 1 day after transfection and a plate reader integration time of 0.1 s. Measurements were performed in technical triplicate, and the experiment was performed three separate times.

Protein modeling
The model of C15ORF48 was built using the structure of NDUFA4 as reported in the human complex IV (28), according to the default energy-based homology model protocol implemented in Prime (Prime, Schrödinger, LLC, New York, NY, 2020). Unsolved atoms in side chains of residues 3 to 11, 42 to 44, and 46 to 83 of NDUFA4 were added. The completed side chains were assigned an energy minimum conformation. Sequences of NDUFA4 and C15ORF48 were retrieved from UniProt (www.uniprot.org). In light of the levels of identity (30%) and similarity (44%) between the two sequences, the default Prime pairwise sequence alignment routine was adopted. During the coordinate transfer, the cardiolipin from the experimental structure was explicitly retained. The multimeric structure encompassing C15ORF48 and all subunits from complex IV except NDUFA4 was obtained superimposing the backbone coordinates of the model to those of the template, deleting the template and minimizing the generated complex. The position and thickness of the mitochondrial membrane were assigned according to the predictions generated by the OPM (orientation of proteins in membranes) server.

One-dimensional BN-PAGE
Mitochondrial-enriched fractions were prepared from unstimulated or LPS-treated human MDMs (1 × 10 6 to 2 × 10 6 cells per condition) by treatment with digitonin (4 mg/ml), centrifugation at 10,000g (4°C), and washing. Mitochondrial membranes were solubilized with 10% digitonin in 1.5 M aminocaproic acid, 50 mM bis-tris/HCl (pH 7.0), followed by centrifugation at 18,000g (4°C). BN sample buffer was added at 10% volume [0.75 M aminocaproic acid, 50 mM bis-tris/ HCl (pH 7.0), 0.5 mM EDTA, 5% Coomassive Brilliant Blue G-250] and a portion of each sample run on NativePAGE 3 to 12% bis-tris gel (Thermo Fisher Scientific, BN1001) according to the manufacturer's instructions. For Western blotting, protein was transferred to PVDF membranes by wet transfer, and protein detection was performed as described above. For in-gel activity assay, the gel was incubated overnight with complex IV activity solution as described (77).

In situ hybridization and immunohistochemistry
RNAscope in situ hybridization was carried out on formalin-fixed paraffin-embedded human joint replacement tissues sections of OA and RA synovium provided by the Royal Orthopaedic Hospital (ethical approval no. 07/H1204/191). Sections were deparaffinized in xylene, followed by 100% ethanol. Detection of human C15orf48 transcript was performed using an RNAscope 2.5 HD manual assay red kit reagents and methods according to the manufacturer (ACD Bio-Techne). Following the last wash for the RNAscope protocol, slides were washed in distilled water for 5 min and then blocked with Bloxall (Vector Laboratories) for 10 min, followed by incubation with 10% normal horse serum in tris buffer for 10 min. Immunohistochemistry was performed by incubating sections overnight at 4°C with biotin-conjugated mouse anti-human CD68 (Novus, NBP2-34661B), followed by streptavidin-HRP (Thermo Fisher Scientific) and ImmPACT DAB Peroxidase (HRP) Substrate (Vector Laboratories). Hematoxylin QS (Vector Laboratories) was used for counterstain. Images were obtained using the Zeiss Axio Scan and analyzed in Zen Blue (both Zeiss).

Metabolic activity assays
Metabolic assays were carried out using an Agilent Seahorse XFe96 Extracellular Flux analyzer. For intact cell assays, macrophages were seeded into Seahorse XFe96 assay plates at 50,000 cells per well (human) or 75,000 (mouse), left for 1 hour at room temperature to settle, and incubated overnight to adhere. Cells were left unstimulated or stimulated with LPS (10 ng/ml) for 24 hours. For normalization of measurements, a viable cell count ratio was determined.

Mitochondrial ROS measurement
Mitochondrial ROS was detected using MitoSOX dye (5 M; Thermo Fisher Scientific, M36008). Staining was performed in culture plates immediately following the stimulation period. Cells were washed with PBS and stained for 20 min at 37°C in Hanks' balanced salt solution + Ca 2+ + Mg 2+ . For positive control, antimycin A was spiked into the well 1 hour before staining. Cells were run on a BD LSRFortessa X-20, and results were analyzed using FlowJo v10. Dead cells were excluded on the basis of positive staining for eBioscience Fixable Viability Dye eFluor 780 (Thermo Fisher Scientific, 65-0865-14).

Nitric oxide measurement
Nitrite concentration in conditioned medium from cultured macrophages was determined using the Griess reagent kit (Thermo Fisher Scientific, G7921) according to the manufacturer's instructions. Briefly, cultured cell supernatant was collected after treatment and centrifuged to remove debris. The supernatant (100 l) was added to 96-well plate containing 20 l of deionized water per well and followed by the addition of an equal volume of N-(1-naphthyl)ethylenediamine and sulfanilic acid mixture to a total volume of 200 l. The reaction was incubated for 30 min at room temperature. Absorbance at 548 nm was measured by microplate reader, and nitrite concentrations were estimated using a standard nitrite curve and correcting for background noise.

NanoString gene expression analysis
Gene expression analysis from NDUFA4 patient and control macrophages was performed using the NanoString nCounter Human Inflammation Panel (NanoString, XT-CSO-HIN2) according to the manufacturer's instructions, using 150 ng of RNA per sample. Analysis and visualization were performed using NanoString nSolver software and Qlucore Omics Explorer 3.6.

Multiplex cytokine analysis
Conditioned medium samples from cultured macrophages were subjected to multiplex Luminex assay (Biotechne R&D Human magnetic luminex assay LXSAHM and Bio-Rad Bio-Plex Pro Human Chemokine Panel, 40-Plex 171AK99MR2) according to the manufacturer's instructions. Differential abundance analysis and data visualization were performed using Qlucore Omics Explorer 3.6.

Statistics
Statistical analyses were carried out using GraphPad Prism v6 unless otherwise stated. Statistical tests and corrections used are indicated in the figure legends. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. N numbers specified in the figure legends indicate biological replicates.

SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at https://science.org/doi/10.1126/ sciadv.abl5182 View/request a protocol for this paper from Bio-protocol.