Interferon Regulatory Factor 5 represses oxidative respiration in human and murine macrophages by inhibition of mitochondrial matrix protein GHITM


 Over the course of diet-induced obesity (DIO), adipose tissue macrophage (ATM) populations transition from highly oxidative and protective to highly inflammatory and metabolically deleterious. Here, we demonstrate that the Interferon Regulatory Factor (IRF)-5 is a key molecular switch mediating repression of macrophage oxidative capacity early in DIO. ATMs from mice with a myeloid-specific deletion of IRF5 are characterised by increased mitochondrial activity and oxidative respiration compared to ATMs from wild-type mice. This hyper-oxidative phenotype is inducible and reversible in vitro by delivery of an inhibitory IRF5-decoy peptide and through IRF5 adenoviral over-expression, respectively. In a data-driven approach, using public IRF5-cistrome data and in-house RNA-sequencing, we identified a transcriptional mechanism by which cellular oxidative capacity is repressed in an IRF5-dependent manner. The hyper-oxidative phenotype of IRF5-deficient macrophages is mediated by the Growth Hormone Inducible Transmembrane Protein (GHITM), known for maintaining mitochondrial architecture for optimal oxidative respiration. Cas9-mediated simultaneous knock-down of GHITM and of IRF5 reverses the hyper-oxidative phenotype associated with IRF5-deficiency in vitro. IRF5-dependent regulation of GHITM expression and mitochondrial activity extends to human ATMs and monocytes from obese and diabetic patients.


Introduction
ATMs are a heterogenous population of cells with a range of bene cial and deleterious functions [1][2][3][4][5][6] . Lipid associated macrophages (LAMs) have been reported to be a key bene cial ATM population, with a high capacity to clear dead and dying adipocytes as well as lipid droplets 1 , while metabolically activated macrophages (MMe) exhibit both bene cial and harmful functions 2 . A recent report by Sárvári et al. has also identi ed 6 distinct subpopulations of ATMs through snRNA-seq, of which two LAM subpopulations were characterised 5 . Predominance of each of these ATM populations is extremely time-dependent. ATMs physiologically attempt to maintain homeostasis in early caloric excess and over time, they become increasingly in ammatory, contributing to systemic in ammation and worsening insulin resistance.
In ammatory ATMs are critical actors in the development of type-2 diabetes (T2D) and other manifestations of metabolic syndrome.
Macrophage cellular metabolism is a major mitigating or optimising factor of immune effector functions 7,8 . Pro-in ammatory functions are dependent on glycolysis, transcriptionally mediated by hypoxia inducible factor (HIF)-1α, and targeted interruptions of the mitochondrial tricarboxylic acid (TCA) cycle. Conversely, anti-in ammatory polarisation or resolution is supported by mitochondrial respiration and oxidative metabolism. The hallmark investigations that brought these mechanisms to light were carried out in model systems of macrophages applying canonical stimulants of polarisation. More recent studies found that, in addition to canonical stimuli, microenvironmental niches also dictate energetic responses of immune cells. For example, highly active mitochondrial respiration gives regulatory T-cells a metabolic advantage over other immune cells in low-glucose high-lactate environments 9 .
In the context of metabolic syndrome, cells of the immune system are exposed to the same dysmetabolism as all peripheral tissues, that is glucolipotoxicity which represents a systemic abundance of metabolic substrates. In diet-induced obesity, ATMs are now known to have their own adaptive metabolism. They exhibit a hypermetabolic phenotype with excessive glycolytic and oxidative uxes compared to lean physiological circumstances [10][11][12][13][14][15] . These important ndings indicate that as sentinel cells, macrophages are reactive beyond in ammatory polarisation and their successful early metabolic adaptation may mitigate in ammatory polarisation and systemic metabolic decline.
Our previous work identi ed the Interferon Regulatory Factor (IRF)-5 as a key molecular switch mediating M1-like polarisation of ATMs 16 . We demonstrated that chronic high-fat feeding increases IRF5 expression in ATMs, in turn promoting pro-in ammatory polarisation and the transcriptional repression of TGFβ signalling. This results in maladaptive white adipose tissue (WAT) expansion and systemic metabolic decline. Physiologically, IRF5 promotes transcription of pro-in ammatory cytokines and chemokines in macrophages in response to bacterial and viral stimuli [17][18][19] . Several lines of evidence implicate deregulated IRF5 expression in conditions of chronic in ammation, including auto-immune and metabolic diseases [20][21][22][23] . Notably, gain-of-function IRF5 risk-variants are associated with a multitude of auto-immune diseases and, more recently, with promoting macrophage glycolytic programming in autoimmunity (systemic lupus, irritable bowel disease, rheumatoid arthritis) 16,24,25 . In-line with these reports, we hypothesised that IRF5 may also orient ATM metabolism, notably by repressing oxidative respiration that supports protective macrophage phenotypes predominant in early caloric excess.
Here, we reveal a non-canonical function of IRF5 in orienting bioenergetic adaptation of ATMs early in the course of diet-induced obesity. Upon short-term caloric excess we demonstrate that IRF5 restricts the cellular lipid pool and downregulates expression of components required to maintain e cient mitochondrial oxidative respiration. Through downregulation of the Growth Hormone Inducible Transmembrane protein (GHITM), also known as Mitochondrial Morphology and Cristae Structure 1 (MICS1), IRF5 contributes to failure in maintaining normal inner membrane architecture required for effective oxidative respiration. Furthermore, the IRF5-GHITM axis of cellular bioenergetic adaptation extends to circulating monocytes and ATMs in diabetic and obese humans.

Results
Caloric excess is associated with ATM metabolic adaptation and increased IRF5 expression To evaluate metabolic adaptation in ATMs, we placed C57BL/6J mice on a high-fat or normal chow diet (HFD, NCD) for 4 weeks, and quanti ed IRF5 expression and metabolic markers in epidydimal WAT (EpiWAT) ATMs. As expected, mice on HFD gained more weight than mice on NCD over the 4 weeks, increasing WAT mass (Fig. S1A) and losing systemic insulin sensitivity and glucose tolerance (Fig. S1B). In EpiWAT, we found increased expression of F4/80 and MCP1 indicating increased macrophage recruitment, and increased expression of IRF5 (Fig. S1C). We carried out cytometric analyses in the EpiWAT stromal vascular fraction (SVF) and con rmed an increase in F4/80 + CD11b + macrophage numbers, also histologically indicated by the presence of crown-like structures following HFD (Fig. 1A). In F4/80 + CD11b + cells, we con rmed an increase in IRF5 expression and M1-like marker CD11c expression, with no difference in M2-like marker CD206 (Fig. 1B). Assessing metabolic substrates, we used BODIPY, a uorescent lipid dye, and 2-NBDG to assess glucose uptake. Upon short-term HFD, ATMs have increased intracellular lipids and increased glucose uptake relative to mice on NCD (Fig. 1C), as previously reported on longer term HFD 11 . Next, we analysed mitochondrial mass and activity with the uorescent JC-1 dye which is sensitive to mitochondrial membrane potential 26 . We found trends to increased mitochondrial mass, decreased activity and decreased activity-to-mass ratio in ATMs from mice on HFD (Fig. S1D).
Correlative analyses revealed that EpiWAT ATM numbers positively correlated to IRF5 expression in ATMs (Fig. 1D). Macrophage glucose uptake and mitochondrial mass positively correlated to IRF5 expression, whereas activity-to-mass ratio was negatively correlated to IRF5 expression (Fig. 1E). As macrophage numbers and IRF5 expression were positively correlated, IRF5 may disproportionately impact in ltrating macrophages. We analysed F4/80 Hi ATMs, previously characterised as monocyte-derived in ltrating macrophages 10,27 , and found that F4/80 Hi ATMs had higher expression of IRF5 on HFD relative to NCD, with increased lipid content and glucose uptake (Fig. 1F, S1E). Interestingly, F4/80 Hi ATMs also had increased mitochondrial mass and decreased mitochondrial activity upon HFD relative to NCD (Fig. 1G). These data suggest that IRF5 plays a role in driving ATM in ltration and metabolic adaptation, including alterations in mitochondrial mass or activity on short-term HFD.
Myeloid-de ciency of IRF5 alters ATM metabolic adaptation to caloric excess To discern whether ATM metabolic adaptation associated with IRF5 expression are dependent on IRF5, we applied the same model of short-term high-fat feeding to mice with a myeloid-de ciency of IRF5 (IRF5-KO) and their wild-type (WT) littermates (Fig. 1H). Despite no difference in systemic glucose homeostasis, body weight nor adipose tissue weight, we observed increased mitochondrial activity and activity-to-mass ratio in ATMs from IRF5-KO relative to WT mice (Fig. 1I, S1G). This was accompanied by a trend to increased lipid content in IRF5-KO but no difference in CD11c expression nor in glucose uptake (Fig. 1J). Unsupervised tSNE analyses con rmed that JC1-red uorescence, indicating mitochondrial activity, was highest in F4/80 Hi ATMs and that these ATMs had higher uorescence in IRF5-KO (Fig. 1K). To determine whether these effects on macrophage metabolism are re ected at the transcriptomic level or are timedependent, we sorted F4/80 + cells from epididymal fat pads of IRF5-KO and WT mice upon HFD for 4 and 12 weeks. Differential analyses between genotypes and GO term enrichment revealed that terms related to cellular metabolism were over-enriched, whilst those related to immune effector functions were underenriched at 4 weeks, comforting our in vivo and cellular results showing that metabolic pathways are altered in an IRF5-dependent manner at 4 weeks of HFD (Fig. 1L). On long-term HFD, we previously reported that IRF5-de ciency results in a protected systemic metabolic phenotype due to bene cial adipose tissue remodelling at 12 weeks of HFD 16 . As expected, differentially expressed genes at 12 weeks enriched terms relating to tissue remodelling and in ammatory processes. Additionally, the Positive Regulation of Metabolic Process term was enriched indicating that IRF5-dependent metabolic adaptation may be sustained over time (Fig. 1L). Gene set enrichment analysis (GSEA) on GO terms relating to the TCA Cycle, Glycolytic Process and Regulation of the Type 1 Interferon Response con rmed that genes related to cellular metabolism were disproportionately impacted by IRF5 de ciency (Fig. 1M). These ndings indicate that IRF5-KO alters ATM metabolic adaptation to caloric excess and this is represented at the transcriptional level.
IRF5-de ciency is associated with increased oxidative respiration and fatty acid oxidation in the SVF of epididymal WAT upon caloric excess To functionally characterise cellular respiration, we carried out metabolic ux analyses on the SVF of EpiWAT from IRF5-KO and WT mice following 4 weeks of HFD. Under conditions favouring mitochondrial respiration, EpiWAT SVF, largely composed of ATMs, had increased oxidative respiration in IRF5-KO relative to WT ( Fig. 2A). This was also re ected by an increased area under the curve (AUC) (Fig. 2B). Interestingly, when evaluating oxygen consumption under conditions favouring glycolysis, we also found oxidative respiration to be increased in IRF5-KO under both basal and glucose-stimulated conditions (Fig.  2C). The energetic plot of basal and maximal respiration con rms the hyperoxidative nature of SVF from IRF5-KO relative to WT (Fig. 2D). To discern whether the increased oxidative respiration in IRF5-KO could be attributed to fatty acid oxidation (FAO), we carried out a palmitate (Palm) oxidation test with or without etomoxir (ETO), an inhibitor of carnitine palmitoyl transferase 1 (CPT1), the enzyme allowing palmitoyl-CoA catabolism 28 . When treated with Palm, oxygen consumption was higher in cells from IRF5-KO mice relative to WT mice, but the difference between the genotypes was normalised when cells were treated with ETO (Fig. 2E, S2A). This indicates that CPT1-dependent FAO contributes to the higher mitochondrial oxidative respiration in SVF from IRF5-KO mice relative to WT mice. Following these ndings in the SVF, we investigated the metabolic effects of IRF5-de ciency in bone marrow-derived macrophages (BMDMs) from IRF5-KO and WT mice. BMDMs were treated with bacterial lipopolysaccharides (LPS), a known stimulant of IRF5 expression, and with Palm to replicate lipotoxicity. We applied metabolic ux analyses and found no genotype-dependent difference in extracellular acidi cation rate in response to LPS or Palm, nor under basal conditions when testing glycolysis (Fig.  S2B). In the same assay, oxygen consumption after glucose injection was increased in IRF5-KO relative to WT BMDMs following treatment with LPS or Palm. This indicates that mitochondrial respiration, aerobic glycolysis, or both, preferentially take place in IRF5-de cient macrophages (Fig. 2F, S2C). Carrying out a test of mitochondrial respiration, we observed no difference in untreated cells (Fig. S2D). Yet mitochondrial respiration was increased in IRF5-KO relative to WT BMDMs following treatment with LPS or Palm (Fig. 2G). The metabolic phenotype of BMDMs mirrors that seen in the SVF following 4 weeks of HFD ( Fig. 2A), indicating that ATMs may be the main drivers of the difference in respiration seen in EpiWAT SVF from IRF5-KO and WT mice. The energetic plot from these experiments indicated that whilst extracellular acidi cation rate was similar between genotypes, oxygen consumption was consistently higher in IRF5-KO relative to WT BMDMs (Fig. 2H). The switch in metabolic adaption in IRF5-KO coincided with lower concentrations of IL6 and TNF, and higher concentrations of IL10 secreted into cell culture media (Fig. 2I), con rming that IRF5-KO BMDMs are more prone to alternative polarisation, as previously described 16,20,29 . Complementary to these experiments, we applied an inhibitory IRF5 decoy peptide (IRF5-DP) to mature IRF5-competent BMDMs. IRF5-DP binds to IRF5 in the cytoplasm, preventing its nuclear translocation 30 . LPS-induced upregulation of TNF is prevented by IRF5-DP, con rming that IRF5-DP blocks the transcriptional activity of IRF5 (Fig. 2J). When treated with Palm, oxygen consumption was increased with IRF5-DP relative to vehicle, which was re ected by increased oxygen consumption rate (OCR) at maximal respiration and increased AUC (Fig. 2K, S2E). The energetic plot re ects increased oxidative respiration in response to Palm and inducible by IRF5-DP (Fig. 2L). The effects of IRF5-DP delivery indicate a mechanism dependent on nuclear translocation and transcriptional activity and rule out a differentiation effect of IRF5-de ciency that may result in altered cellular respiration. We next used adenoviral delivery to re-introduce IRF5 expression in BMDMs from IRF5-KO mice. Treatment with an IRF5 adenovirus (adIRF5) resulted in a 1.4-fold increase in IRF5 expression (Fig. S2F) compared to the control adenovirus (adGFP). When metabolic ux was analysed following Palm treatment, OCR was decreased in cells treated with adIRF5, but not in cells treated with adGFP ( Fig. 2M, S2G). To further link IRF5 to altered mitochondrial function, we ectopically overexpressed IRF5 in primary brown adipocytes as cells rich in mitochondria without an overt immune function. Adenoviral delivery of IRF5 in brown adipocytes resulted in a 5-fold increase in IRF5 expression relative to samples treated with adGFP (Fig. S2H). AdIRF5 treated cells were irresponsive to oligomycin and FCCP, had increased proton leak, and decreased ATP production and spare capacity (Fig. 2N, S2I). In adIRF5 treated cells, mitochondrial machinery is less able to respond to increased energy demand and respiration was no longer coupled to ATP production.
Altered destination of glycolysis end-products and accumulation of alpha-hydroxyglutarate characterise the TCA cycle in IRF5-de cient macrophages Evaluating elements of mitochondrial respiration, we carried out targeted quantitative metabolomics of the tricarboxylic acid (TCA) cycle intermediates in IRF5-KO and WT BMDMs treated for 2 h or 24 h with Palm or with LPS. Global correlative analyses of metabolite concentrations resulted in a strong positive correlation (p<0.0001; R 2 =0.42) between genotypes and indicated overall higher concentrations of TCA metabolites in IRF5-KO BMDMs (Fig. 3A). PCA score plot obtained from metabolite concentrations revealed genotype-dependent separation of the TCA cycle metabolite pro le with Palm treatment for 2 h and following LPS treatment for 24 h (Fig. 3B, S3A). Variable ranking per principal component revealed that alpha-hydroxyglutarate (AHG) was the highest contributor to the IRF5-dependent response to 2 h Palm (Fig. 3C). Accordingly, AHG concentration was increased in IRF5-KO BMDMs (Fig. 3D). Surprisingly little is known about AHG in macrophage metabolism. Previous reports link this metabolite to inhibition of alpha-ketoglutarate (AKG) dehydrogenase 31 , which is supported by our ndings of reciprocal regulation of AKG and accumulation of its precursor glutamine in IRF5-KO. As re ected by PCA plots following 2 h of treatment, IRF5-dependent remodelling of the TCA metabolite landscape is a speci c and early response to Palm. In response to LPS, most metabolites in IRF5-KO cells are normalised to WT levels (Fig. S3B). Taking TCA metabolite pro le into consideration with correlative analyses per metabolite (Fig. 3E), IRF5-de ciency seems to modify the destination of glycolysis end-products (lactate/pyruvate) as well as AHG or glutamine to AKG conversion (Fig. 3F). Despite considerable remodelling of TCA cycle intermediates in IRF5-KO cells, this alone cannot explain increased mitochondrial respiration nor the in ammatory pro le of IRF5-KO cells. Thus, we characterised structural aspects of mitochondria that may alter function. We rst evaluated mitochondrial density through electron microscopy and found no genotype-dependent differences (Fig. 3G, S3C). We also observed no difference in form factor nor aspect ratio, proxies of mitochondrial dynamics 32 (Fig. S3D). However, analysing membrane associations with mitochondria revealed that IRF5-KO BMDMs are characterised by lower proportions of distant (> 30 nm) mitochondria-endoplasmic reticulum contacts (MERC), relative to WT BMDMs upon treatment with Palm ( Fig. 3H, S3E). Taken together, these data indicate that IRF5 plays a role in maintaining normal TCA cycle responses to Palm or LPS. Frequency of distant MERCs is also decreased in IRF5-KO relative to WT in response to Palm. However, these ndings cannot explain the increased mitochondrial respiration in IRF5-KO BMDMs in response to metabolic stress.

IRF5 binds and regulates mitochondrial matrix protein GHITM in macrophages
To investigate potential mechanisms, we rst made use of previously published IRF5 ChIP-seq data 33 . This dataset originates from a chromatin immunoprecipitation of IRF5 in BMDMs treated with LPS for 30 and 120 minutes. At 30 and 120 minutes of LPS stimulation, we mapped 3404 and 526 genes with IRF5 binding, respectively. Functional classi cation revealed that 2.3% and 5.1% of bound genes, at 30 and 120 minutes of LPS treatment, are part of the Immune System Process GO term; whereas 17.8% and 14.6% were in the Metabolic Process GO term (Fig. 4A). These results indicate that IRF5 directly binds to genes that control metabolic as well as immune processes in macrophages.
Whilst the above indicated that IRF5 binds to genes that can impact cellular metabolism upon LPS treatment, our current data indicated a time-dependent and divergent metabolic response to LPS and to Palm under the control of IRF5. To investigate these responses at the transcriptional level, we carried out RNA-seq on IRF5-KO and WT BMDMs following 2 h or 24 h treatment with LPS or with Palm. To ensure in vivo relevance of these analyses, we also included our RNA-seq analyses on F4/80 + ATMs sorted from EpiWAT of IRF5-KO and WT mice on HFD for 4 weeks and 12 weeks (Fig. 1L). To genes ltered on differential expression, we applied clustering analyses to de ne 6 clusters based on expression trajectory over time (Fig. 4B). Of the 6 clusters de ned, 4 were differentially regulated between IRF5-KO and WT BMDMs upon LPS or Palm treatment; whereas upon HFD, 3 clusters were differentially regulated between genotypes (Fig. 4B). GO term enrichment of differentially regulated clusters indicated several clusters related to the regulation of transcription, cellular metabolic processes as well as myeloid cell differentiation and tissue remodelling (Fig. S4A). Differentially regulated clusters upon Palm treatment related to response to lipids, induction of immune tolerance as well as mitochondrial respiration, these were also enriched in ATMs upon HFD. Interestingly, correlative analyses of gene expression also found that Palm treatment resulted in the most divergent IRF5-dependent response (Fig. S4B). To de ne target genes differentially regulated upon in ammatory, lipotoxic and in vivo metabolic stress, we carried out differential expression analyses between genotypes on the different models. Importantly, Palm treatment resulted in the highest number of differentially expressed genes between IRF5-KO and WT cells, followed by LPS and HFD (Fig. 4C). One percent of differentially expressed genes were represented in all conditions (Fig. 4D), and integrating these data with the previously reported IRF5 ChIP-seq allowed us to identify a subset of 6 genes that are differentially regulated between IRF5-KO and WT cells and are also bound by IRF5 at, or upstream of, their transcription start sites (ATF5, SYCE2, ABCG1, LRRC27, FNIP2, GHITM) (Fig. 4E). Of these genes, GHITM has been reported to have an overt function in maintaining the mitochondrial network as well as inner membrane cristae 34 . Interestingly, correlative analyses in IRF5competent cells revealed that GHITM was negatively correlated with IRF5 expression in ATMs (Fig. 4F). These data indicate that IRF5 controls the expression of GHITM in macrophages and this may be important in mediating repression of mitochondrial respiration. In line with the above results, we quanti ed the number and length of cristae per mitochondrion in ultra-structural analysis of IRF5-KO and WT BMDMs following Palm treatment and found indeed that both number and length of cristae per mitochondrion were increased in IRF5-KO BMDMs relative to WT (Fig. 4G). These data support that IRF5 may transcriptionally repress GHITM expression, and function, upon metabolic stress.
GHITM is highly expressed in monocytes and ATMs and its silencing reverses highly oxidative respiration in IRF5-de ciency We procured previously published single cell RNA-seq data from EpiWAT SVF of mice on HFD for 6 and 12 weeks 1 . Single cell sequencing data con rmed that IRF5 and GHITM are highly expressed in monocytes as well as in macrophages from EpiWAT SVF, expression levels were higher than all other targets identi ed (Fig. 5A, 5B, S5A). Correlative analyses revealed that GHITM expression was negatively correlated to IRF5 expression at the single cell level (R= -0.44; p<0.001) (Fig. 5C). IRF5 expression was highest in monocytes, with its expression increasing over time. The negative correlation with GHITM was also sustained over time (Fig. 5B, 5D). Our analyses of this dataset, with our own data, suggests that that an interaction between IRF5 and GHITM may impact cellular respiration and metabolic adaptation of macrophages in diet-induced obesity. Thus, with guide RNAs (gRNA) targeting GHITM (gGHITM), we transduced BMDMs from mice with myeloid-restricted expression of the clustered regularly interspaced short palindromic repeats (CRISPR)-associated protein 9 (Cas9) linked to EGFP. We rst con rmed Cas9-EGFP expression in mature BMDMs (Fig. S5B). Transfection with gGHITM resulted in a 40 % decrease in GHITM expression relative to controls (Fig S5C). We subjected this model to Palm treatment, where we con rmed decreased GHITM expression in response to Palm and upon transfection, by both qPCR and Western blotting (Fig. 5E, S5D). We also con rmed increased TNF expression in gGHITM transfected cells and upon Palm treatment (Fig. 5F). Metabolic ux analyses revealed that gGHITM transfection resulted in decreased oxidative respiration relative to controls, both under basal conditions and following Palm treatment, this effect was most pronounced at maximal respiration (Fig. 5G). To determine whether GHITM contributes to increased oxidative respiration in IRF5-de cient macrophages following lipotoxicity (Fig. 2G), we carried a double transfection with a gRNA targeting IRF5 (gIRF5) and with gGHITM. We rst con rmed e cient knockdown of targeted genes, where gIRF5 resulted in a 40 % knockdown of IRF5 when used alone or simultaneously with gGHITM, which itself lowered GHITM expression to 30% (Fig.  5H). This model was treated with Palm and subjected to metabolic ux analysis. Transfection with gIRF5 alone increased OCR relative to controls, con rming our ndings from IRF5-KO and WT mice and from treatment with IRF5-DP (Fig. 5I, 2G, 2K, S2E). This increase is normalised to control levels in cells doubly transfected with gGHITM and gIRF5 (Fig. 5I). Taken together, these results indicate that GHITM contributes to increased oxidative respiration in IRF5-de cient macrophages.
IRF5 expression is associated with decreased GHITM expression and altered human monocyte and ATM metabolic phenotype When we observed that IRF5 and GHITM were highly expressed in monocytes as well as macrophages, we sought to validate our results in monocytes from patients with T2D. We sorted CD14 + monocytes based on their expression of IRF5, and the sorted cells were subjected to RNA-seq (Fig. 6A). Paired differential expression analysis of IRF5 + versus IRF5negative monocytes revealed 3211 genes upregulated in IRF5 + monocytes versus 295 genes downregulated. GO term-enrichment of each gene set found that upregulated genes enriched terms relating to cell adhesion, development and motility (Fig. 6B).
Interestingly, terms enriched amongst downregulated genes were related to mitochondrial organisation and protein localisation to mitochondria (Fig. 6B). Moreover, GHITM was consistently downregulated in IRF5 + relative to IRF5monocytes (Fig. 6C). Next, we stained cytospin prepared monocytes, from patients with T2D, with IRF5 and with an antibody against the oxidative phosphorylation (OXPHOS) enzyme complexes, known to be anchored to the mitochondrial cristae structures maintained by GHITM (Fig. 6D). We found that monocytes with primarily nuclear localisation of IRF5 (Nuc) had lower OXPHOS enzyme staining density relative to monocytes with cytoplasmic IRF5 staining (Cyt). Investigating adipose tissue, we quanti ed IRF5 and GHITM mRNA expression in CD14 + cells sorted from human WAT biopsies from obese patients. We separated patients based on IRF5 mRNA expression level and found that IRF5 Hi expressers have a lower expression of GHITM than IRF5 Lo expressers (Fig. 6E). Following these additional lines of evidence that IRF5 and GHITM share the same counter-regulation in human monocytes and ATMs as in mice, we sought to verify that IRF5 can indeed bind to GHITM in human macrophages as it does in murine macrophages (Fig. 4G). We used the University of California Santa Cruz (UCSC) genome browser to visualise IRF5 binding regions on or around the GHITM gene (Fig. 6F). Indeed, IRF5 can bind to several regions on and upstream of the GHITM gene. Furthermore, mRNA expression of GHITM is decreased in LPS-treated human monocyte-derived macrophages (HMDM) relative to untreated HMDMs; and this coincides with a decrease in active transcription histone mark H3K27ac in LPS-treated HMDMs relative to untreated HMDMs. The above analyses show that IFR5 and GHITM are expressed in human monocytes and macrophages, are reciprocally regulated, and that IRF5 can also bind to the GHITM gene. We next sought to functionally validate our hypothesis that IRF5 expression is associated with functional differences in cellular respiration in human monocytes and ATMs. We carried out ow cytometric analyses on human venous blood and AT SVF to quantify IRF5 expression in CD14 + monocytes and macrophages, and to analyse their metabolic adaptation. Correlative analyses indicated that ATM expression of IRF5 tends to negatively correlate with cellular lipid content, mitochondrial mass, activity, and activity-to-mass ratio, trends for lipid content and mitochondrial activity-to-mass ratio are also found in monocytes (Fig. 6G). We next quanti ed lipid content in IRF5 + and IRF5 -ATMs and found, similar to in vivo results, that IRF5 + ATMs were characterised by a lower lipid content relative to IRF5 -ATMs (Fig. 6H). We plotted monocyte and ATM populations onto an IRF5 and BODIPY quadrant, analyses con rmed that IRF5 + cells are characterised by a restricted lipid pool and the IRF5 + BODIPY + population has the lowest frequency in both monocytes and ATMs (Fig. 6I). Taken together these ndings indicate that IRF5 can bind to and transcriptionally regulate GHITM expression under metabolic stress in human monocytes and macrophages. This regulation has functional consequences on mitochondrial activity and on cellular lipid content which may be a key substrate for mitochondrial metabolism.

Discussion
Visceral WAT is a key responder to caloric excess. Adaptive responses dictate disease course in metabolic syndrome and a major determinant of tissue adaptation is the phenotype and function of ATMs. ATMs are a heterogenous population of cells ranging from regulatory to highly in ammatory, the latter being a key contributor to systemic metabolic decline in obesity and T2D. As sentinel cells playing a homeostatic role in normal physiology, the molecular mechanisms of ATM adaptation to early caloric excess remain to be fully understood. Here we demonstrate that ATMs undergo extensive IRF5-dependent metabolic rewiring upon short-term caloric excess. ATM oxidative capacity is limited by IRF5-mediated transcriptional repression of GHITM, the inner mitochondrial membrane protein that maintains mitochondrial structure for e cient oxidative respiration. Decreased expression of GHITM and loss of cristae organisation bias macrophage metabolism away from mitochondrial respiration to promote in ammatory polarisation.
Previous studies show that in ammation arises in WAT and is mediated by ATMs. The key implication of IRF5 in metabolic in ammation and T2D has been demonstrated in both mice and humans 16 . Gain-offunction IRF5 risk-variants have also been associated with increasing macrophage glycolytic ux 25 , a cellular process that supports in ammatory effector functions. A recent study has also demonstrated that IRF5 regulates airway macrophage metabolic response to viral infection 35 . Whilst we have previously reported the in ammatory responsiveness of IRF5 in ATMs in metabolic disease, the role that this transcription factor plays in adapting ATM metabolism remains to be investigated. First, we investigated the potential role of IRF5 in regulating cellular metabolic adaptation through a model of short term highfat feeding. A study by Lee et al. 36 demonstrated that immunocompromised mice developed insulin resistance upon short-term high-fat feeding, indicating that in ammation is not required for insulin resistance at this stage. A further study by Shimobayashi et al. 37 con rmed this nding, demonstrating that WAT was disproportionately affected and that early insulin resistance precedes in ammation. In our ndings, we con rm that upon 4 weeks of high-fat feeding mice develop altered glucose homeostasis, whilst accumulating macrophages that are metabolically different to ATMs from mice on a NCD.
Reports in immunometabolism have characterised macrophage metabolic responses to infectious pathogens 38 . Recent studies demonstrate that tissue macrophages have their own metabolic speci cities in-line with microenvironmental pressures 39,40 . Moreover, in T2D, a systemic abundance of metabolic substrates in the form of hyperglycaemia and dyslipidaemia presents its own challenges. In the case of ATMs, they reside in a lipid-rich environment and take on an overall hypermetabolic phenotype upon diet-induced obesity, increasing glycolysis as well as mitochondrial respiration 10 . More recent studies report speci c LAM expansion within 6 weeks of HFD, with LAMs being metabolically protective 1 . Similarly, CD11c + , CD206 + double positive macrophages expressing the scavenger receptor MARCO are highly oxidative and expand in short-term HFD 41 . Whilst these two populations of ATMs are highly oxidative and represent physiological adaptation to caloric excess, the predominance of glycolysis has been associated with supporting ATM in ammatory effector function 11 . In the current study, we demonstrate that IRF5de cient macrophages have a higher rate of mitochondrial respiration relative to IRF5-competent macrophages. This indicates that IRF5 plays an important role in tilting the metabolic balance in ATMs towards glycolysis by impairing mitochondrial respiration, thus promoting metabolism that supports in ammation. Analysing the ways in which mitochondria are affected in IRF5-de ciency, we found that TCA metabolites are modi ed in IRF5-KO relative to WT macrophages upon stimulation. We initially expected typical perturbations to the TCA cycle in M1-like macrophages to be reversed in IRF5-KO.
Notably, for itaconate and succinate to be depleted, as their accumulation is a hallmark of M1-like metabolism 42 . Yet, upon LPS stimulation, we were surprised to nd that only succinate was decreased in IRF5-KO. Typically, succinate accumulation in M1-like macrophages is a result of the inhibition of succinate dehydrogenase by itaconate 43 , yet we do not observe such co-regulation in our study. Indicating that IRF5-dependent mechanisms may be circumventing the characterised itaconate-succinate axis in macrophage polarisation. Some hallmarks of the IRF5-KO BMDM TCA cycle reproduce previous reports on TCA dynamics in the M2-like state. IRF5-KO BMDMs accumulate lactate in preference to pyruvate conversion at the end of glycolysis, this has been previously reported as characteristic of M2 macrophages in the tumour microenvironment [44][45][46][47] . In IRF5-KO BMDMs decreased pyruvate means it is less available as a substrate for mitochondrial metabolism. This is coherent with our ex vivo and human observations that IRF5-KO or IRF5 Lo macrophages have a higher lipid content, which fuels mitochondrial FAO. The highest-ranking metabolite driving the IRF5-dependent response to Palm was AHG. AHG is a relatively little-known metabolite, it is produced as a result of non-canonical enzyme function at low pH (possibly due to increased lactate in IRF5-KO BMDMs) 48 . AHG has been reported to accumulate in IL4primed macrophages and is also found as an oncometabolite produced from mutated tumour-derived isocitrate dehydrogenase 31,49,50 . AHG is a competitive inhibitor of AKG-dependent deoxygenases, occupying the same active sites on AKG 31,48,49 . This action of AHG results in genome-wide histone and DNA methylation alterations. This mechanism is supported by our data as we do observe a concomitant decrease in AKG as well as an accumulation of glutamine in IRF5-KO. With the TCA metabolite landscape being altered, but not fully explaining the metabolic phenotype of IRF5-de cient macrophages, we carried out structural analyses of mitochondria in search of a structural basis for increased mitochondrial metabolism. With no difference in mitochondrial density nor dynamics, we investigated membrane interaction with the endoplasmic reticulum. MERCs play diverse roles, including in metabolic sensing, and lipid and calcium ion transport [51][52][53] . MERCs have been extensively studied in other cell types, and only to a limited extent in macrophages 54 . Studies in broblasts show that narrower MERCs are associated with more e cient calcium ion and lipid exchange with mitochondria; whilst increasing MERC distance is associated with dormancy and loss of lipid homeostasis 51,55 . The proportion of distant MERCs (> 30 nm) was decreased in IRF5-KO relative to WT macrophages upon stimulation with Palm. However, the stability of narrower associations indicates that increased respiration is independent of MERC-mediated substrate transfer in our model.
In a data-driven approach, we combined public datasets with our own RNA-seq data to reveal a previously undescribed IRF5-dependent mechanism that impairs macrophage mitochondrial respiration. Impaired mitochondrial respiration is, at least in part, attributed to decreased expression of GHITM. GHITM was initially characterised as an inner mitochondrial membrane protein required to maintain cristae structures, the anchoring points for OXPHOS enzyme complexes 34 . Previous studies found that GHITM downregulation causes mitochondrial fragmentation and cristae disorganisation 34 . Mitochondrial fragmentation has previously been associated with in ammatory polarisation, both in response to LPS and to fatty acids 56, 57 . Studies in lymphocyte cell lines stimulated with in ammatory cytokines and from virus-exposed monocytes also report downregulation of GHITM 58,59 . These previous reports are in line with our current ndings that loss of GHITM is associated with in ammation. We additionally report a role for GHITM in macrophages and in ATMs in both humans and mice in response to metabolic stress.
In summary, we have deciphered a novel mechanism through which IRF5, a well characterised proin ammatory transcription factor, alters cellular mitochondrial dynamics and respiration. Having identi ed this IRF5-dependent mechanism to control cellular metabolism, a number of questions remain unanswered. For example, to elucidate how and through which regulatory elements IRF5 may be binding to such non-in ammatory targets as GHITM. Whilst it is widely accepted that IRF5, and other IRFs, target interferon sensitive regulatory elements (ISRE), the presence and signi cance of ISREs on metabolically relevant genes remains to be investigated. Furthermore, the speci c functional contribution of GHITM downregulation to effective in ammation is unknown, for example consequent mitochondrial fragmentation may be a source of reactive oxygen species required for bacterial killing. Similarly, whilst we observe increased oxidative respiration in IRF5-de ciency, we also observe an increase in the intracellular lipid pool, the signi cance and composition of this lipid pool remains to be understood, future investigations can decipher whether the interferon pathway can also act on cellular lipid dynamics.
Lastly, despite several lines of evidence implicating IRF5 in metabolic decline associated with dietinduced obesity, the metabolic stressors that induce IRF5 expression remain unknown. Future work on the above questions will be of important mechanistic insight into how this in uential pathway can be modulated in diabetes and in other in ammatory conditions.

Human samples
Blood samples and adipose tissue biopsies were obtained from different populations admitted to the Lariboisière and Geoffroy Saint Hilaire hospitals (Paris, France). Studies were conducted in accordance with the Helsinki Declaration and were registered to a public trial registry (Clinicaltrials.gov; NCT02671864). The Ethics Committee of CPP Ile-de-France approved the clinical investigations for all individuals, and written informed consent was obtained from all individuals. The principal investigator of this clinical trial is Prof. Jean-François Gautier: jean-francois.gautier@aphp.fr. Adipose tissue biopsies were obtained from obese subjects during bariatric surgery.

Experimental animals and In vivo studies
Male C57BL/6J mice (5-7 weeks) were purchased from Charles River. To generate mice with a myeloidspeci c deletion of IRF5, IRF5 ox/ ox mice (C57BL/6-Irf5tm1Ppr/J; stock no. 017311) were crossed with For oral glucose tolerance test (GTT), mice were fasted overnight before being gavaged with glucose (2g/kg of body weight). Tail vein blood was collected to measure glycaemia with a glucometer (Verio, One touch). For insulin tolerance test (ITT), mice were fasted during 5 h before being i.p injected with insulin (0,5U/kg of body weight). For pyruvate tolerance test (PTT), mice were fasted overnight before being i.p. injected with pyruvate (2g/kg of body weight). Glycaemia was monitored for 120 min after insulin or pyruvate injection.

Organ collection and histology
Mice were sacri ced by cervical dislocation. Upon dissection, tissues were weighed. Immediately after collection, samples were either digested with collagenase, snap-frozen for further analysis or drop-xed into 10% formalin (Sigma Aldrich) for 24 h for histological analysis. For histological analysis, tissues were processed for dehydration, clearing and para n embedding with an automated carousel (Leica).
Sections (6µM thick) were stained with hematoxylin and eosin according to standard procedures. Images were acquired with a slide scanner (Zeiss Axio Scan Z1).
Acquisition was performed on a MACSQuant cytometer (Miltenyi Biotech). Cell sorting was performed on a FACSAria III (BD Biosciences). Cells were directly sorted in RLT lysis buffer supplemented with βmercaptoethanol for RNA extraction (Qiagen). Data were analysed with FlowJo software (Tree Star).
Cells from the previously isolated SVF were stained for immunoselection of F4/80 + cells according to manufacturer's protocol (MACS, Miltenyi Biotec). Cells were resuspended in MACS buffer (1X PBS supplemented with 0.5% BSA and 2mM EDTA) containing the appropriate dilution of anti-F4/80 microbeads (130-110-443, Miltenyi Biotec), for 10 min at 4°C. Automated magnetic cell separation was performed with the Multi-MACS Cell Separator. Once collected, the F4/80 + cell fraction was washed and directly resuspended in RLT lysis buffer supplemented with β-mercaptoethanol for RNA extraction (Qiagen).

In vitro macrophage studies
Bone marrow-derived macrophages: Murine bone marrow cells were isolated from femurs and tibias. Cells were plated in DMEM (Gibco) supplemented with 10% FBS, 1% P/S and 30% L929 conditionedmedia and were allowed to differentiate for 8-10 days into bone marrow-derived macrophages.
Decoy peptide: Fully differentiated BMDMs were pre-treated with an IRF5 decoy peptide 30 (50µg/ml) for 30 min, before being treated for further analysis. Cytokine quanti cation: Pro-in ammatory cytokines in the cell culture supernatants were quanti ed with the Mouse In ammation Kit (552364, BD Cytometric Bead Array) according to manufacturer's protocol.
Quantitative PCR with reverse transcription RNA was extracted from cells or tissue using RNeasy Plus Mini or Micro kit (Qiagen). Complementary DNA was synthesized with M-MLV Reverse Transcriptase kit (Promega). SYBR Green qRT-PCR reactions were performed with MESA green MasterMix (Eurogentec) and sequence-speci c primers (Table S3), using QuantStudio 3 Real-Time PCR Systems (ThermoFisher Scienti c). 18S was used for normalization to quantify relative mRNA expression levels.

Electron microscopy and structural analyses
BMDMs were scraped and xed in 2 % glutaraldehyde for 2 h at 4°C, post xed in 1 % Osmium tetroxide for 1 h at 4°C, dehydrated, and embedded in Epon. Samples were then cut using an RMC/MTX ultramicrotome (Elexience), and ultrathin sections (60-80 nm) were mounted on copper grids, contrasted with 8% uranyl acetate and lead citrate, and observed with a Jeol 1200 EX transmission electron microscope (Jeol LTD) equipped with a MegaView II high-resolution transmission electron microscopy camera. Pictures of cells sections were taken at 45000 × magni cation. For MERC analyses, mitochondria and ER limits were traced on ImageJ® (Fiji). The fraction of mitochondrial membrane in contact with ER within a 50-nm range was calculated and normalized to the mitochondrial perimeter and expressed as the total percent of contact according to different ranges of distance between both organelles 60 . Mitochondria number per section was also measured to evaluated mitochondria density.
For cristae analysis, mitochondria and cristae were outlined using ImageJ® (Fiji) and both the total length and number of cristae in each mitochondrion was calculated, as previously described 61 . For the analysis of mitochondria dynamics, the long and short axis of each mitochondria, as well as their perimeter and area, were measured. From these values, aspect ratio (major axis/minor axis) and form factor (perimeter)²/(4xpixArea) were calculated. TEM analyses were performed in triplicate and a minimum of 11 images per sample were taken.
Quanti cation of TCA metabolites by liquid chromatography coupled to high resolution mass spectrometry (LC-HRMS) Metabolite extraction. A volume of 170 μL of ultrapure water was added to the frozen cell pellets. At this step, 20µL of each sample were withdrawn for further determining the total protein concentration LC-HRMS analysis. Targeted LC-HRMS experiments were performed using an U3000 liquid chromatography system coupled to a Q Exactive Plus mass spectrometer (ThermoFisher Scienti c). The software interface was Xcalibur (version 2.1) (ThermoFisher Scienti c). The mass spectrometer was externally calibrated before each analysis in ESI-polarity using the manufacturer's prede ned methods and recommended calibration mixture. The LC separation was performed on a Sequant ZIC-pHILIC 5 µm, 2.1 x 150 mm column (HILIC) maintained at 45°C (Merck, Darmstadt, Germany). Mobile phase A consisted of an aqueous buffer of 10 mM of ammonium acetate, and mobile phase B of 100% acetonitrile. Chromatographic elution was achieved with a ow rate of 200 µL/min. After injection of 10 µL of sample, elution started with an isocratic step of 2 min at 70% B, followed by a linear gradient from 70 to 40% of phase B from 2 to 7 min. The chromatographic system was then rinsed for 5 min at 0% phase B, and the run was ended with an equilibration step of 9 min. The column e uent was directly introduced into the electrospray source of the mass spectrometer, and analyses were performed in the negative ion mode. The Q Exactive Plus mass spectrometer was operated with capillary voltage set at −2.5 kV and a capillary temperature set at 350°C. The sheath gas pressure and the auxiliary gas pressure (nitrogen) were set at 60 and 10 arbitrary units, respectively. The detection was achieved from m/z 50 to Metabolomic data processing and quanti cation. Xcalibur software was used for peak detection and integration. Metabolite quanti cation was perform using calibration curves established from peak area ratios between metabolites and their respective internal standard. Each metabolite amount was normalised by the protein quantity measured in each sample by BCA assay.

RNA-sequencing of BMDMs and F4/80 + ATMs
After extraction, total RNA was analysed using Agilent RNA 6000 Pico Kit on the Agilent 2100 Bioanalyzer System. RNA quality was estimated based on capillary electrophoresis pro les using the RNA Integrity Number (RIN) and DV200 values. RNA sequencing libraries were prepared using the SMARTer Stranded Total RNA-Seq Kit v2 -Pico Input Mammalian (Clontech/Takara) from 10ng of total RNA. This protocol includes a rst step of RNA fragmentation using a proprietary fragmentation mix at 94°C. The time of incubation was set up for all samples at 4 min, based on the RNA quality, and according to the manufacturer's recommendations. After fragmentation, indexed cDNA synthesis and ampli cation were performed followed with a ribodepletion step using probes targeting mammalian rRNAs. PCR ampli cation was nally achieved on ribodepleted cDNAs, using 12 cycles estimated in accordance to the input quantity of total RNA. Library quanti cation and quality assessment were performed using Qubit uorometric assay (Invitrogen) with dsDNA HS (High Sensitivity) Assay Kit and LabChip GX Touch using The analyses were performed using the Eoulsan pipeline 66 , including read ltering, mapping, alignment ltering, read quanti cation, normalisation and differential analysis: Before mapping, poly N read tails were trimmed, reads ≤40 bases were removed, and reads with quality mean ≤30 were discarded. Reads were then aligned against the hg19 genome from Ensembl version 91 using STAR (version 2.5.2b) 62 .
Alignments from reads matching more than once on the reference genome were removed using Java version of samtools 67 . To compute gene expression, hg19 GTF genome annotation version 91 from Ensembl database was used. All overlapping regions between alignments and referenced exons were counted and aggregated by genes using HTSeq-count 0.5.3 68 . The sample counts were normalized using DESeq2 1.8.1 69 . Statistical treatments and differential analyses were also performed using DESeq2 1.8.1.

Statistics
Data analysis were performed using Microsoft Excel for Mac 16.47. Statistical analysis was performed using a two-tailed t-test for two groups, an ordinary one-way ANOVA followed by Tukey's multiplecomparisons test for multiple groups and a two-way ANOVA followed by Bonferroni's multiple comparison test on Prism 9 for macOS (GraphPad). Correlative analyses were performed on Prism 9 for macOS, computing Pearson coe cients for normally distributed data or Spearman coe cients for non-normally distributed data (GraphPad). PCA analyses were carried out on Prism 9 for macOS. Trajectoryresolved clustering was carried out on the Orange (v. 3.28.0) Python toolbox 70 . Statistical approaches per data panel are detailed in gure legends.

Public data
Single cell sequencing data: Single cell sequencing data from Jaitin et al. 1 , were downloaded and treated using BioTuring BBrowser (v. 2.7.48) 71 . Data were ltered in BBrowser and exported in tabular format for subsequent treatment with Microsoft Excel for Mac and Prism9 for macOS. UCSC genome browser: Gene tracks in Fig. 6E were visualised with the UCSC genome browser http://genome.ucsc.edu 72,73 , using the track hubs 74 . JASPAR2020 was used to visualise transcription factor binding sites 75 . The BLUEPRINT track-set was used for RNA expression and H3K27Ac lines 76,77 . Sample lines and tracks available through this session link / live link. The Human Dec. 2013 (GRCh38/hg38) assembly was used 78, 79 .

Data availability
Data that support ndings of this study are available from the corresponding authors upon request. RNAsequencing of IRF5 +/human monocytes: RNASeq gene expression data and raw fastq les are available on the GEO repository (www.ncbi.nlm.nih.gov/geo/) under accession number: GSEXXX. Source data are provided with this paper.

Competing interests
The authors declare no competing interests  Figure 1 Epididymal white adipose tissue macrophages undergo IRF5-dependent metabolic adaptation upon short-and long-term high-fat feeding. A. C57BL/6J mice were placed on a high-fat diet (HFD) or normal chow diet (NCD) for 4 weeks (n=10 per group), epididymal white adipose tissue (EpiWAT) was disrupted and F4/80+CD11b+ adipose tissue macrophages (ATMs) were quanti ed by ow cytometry (**p=0.0023 unpaired t-test). Haematoxylin and eosin (HE) staining of epidydimal fat pad at 5 x magni cation to