Immune privilege of adipocyte mitochondria protects from obesity


 Infant nutrition is rich in lipids, and the adipose tissue has been adapted to properly break down neutral lipids and oxidize fatty acids in infancy. Accordingly, infant adipose tissue contains so-called beige adipocytes, which burn off lipids to heat, and impede fat storage and obesity. We show here that infant adipocytes are immune privileged sites for mitochondria due to a blockade in interferon regulatory factor 7 (IRF7)-signaling, which allows mitochondrial RNA to trigger beige adipocyte differentiation through mitochondria-to-nucleus signaling. These mechanisms serve to maintain an extensive mitochondrial network in beige adipocytes and protect against obesity. By contrast, fat storing white adipocytes lack these mechanisms and respond to their mitochondrial content with inflammation. We show that obesity subverts the immune privilege for mitochondria in adipocytes, which reduces mitochondrial mass and abrogates beige adipocyte development. In turn, suppressing IRF7 signaling and restoring the RNA-mediated mitochondria-to-nucleus signaling in adipocytes effectively reduces obesity.

We next examined the expression of the STING/AIM2 pathways and IRF7 in the inguinal 104 adipose tissue (iAT) of human infants and children (0.3-6.9 years of age, N=26). Overweight 105 (BMI-SDS>1. 28) and obesity (BMI-SDS>1.88) strongly increased the expression of IFI16, ZBP1, 106 and IRF7, and moderately increased TMEM173 level (Fig. 1E, S6A-C), which was coherent with 107 the loss of beige adipocytes in childhood obesity (3,7,21). IFI16 protein level positively correlated 108 with adipocyte size (Fig. 1E). IRF7 and IFI16 expression was triggered by in vitro white 109 adipogenesis (Fig. 1E, Fig. S6D), and TMEM173 expression positively correlated with IFI16 and 110 IRF7 levels and was increased by premature loss of beige fat (Fig. 1E, Fig. S6D). We next extended 111 the age group of our analysis (7.0-11.0 years, N=73; 11.1-20.5 years, N=155) and found that in 112 lean subjects the STING/AIM2 pathways moderately increased with age, matching the time scale 113 of the physiological WAT expansion (Fig. S6E). 114 In summary, immune response to cytosolic mtDNA and mtRNA was lacking in P6 115 adipocytes and was dependent on IRF7 (Fig. 1F, Fig. S3D, Fig. S4C mitochondrial mass and led to inflammation in P6 adipocytes (Fig S7D-G). On the contrary, 120 cGAMP triggered IFN-response in P56 adipocytes (Fig. 1F, Fig. S3D). 121 STING stimulates IFN-response against mtDNA (22), however it is known that the STING 122 signaling may also induce autophagy (23,24). Mitophagy is a form of autophagy and protects the 123 cytosol from leaking mtDNA (25). Our data show that an autophagy-inducer effect of STING  Infant adipocytes employ mtRNA as a paracrine signal for beige fat development 128 We found that P6 adipocytes secreted mitochondrial contents in extracellular vesicles (EVs). 129 Adipocyte EVs were generated in the endosomal pathway, by inverse budding of endosomes, 130 leading to the formation of multivesicular bodies (MVBs) ( Fig. 2A,B; S8A-G). In line with this, 131 transcripts necessary for inverse budding of endosomes and the generation of MVBs were over- 132 represented in iAT at P6 (Fig. 2B). Inverse budding allows cytosolic nucleic acids to be delivered 133 to MVBs, and this process is a form of micro-autophagy (26). Endosomal content can be further 134 targeted for degradation in the lysosomes; however lysosomal genes were underrepresented in P6 135 iAT and by contrast, transcripts required for exocytosis were over-represented in P6 iAT (Fig. 2B). 136 P6 EVs were packed with mtDNA molecules and mitochondrial mRNA and rRNA species (Fig. 137 2C-E). Some of the EV cargo mRNAs, including Nd5, Co1 and Cytb, are known to generate non- 138 coding mtRNA species (27,28). The adipose tissue mesenchymal stem cell EV-specific 139 microRNA miR29a-5p was absent in P6 EVs (Fig. S8H). P6 EVs also contained minimal amounts  (29,30). Cytosolic single stranded RNA, or stimulation of 150 10 TLR3 did not mirror the effects of EVs (Fig. S10A-C), unlike the activation of RIG-I/MDA5 which 151 induced strong beige adipocyte gene transcription (Fig. 2I, S10D-H). 152 Beige-inducing effect of EVs was dependent on IL-6/STAT3 and RIG-I/MDA5 signaling 153 (Fig. 2J, Fig. S11A-E), and the lack of RIG-I or MDA5 led to the loss of beige adipocytes and 154 compromised mitobiogenesis, and compromised the expression of the nucleus-encoded 155 mitochondrial succinate dehydrogenase complex (Fig. 11F,G). 156 Nucleic acids in EVs are protected from extracellular nucleases by the surrounding 157 membrane and they may function as intercellular messengers (31). Accordingly, delivery of total 158 mtRNA into the cytosol induced beige adipocyte gene expression, mitobiogenesis and 159 mitochondrial thermogenesis (Fig. 2I, Fig. S10G,H) in a RIG-I/MDA5-dependent manner (Fig. 160 2J,K, Fig. S11H). Cytosolic mtDNA stimulated mitophagy in infant adipocytes (Fig. S12A,B). In 161 summary, EVs of infant adipocytes conveyed mtRNA and mtDNA to recipient adipocytes and 162 triggered beige adipocyte differentiation and mitophagy, respectively. 163 Breast milk is a known beige-inducing signal (7), and we found that human breast milk 164 EVs were rich in mtRNA (Fig. S12C). Eventually, breast milk EVsunlike formula milk EVs -165 induced beige adipocyte gene expression, mitobiogenesis and mitochondrial thermogenesis, and 166 in turn reduced IRF7 abundance in human adipocytes (Fig. S12D,E).  Suppressed IRF7 signaling permits beige adipogenesis by mtRNA   168   P56 adipocytes expressed IRF7, unlike P6 adipocytes. Activation of the STING/AIM2 and RIG-169   I/MDA5 pathway was strong in P56 adipocytes with synthetic ligands, with mtRNA or with   170 mtDNA, leading to Ifnb expression (Fig. S13A). Ultimately, IFN damaged adipocyte 171 mitochondria (Fig. S13B,C). In turn, IRF7-deificient adipocytes were immune privileged for 172 mitochondria (Fig. 3A, S4E), and mice lacking IRF7 retained their beige adipocytes to adulthood 173 (Fig. 3B). This is coherent with the protection of IRF7-deficient mice from obesity (32). Moreover, 174 P6 EVs reduced Irf7 mRNA and IRF7 protein levels in adipocytes (Fig. 3C) and did not induce 175 IFN-response (Fig. S13D). On the contrary, P56 EVs induced IFN-response and triggered Irf7 176 expression, and reduced mitochondrial content in adipocytes (Fig. S13D,E). 177 Vitamin D receptor (VDR)-controlled gene networks were highly expressed in P6 iAT 178 (Fig. S2A). The known VDR-target Camp, encoding cathelicidin, an adipose tissue enriched 179 antimicrobial peptide (33), was highly expressed at P6. In turn, the VDR-repressed gene Coro1a 180 had a low transcript level at P6 (Fig. 3D). Coro1a encodes coronin A1, also known as tryptophan-181 aspartate containing coat protein (TACO), which inhibits autophagosome formation (34). Low 182 levels of coronin A1 allow autophagy (34), which is in accordance with the prominent autophagy 183 we found in P6 iAT (Fig. 1H). The transcription of vitamin D metabolizing enzymes favored the 184 storage of vitamin D3 (Vit-D3) and the synthesis of the potent VDR-agonist calcitriol in P6 iAT 185 (Fig. 3D). Moreover, miR434-3p, a VDR-controlled miRNA which had complementarity to Irf7 186 mRNA (35) was also highly expressed in P6 iAT (Fig. 3E). IRF7 level and inflammasome 187 activation was effectively reduced by miR434-3p in adipocytes (Fig. 3E). Moreover, P6 EVs were 188 rich in Vit-D3, and cytosolic mtRNA increased the transcription of the calcitriol synthesis gene 189 Cyp27b1 in adipocytes (Fig. 3F). VDR protein expression was higher in P6 than in P56 iAT, and 13 Vit-D3 effectively suppressed Irf7 transcription in a VDR-dependent manner in adipocytes (Fig. 191 3F). Diet-induced obesity diminished adipocyte Vdr expression, and concomitantly upregulated 192 Irf7 in mice (Fig. 3G). Accordingly, inhibition of VDR signaling in young mice led to the loss of 193 beige adipocytes in iAT, along with increased IRF7 level in adipocytes (Fig. 3H). In turn, 194 suppression of IRF7 level with miR434-3p protected from inflammasome activation in adipocytes 195 of HFD-fed mice (Fig. 3I). 196 IRF7 is a hub for the transcription of AIM2/STING pathway (Fig. 13F,G), and thus 197 repression of IRF7 expression is a potential mechanism that protects infant adipocytes from an 198 IFN-response to cytosolic mtDNA/mtRNA (Fig. 13H). We found that VDR signaling suppressed  207 We found that childhood obesity compromised VDR-controlled gene networks and decreased the 208 expression of the calcitriol producing CYP27A1 (Fig. 4A), and increased IRF7 expression in the 209 iAT (Fig. 1E). Similarly, diet induced obesity compromised Vdr and increased Irf7 expression in 210 mouse, and inhibition of VDR signaling in infant mice led to the loss of beige fat cells (Fig. 3G,H). 211 Next, we studied a mouse model of childhood obesity, using infant mice which were nursed by 212 dams fed with HFD ( Fig. 4B) (36). In the offspring of HFD-fed dams adipocytes had a 213 14 compromised Vdr, and a robust Irf7 expression (Fig. 4C), and beige adipocytes were lacking from 214 the iAT (Fig. 4D). Eventually obesity developed and the adipocytes had a sustained inflammasome 215 activation (Fig. 4E). Moreover, the mitochondrial network was compromised in adipocytes ( Fig.   216 4F), and AIM2/STING pathway proteins were expressed in the cytosol and in the nuclei of 217 adipocytes of mice nursed by HFD-fed dams (Fig. 4F). In turn, Vit-D3 reverted these adverse 218 effects and protected the beige adipocyte content in infant mice (Fig. 4G), reduced obesity and 219 adipocyte inflammation (Fig. 4H). In adult HFD-fed mice, cytosolic delivery of mtRNA into the 220 iAT, combined with Vit-D3 treatment, reduced IRF7 level and increased beige adipocyte content

238
Adipose tissue inflammation is considered deleterious for metabolism (37). However, various lines 239 of evidence show that differentiation of thermogenic adipose tissue requires JAK/STAT3 signaling 240 (7,38,39), and an autocrine IL-6/STAT3 signaling loop is sustained by breast milk-derived lipid 241 signaling in the newborn adipose tissue (7). Some inflammatory signal mechanisms that cause 242 obesity-associated metabolic impairment also sustain beige adipocytes (40,41). Here we report 243 the unexpected finding that beige adipocyte development is promoted by a potentially 244 inflammation-evoking cytosolic RNA signal, released by the mitochondria of infant adipocytes. . Eventually, mtRNA serves as endogenous signal for beige adipogenesis in neighboring cells through the RIG-I/MDA5/IL-6/STAT3 pathway. In turn, mtDNA content of the EVs triggers mitophagy through STING signaling. (B) Albeit cytosolic mtRNA and mtRNA are noxious signals, they can act as metabolically beneficial mitochondriato-nucleus signals when IRF7 expression is suppressed. VDR is an effective suppressor of IRF7 and abrogates IFN-response to cytosolic mtRNA and mtDNA in infant adipocytes. Infant adipocytes are hence immune privileged sites for mitochondria, allowing a retrograde mitochondria-to-nucleus signaling through mtRNA, which is key for mitobiogenesis and beige fat development. 18 The endosymbiotic origin of mitochondria has led to a metabolic co-dependence of the 247 mitochondria and the host cell (42). This is driven by a retrograde, mitochondria-to-nucleus 248 signaling pathway, as the majority of genes required for the maintenance of mitochondria are 249 encoded in the nuclear genome. We show that, analogous to a parasite-host interaction, 250 mitochondrial nucleic acids are released by EVs, and are taken up by surrounding adipocytes to 251 activate cytosolic RNA sensors that stimulate an autocrine IL-6/STAT3 signaling loop, ultimately 252 triggering the nuclear expression of beige adipocyte genes (Fig. 2K, Fig. 5A). Non-coding RNA 253 species of mitochondria are known to increase the transcription of mitochondrial genome-encoded 254 genes (27). As an equivalent mechanism, we show that mtRNA species boost the transcription of 255 nuclear genome-encoded genes for mitochondrial biogenesis and thermogenesis. This is key for 256 mitobiogenesis since the majority of the mitochondrial genes are encoded in the nuclear genome 257 (42). The release of EVs containing mitochondrial nucleic acids resembles the recently explored 258 mechanism that allows nucleic acid delivery from bacteria to host cells in membrane microvesicles 259 (43,44). 260 The primary sensors of cytoplasmic mtRNA are RIG-I and MDA5. RIG-I detects dsRNAs 261 with or without a 5′-triphosphate end; MDA5 binds uncapped RNA; and RIG-I and MDA5 262 selectively recognize short and long dsRNAs, respectively (29,30). Given the prokaryote origin 263 of mitochondria, various mtRNA species such as mitochondrial ribosomal RNAs, uncapped 264 mitochondrial mRNA, and non-coding mtRNAs, can potentially stimulate the cytoplasmic RNA 265 sensor system (45,46). Beige adipocyte gene transcription was achievable by indirect RIG-I 266 activation using cytosolic p(dA:dT), and also by MDA5 activation using cytosolic high molecular 267 weight p(I:C), but not with cytosolic ssRNA. Coherently, lack of RIG-I and MDA5 signaling 19 compromised the mtRNA-mediated beige adipocyte development, and abrogated nucleus-encoded 269 SDH-A expression and mitobiogenesis, and promoted the loss of beige adipocytes in mice. 270 Nevertheless, excessive release of mitochondrial content is a danger signal, and activates 271 an IFN-response, which is detrimental for thermogenic fat development (16,47,48), triggers 272 obesity, mitochondrial dysfunction and the mitochondrial pathway of adipocyte apoptosis (49,50), 273 and may aggravate obesity-associated metabolic diseases (51,52). We show here that beige VDR may also skew IFN-response and IRF7 expression (53,54). Vit-D3 supplementation is today 288 routine in postnatal care, however, Vit-D3 deficiency is prevalent among obese children and 289 adolescents and is a risk factor for metabolic diseases (55-57). Vit-D3/VDR is proposed to inhibit 290 weight gain by activating UCP3 in the muscles (58), albeit VDR overexpression promotes weight 291 20 gain in mouse (59). Indeed, promotion of formula feeding originally served to increase Vit-D3 292 supply and induce weight gain (60). Formula milk lacks maternal lipid species that maintain beige 293 fat and has obesogenic effects (7). We also show here that formula milk lacks beige-inducing 294 mtRNA signals. Moreover, VDR signaling was impaired in the adipose tissue of obese children, 295 therefore despite its increased Vit-D3 level, formula milk is not sufficient to trigger beige 296 adipogenesis. However, when Vit-D3 supplementation is combined with stimulation of cytosolic 297 mtRNA signaling, beige adipocytes develop and obesity is reduced. 298 In summary, beige adipocyte development is dependent on a mtRNA-mediated signaling 299 and the suppression of IFN-response. Restoring the mtRNA-mediated mitochondria-to-nucleus 300 signaling may represent a novel and effective mechanism to increase beige fat and reduce obesity. 303 We used wt male C57BL/6 (Charles River Laboratories, Wilmington, MA), Irf7 -/-(RIKEN, Wako, 304 Japan), Ddx58 -/and Mda5 -/-(kindly provided by Gunther Hartmann, University of Bonn, 305 Germany) mice. All mouse lines were housed under SPF conditions. Animal experiments were 306 approved by the local ethics committees. Primary mouse adipocytes were isolated by collagenase 307 digestion and separation of cell fractions and subsequently analyzed or cultured, as described (7).   Red symbols indicate DEGs overrepresented in P6 iAT. Beige/brown adipocyte-associated genes were overrepresented in P6 iAT. (E) Transcription of Prdm16 in mouse adipocytes at P6 and P56. Heat map summarizing the transcription level of beige/brown adipocyte marker genes and white adipocyte marker genes in P6 and P56 iAT. Ucp1 is necessary for thermogenesis; Ppargc1a for mitochondrial biogenesis; Cidea, Cox7a1, Dio2, Zic1 are associated with brown/beige adipocytes; Tmem26 and Tbx1 are beige adipocyte markers; Eva1a is a brown adipocyte marker (4-9); Myf5 is expressed by progenitors of brown adipocytes (10). Levels of Hoxc8 and Hoxc9 increase along white adipocyte development (4), although Hoxc9 may also be a marker of beige adipocytes (9). Lep, Fabp4, Plin2, Adipoq, Gpd1, Slc2a4 and Pparg are associated with white adipocyte maturation (11). See also (12). (F) Correlation of UCP1 levels with beige/brown adipocyte-associated transcripts (PPARGC1A, TMEM26, CIDEA, LHX8) and white adipocyte markers (HOXC8, HOXC9) in the iAT of human male infants (4,13). P values were determined with linear regression analysis. Age 0.2-3.5 years. Further details regarding beige adipocyte content in mouse and human fat depots are provided in (4,(12)(13)(14), and reviewed in the introduction section of (3).

Supplemental Figure 2. Expression of the STING/AIM2 pathways in P6 and P56 iAT
(A) Gene ontology and STRING protein-protein association network of DEGs overrepresented in P6 iAT. Further analysis is available in (12). Vdr and its gene network were overrepresented at P6. (B) Gene ontology and protein-protein association network of underrepresented DEGs at P6 (15). (C) Structure of the DNA-sensor p204. The three DNAbinding domains are labeled A, B and C. p204 is encoded by Ifi204 in BALB/C mice. In C57/BL6, however, Ifi204 has a frameshift mutation and its function is taken over by Ifi205 (16)(17)(18). In 3T3-L1 cells, which have a BALB/C origin, we measured Ifi204, whereas we measured Ifi205 in adipocytes from C57/BL6 mice. Level of Ifi204 in P6 and P56-derived adipocytes mirrored that of Ifi205, shown in Figure   Vit-D3 and calcitriol were purchased from Sigma-Aldrich; IL-6, IFN and IFN from ImmunoTools (Friesoythe, Germany), NPVF, human and mouse NPFF from Tocris Bioscience (Bristol, UK). Isoproterenol and fumonisin B1 were purchased from Sigma-Aldrich and from Cayman Chemical Company, respectively. To test the inhibitory effect of Vit-D3 on IRF7 signaling, 3T3-L1 cells were treated with 1 M Vit-D3 for 48 h, and treated further with vehicle or 5 g/ml cGAMP for 6 h, or were transfected with mtRNA for 18 h. VDR was inhibited with PS121912, as described (42). Cellular uptake of cGAMP is dependent on the transporter Slc19a1 (20), whose level was similar in P6 and P56 adipocytes (GEO submission #GSE154925).

Isolation of extracellular vesicles from cell culture media, breast milk and formula milk
Extracellular vesicles (EVs) were collected from adipocyte culture media, human breast milk, or from commercially available cattle milk-based infant formula. Human breast milk was collected from healthy volunteers. For cell culture, to avoid contamination with bovine EVs, we used EV-depleted fetal calf serum throughout the study (Gibco). EVs were precipitated with the EPStep exosome precipitation solution (Immunostep, Centro de Investigación del Cáncer, Campus Miguel de Unamuno, Salamanca, Spain) and concentrated by centrifugation. EVs were analyzed with FACS using capture beads and labeling for CD63 (Immunostep). EV pellets were used for treating recipient cells, to extract DNA/RNA, or were processed for FACS. Fractions of EV pellets and adipocytes were also fixed in paraformaldehyde/glutaraldehyde, and processed for transmission electron microscopy (TEM) analysis, as described (43). Morphology of EVs was analyzed with conventional TEM, and with negative staining for TEM (44). EV diameter and area was measured with ImageJ (NIH) with manual annotation, and EVs were classified according to their morphology and electron density, as described (26, 27).

Phagocytosis and endocytosis assays
Uptake of naked nucleic acids was assessed microscopically by incubating adipocytes with rhodamine-conjugated p(dA:dT) or FITC-conjugated ODN 1668 CpG (both from InvivoGene) for 1 h. Endocytosis by means of pinocytosis was assessed by incubating adipocytes with FITC-conjugated dextran, followed by FACS analysis or fluorescence microscopy. Uptake of solid particles was assessed with the use of fluorescent latex beads (Sigma-Aldrich) and FACS analysis (BD LSR II).

ELISA assays
Tissue samples were weighed and homogenized in RIPA buffer using a Roche bead mill homogenizer at 6,500 rpm for 1 min. Cell culture supernatants and plasma samples were centrifuged at 0.8 g for 10 min to remove cell debris, and supernatants were used for analysis. We used commercial ELISA kits to measure the levels of IL-6, TNF (Fisher Scientific), Vit-D3, calcitriol and VDR (MBS268259-48, MBS2701844-24, MyBioSource). All samples were stored at -80°C until analysis.

mtRNA isolation and in vitro transfection
Adipocyte mitochondria were isolated with a commercial mitochondrial isolation kit (Thermo Fisher Scientific, Waltham, MA). Mitochondrial RNA (mtRNA) was isolated by lysing the mitochondrial pellet with TRI Reagent (Sigma-Aldrich), as described (1). 3T3L1 cells were transfected with 2 µg of mtRNA in 6-or 24-well plates with cells at 80-90% confluency. As a transfection reagent we used Lipofectamine 3000 (Invitrogen) at a 1:3 ratio. Control cells received transfection reagent only. Cells were analyzed 18 h after transfection.

mtDNA isolation and transfection
Mitochondrial DNA (mtDNA)was isolated from mitochondria pellets using TRI Reagent (Merck Sigma-Aldrich) and reconstituted in TE buffer (10 mM Tris-HCL, 1 mM EDTA, pH 8.0). 3T3L1 cells were transfected for 18 h with 1 µg/ml mtDNA using the TurboFect Transfection Reagent. Control cells received transfection reagent only. Agarose gel electrophoresis was used to examine mtDNA integrity.

Histology and image analysis
Tissues were fixed with 4% paraformaldehyde and embedded in paraffin, as described (1). Sections were stained with hematoxylin and eosin (Carl Roth, Karlsruhe, Germany). Antibodies are listed in Supplemental Table 2. UCP1, IFI16, AIM2 and NPFFR1 immunohistochemistry was performed on paraffin-embedded tissue sections. For histomorphometry of fat cells we used Image J, with an image-processing algorithm that incorporated the Euclidean distance-based Watershed transformation to segment the images. Briefly, binarized images were generated using Otsu's method for thresholding; enhanced images were generated using contrast limited adaptive histogram equalization (CLAHE), and finally segmented images were generated using the Watershed transformation (Supplemental Figure 20). Negative control specimens of our fluorescent imaging and immunostaining are shown in Supplemental Figure 15. Mitochondrial content and morphology was analyzed with ImageJ, as described (14). Beige adipose area was measured with our custom-developed image analysis software (BeAR©, (14)).

Oil Red-O staining and quantification of UCP1 staining
The triglyceride content of cultured adipocytes was examined by Oil Red-O using a commercial kit from BioOptica (Milan, Italy), as described (24). In vitro UCP1 immunostaining was performed in 6-well culture plates, and samples were imaged and the optical density was measured using digital image analysis. Original images are available upon request through Figshare. Mitochondria were also labeled using an SDH-A histochemistry assay (BioOptica).

Adipocyte differentiation
Mouse preadipocytes of the stromal vascular fraction (SVF) were isolated and maintained as described (24,43,46). To ensure the depletion of adipose tissue macrophages (ATMs) from the harvested preadipocytes, we used magnetic bead cell purification of the SVF with an antibody against the F4/80 antigen (Miltenyi Biotec, Bergisch Gladbach, Germany) (47).
Human subcutaneous adipose tissue preadipocytes were harvested as described (24,43). Preadipocytes were maintained in cell culture medium supplemented with 20 g/mL insulin. To induce white differentiation of preadipocytes of the SVF, we treated the cells with 50 M IBMX, 1 M dexamethasone, 1 M rosiglitazone and 20 g/ml insulin (all from Merck Sigma-Aldrich), as described (14).
Flow cytometry analysis of DNA sensors, mitochondrial biogenesis, mitochondrial content and mitochondrial uncoupling Mitochondrial content was analyzed with MitoTracker dyes (Thermo Fisher Scientific). Mitochondrial biogenesis was detected with the MitoBiogenesis™ Flow Cytometry Kit (Abcam, Cambridge, UK). MitoThermo Yellow (MTY), a temperature-sensitive fluorescent probe (48) was used to assess mitochondrial thermogenesis and uncoupling, as described (49,50). Temperature difference between the control and the test groups was expressed as Mito-T, and shown in the respective figures. MTY was developed and provided by Dr. Y-T.

miRNA detection
Total RNA was extracted by TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions and was quantified using the NanoDrop™ 8000 Fluorospectrometer (Thermo Fisher Scientific). In total, 50 ng of purified RNA was subjected to reverse transcription using a TaqMan miRNA Reverse Transcription Kit and TaqMan® MicroRNA Assays (Applied Biosystems, Foster City, CA) according to the manufacturer's instructions (Assay ID: mmu-miR-434-3p, 002604; mmu-miR-29a-5p, 002447; RUN6B, 001973). Quantification of individual miRNAs was using a QuantStudio™ 12K flex realtime PCR system (Applied Biosystems) and the relative expression values were calculated by using the 2−ΔΔCt method and normalized to RUN6B. miR434-3p was overexpressed using a custom-synthesized RNA (Sigma-Aldrich) and transfected with Turbofect transfection reagent (Fisher Scientific). To identify potential Irf7-interacting miRNA species, we searched the TargetScan database for miRNAs with complementarity to Irf7 mRNA. In the next step, we used miRBase to identify precursor-, and mature sequences of the candidate miRNA species (52).

Cell viability assay
We used the Presto Blue Cell Viability Assay (Thermo Fisher Scientific) and the Rotitest Vital (Carl Roth) assays according to the manufacturers' instructions.

Western blotting
Cells were lysed in ice-cold RIPA buffer supplemented with Pierce™ protease and phosphatase inhibitor mini tablets (Thermo Scientific). Protein concentration was measured by the Pierce™ Rapid Gold BCA Protein Assay Kit and 30-40 µg protein samples were run on 16% SDS gels for protein separation, followed by blotting the gels on 0.2-µm nitrocellulose blotting membrane (Amersham, Freiberg, Germany) at 300 mA for 1 h in a cold room. After blotting, membranes were blocked with 5% skimmed milk for 1 h. Providers of the -actin and LC3 antibodies are listed in Supplement Table 2. Antibody concentrations used were as follows: -actin, 1:10,000, LC3, 0.2 µg/ml.

Quantification of nucleic acids in extracellular vesicles
We collected EV pellets from cells, from formula milk or infant formula in a clean Eppendorf tube, which was centrifuged at 0.8 g to remove cell debris. To isolate the EV-associated DNA from the pellets or from the cell culture media, we used the Zymo Quick DNA Microprep Kit (Zymo Research, Irvine, CA). After determination of the DNA concentration, we used 5 ng for qPCR assays. EV-depleted cell culture media was used as a reference. For comparison between groups, we used the Ct method to determine relative changes in mtDNA levels. For extraction of mtRNA and other EV-associated RNA species from cell EV pellets and culture media, we used Trizol Reagent. After determination of the RNA concentration, we used 50 ng of RNA to generate cDNA. mtDNA copy number in the inguinal adipose tissue We used Trizol Reagent DNA isolation from iAT at P6 and P56. DNA was reconstituted in TE buffer and adjusted to 10 ng/µl. We performed qPCR using HK2 as a reference nuclear genome-encoded gene, and measured the DNA copy number of mtDNA-encoded 16S and Nd1. We calculated the copy number according to the formula: Ct=CtTarget gene-CtReference gene mtDNA copy number=2X2 (Ct) (1)

Magnetofection of mtRNA
In vivo delivery of mtRNA into the cytosol of adipocytes was achieved with magnetofection, using mtRNA-magnetic nanoparticle complexes (DogtorMag, OzBiosciences, San Diego, CA). Briefly, mtRNA-nanoparticle complexes were injected into the inguinal adipose tissue of mice, and enrichment of the magnetic nanoparticles was ensured by magnetic exposure of the fat depot, as described (53). MicroRNA was transfected using Lipofectamine 3000 (Thermo Fisher).