Hepatic Rab24 controls blood glucose homeostasis via improving mitochondrial plasticity

Non-alcoholic fatty liver disease (NAFLD) represents a key feature of obesity-related type 2 diabetes with increasing prevalence worldwide. To our knowledge, no treatment options are available to date, paving the way for more severe liver damage, including cirrhosis and hepatocellular carcinoma. Here, we show an unexpected function for an intracellular trafficking regulator, the small Rab GTPase Rab24, in mitochondrial fission and activation, which has an immediate impact on hepatic and systemic energy homeostasis. RAB24 is highly upregulated in the livers of obese patients with NAFLD and positively correlates with increased body fat in humans. Liver-selective inhibition of Rab24 increases autophagic flux and mitochondrial connectivity, leading to a strong improvement in hepatic steatosis and a reduction in serum glucose and cholesterol levels in obese mice. Our study highlights a potential therapeutic application of trafficking regulators, such as RAB24, for NAFLD and establishes a conceptual functional connection between intracellular transport and systemic metabolic dysfunction. Non-alcoholic steatohepatitis (NASH) is characterized by lipid accumulation within hepatocytes and fibrosis. Seitz et al. show that the GTPase protein Rab24 is increased in the livers of people who are obese or have NASH.

I ntracellular transport controls the internalization of nutrients into cells, the fine-tuning and downregulation of signalling receptors and the packaging and secretion of proteins, lipids and other metabolites [1][2][3][4][5][6] . These functions are mediated via vesicular trafficking, which is controlled by coat proteins, SNAP receptor proteins (SNAREs) and Rab GTPases that collectively enable specificity in the intracellular distribution of cargoes 4,[6][7][8][9] . This trafficking network is sensitive to extracellular cues and alters its kinetics and fates of transport depending on different environmental stimuli, including metabolic challenges 10 . Controlling the internalization of nutrient transporters and activated signalling receptors as well as the secretion of metabolically active proteins places this system theoretically on central stage of metabolic control. Surprisingly, although this connection seems obvious, the tight link between metabolite channelling and intracellular trafficking is still mainly unexplored 6 . Hundreds of trafficking components are involved in intracellular transport, providing a large population of potential metabolic regulators waiting to be characterized.
In this study, we present Rab24, which was originally described in the late endosomal pathway 11,12 , as a regulator in the mitochondrial fusion/fission cycle through direct interaction with mitochondrial fission 1 protein (FIS1). Reduction of Rab24 reduces mitochondrial fission, resulting in elongated and more connected mitochondria, and increases mitochondrial respiration. Since Rab24 knockdown in wild-type, high-fat diet (HFD) and methionine-choline-deficient (MCD) HFD-treated mice strongly improves glucose homeostasis and liver steatosis, our data highlight an unknown role of Rab24 in metabolic control.

Results
Rab24 is highly upregulated in patients with fatty liver. To identify trafficking candidates with functional impact on metabolic control, we screened over 1,200 knockout mice for their glucose sensitivity using an oral glucose tolerance test 13 . Out of five trafficking regulators with alteration (±15%) in systemic glucose clearance, wholebody knockout of Rab24 led to the most robust improvement in glucose tolerance, supporting a critical regulatory function of Rab24 in systemic glucose homeostasis. To ensure that Rab24 is also necessary for human metabolic control, we studied its abundance in two independent cohorts of patients with metabolic diseases. Alterations in liver metabolism are known to affect systemic glucose homeostasis and are associated with obesity-related type 2 diabetes 14,15 .
Thus, we first tested hepatic RAB24 expression in a cohort of patients who are obese versus healthy controls. Interestingly, RAB24 was upregulated threefold in the liver of patients who were obese (Fig. 1a). This was associated with a positive correlation of Rab24 with body mass index (BMI) and a negative correlation with the clamp glucose infusion rate GIR (GIR; Extended Data Fig. 1a,b) across the entire patient cohort. In addition, we observed a positive correlation with visceral fat (r = 0.375, P = 0.02), liver fat (r = 0.346, P = 0.03), high-density lipoprotein (HDL) cholesterol levels (r = 0.367, P = 0.02), free fatty acids (r = 0.34, P = 0.03) and leptin (r = 0.46, P = 0.005) levels in these patients, indicating that hepatic RAB24 levels are tightly associated with glucose and lipid homeostasis in humans.
Excess lipid accumulation leads to NAFLD with a possible progression to non-alcoholic steatohepatitis (NASH). To investigate the importance of Rab24 in more severe liver conditions, we checked the abundance of RAB24 levels in independent liver samples of patients with NAFLD (±steatosis), and NASH compared to healthy controls 16 . Interestingly, we found RAB24 to be upregulated 64 and 75% in patients with NAFLD plus steatosis and NASH patients, respectively (Fig. 1b). The alterations in RAB24 were negatively correlated with whole-body insulin sensitivity and positively correlated with hepatocellular lipids (Extended Data Fig. 1c,d). In addition, we observed a positive correlation with liver 8-oxoguanosine (r = 0.16, P = 0.036) and interleukin-6 (r = 0.39, P = 0.039) levels in these patients, markers that correspond to increase oxidative DNA damage and activation of cytokine pathways, respectively. Altogether, these data highlighted a relationship between higher Rab24 levels and an impaired metabolic state (lower whole-body insulin sensitivity, high fat accumulation and inflammation) in humans.

Rab24 knockdown improves glucose tolerance and serum lipid parameters.
To functionally explore Rab24, we administered lipid nanoparticles (LNPs) containing short interfering RNA (siRNA; against Rab24 or luciferase as control at 0.5 mg kg −1 each) via tail vein injection to silence Rab24 in the liver 14,15,17 . Five days after injection, treatment with LNPs resulted in a 60% reduction in Rab24 messenger RNA specifically in the liver ( Supplementary  Fig. 1a) and 75% reduction in Rab24 protein levels compared to control ( Supplementary Fig. 1b,c). Rab24 knockdown mice had similar body weight ( Supplementary Fig. 1d), but showed a decrease in the liver-to-body weight ratio ( Supplementary Fig. 1e). In agreement with the oral glucose tolerance test data from wholebody knockout mice, we observed an improvement in glucose clearance and a 15% reduction in the area under the curve (AUC) (Fig. 1c,d) of Rab24 knockdown mice without affecting serum insulin levels and homeostatic model assessment of insulin resistance ( Supplementary Fig. 1f,g), highlighting the contribution of hepatic Rab24 to systemic glucose homeostasis. Insulin responsiveness was unchanged; Rab24 knockdown mice exhibited similar insulin tolerance compared to controls ( Supplementary Fig. 1h). Surprisingly, insulin-induced protein kinase B (Akt) activation in skeletal muscle, but not in the liver or fat, was enhanced in Rab24 knockdown mice, pointing towards a Rab24-dependent inter-organ communication pathway (Fig. 1e,f and Supplementary Fig. 1i-k). Indeed, expression and secretion of fibroblast growth factor 21 (FGF21) was elevated in primary mouse hepatocytes and mouse liver upon Rab24 knockdown (Fig. 1g-j). Importantly, reduction of Rab24 in FGF21 homozygous knockout mice caused no improvement in glucose clearance and insulin signalling in skeletal muscle compared to their heterozygous littermates, demonstrating an FGF21-dependent mechanism (Fig. 1k-p). Interestingly, the liver but not fat of heterozygous FGF21 knockout mice showed enhanced insulin-induced Akt activation upon Rab24 knockdown, which was abolished in the homozygous controls, suggesting FGF21dependent autocrine regulation ( Supplementary Fig. 1l-s). We did not observe any alterations in brown adipose tissue activation upon Rab24 knockdown, indicating a brown adipose tissue-independent mechanism (Supplementary Fig. 2a-h).
Interestingly, knockdown of Rab24 also led to a decrease in serum total and low-density lipoprotein (LDL) cholesterol as well as apolipoprotein B (Apo B; Table 1), suggesting an alteration of LDL uptake or secretion by the liver. Thus, we measured the LDL internalization kinetics in primary hepatocytes with 60% reduction in mRNA and protein levels of Rab24 ( Supplementary  Fig. 3a-c) using a continuous uptake assay of fluorescently labelled 1,1′-di-n-octadecyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate (DiI)-LDL for various time points 14,18 . Rab24 knockdown caused a small increase in LDL endocytosis, which contributed to the improved serum LDL parameters (Fig. 1q,r) without affecting the expression of cholesterol transporters ( Supplementary Fig. 3d), suggesting an increase in LDL trafficking. In addition to uptake, the liver is a major source of circulating cholesterol 19 . Interestingly, Rab24 knockdown resulted in reduced cholesterol secretion from primary hepatocytes (Fig. 1s) and an increase in liver bile acid levels after 6 h starvation in Rab24 knockdown mice (Fig. 1t). Altogether, these data provided in vivo evidence for an as-yet-unknown role for hepatic Rab24 in the regulation of glucose and lipid handling.

Upregulation of mitochondrial proteins upon Rab24 reduction.
To study the mechanisms of Rab24 metabolic control, we performed quantitative proteomics analysis of liver tissues from control and Rab24 knockdown mice. The tissues were subjected to liquid chromatography-tandem mass spectrometry (LC-MS/MS) combined analysis of the spectra from all samples resulted in the quantification of almost 5,000 proteins at a false discovery rate (FDR) of 1% using the label-free quantification algorithm in MaxQuant (v.1.5.7.9; Supplementary Table 1). In a stringently filtered dataset for valid values of 3,600 proteins, we detected 622 differentially expressed proteins, of which 287 were upregulated and 335 downregulated in the liver of patients who are obese (a) and patients with NALFD ± steatosis and NASH (b) versus healthy controls. c,d, IPGTT (2 g kg −1 ) (c) and AUC (d) after 6 h starvation on day 5 after RNAi (n = 6 animals). e,f, Representative immunoblots for Akt in muscle of insulin-injected mice (0.75 U kg −1 , 7 min) after 6 h fasting (e) and quantification thereof (f). The experiment was done twice with similar results. g-i, FGF21 expression (g) and FGF21 secretion (i) in primary hepatocytes after 3 d of knockdown. g, Twelve wells pooled into n = 4 replicates. h, FGF21 expression in the liver (n = 10 (control) and n = 12 (knockdown) animals). i, n = 12 wells per condition. j, Amounts in serum (n = 6 (control) and n = 8 (knockdown) animals) after 6 d of RNAi. k,l, IPGTT of heterozygous (k) (n = 8 animals) and homozygous (l) (n = 9 animals) FGF21 knockout mice after 5 d of RNAi. m-p, Representative immunoblots for Akt in muscle of insulin-injected heterozygous (m) and homozygous (o) FGF21 knockout mice (0.75 U kg, 7 min) after 6 h fasting and quantification thereof in n and p, respectively. q,r, Representative confocal images (maximal projection of three confocal slices) of primary hepatocytes internalizing DiI-LDL (grey) for 60 min stained with DAPI (q) and quantification thereof with Fiji (r). a.u., arbitrary unit. The images are representative of 3 independent wells of a 24-well plate. The experiment was repeated twice with similar results. Scale bar, 20 µm. s, Cholesterol secretion assay in primary hepatocytes after 4 d of knockdown from n = 11 wells per condition. t, Liver bile acids after 6 d of RNAi (n = 6 animals). All animals treated with control and Rab24 (knockdown) LNPs (0.5 mg kg −1 ). e,f,m-p, n = 3 animals per condition. Primary hepatocytes treated with control or Rab24 (knockdown) siRNA (40 nM; mean ± s.e.m.). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, by two-tailed unpaired Student's t-test. Only data that reached statistical significance are indicated. compared to the control t-test (P < 0.05) (Supplementary Table 1 and Fig. 2a). Next, we subjected the differentially expressed proteins to pathway enrichment analysis using Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) annotation (Fig. 2b,c). Most of the upregulated proteins were involved in metabolic processes, especially carbon and pyruvate metabolism, amino acid degradation and tricarboxylic acid cycle. Interestingly, most of those metabolic reactions are resident in the mitochondria. In agreement, we observed a strong upregulation of mitochondria based on GO annotation. Interestingly, 101 mitochondrial proteins and 21 proteins involved in carbon metabolism were increased in abundance in the individual Rab24 knockdown liver samples  Table 2). Proteins of the ribosomal pathway were downregulated. The remarkable increase in mitochondrial components prompted us to visualize the distribution of mitochondria in the liver. Paraffin sections of Rab24 knockdown livers revealed an approximately 20% increase in the staining of the mitochondrial inner membrane marker prohibitin (Fig. 2d,e), suggesting an increase in mitochondrial mass. Altogether, these data indicated an upregulation of carbon metabolism and a pronounced increase in mitochondrial proteins upon Rab24 knockdown.
Loss of Rab24 causes an increase in mitochondrial activity. The striking effect on mitochondrial proteins inspired us to investigate the role of Rab24 on mitochondrial function in primary hepatocytes in vitro and liver in vivo. Interestingly, we observed an enhancement in the staining of mitochondrial import receptor subunit TOM20 (TOM20) by quantitative immunofluorescence (IF) analysis in vitro and in vivo, supporting an increase in mitochondrial mass ( Fig. 3a-d). This was confirmed by an up to 70% increase in MitoTracker Green staining in primary and isolated hepatocytes from control and Rab24 knockdown animals ( Fig. 3e,f). To investigate whether the increase in mitochondrial mass was associated with an improvement in mitochondrial function, we performed Seahorse analysis to measure the oxygen consumption rate (OCR) in primary mouse hepatocytes after Rab24 knockdown. Interestingly, reduction of Rab24 in vitro led to an increase in the OCR, including basal respiration, ATP production and maximal respiration (Fig. 3g,h), revealing a regulatory role on mitochondrial function. Intriguingly, isolated primary hepatocytes from mice after in vivo knockdown of Rab24 displayed a similar induction in mitochondrial respiration, indicating an enhanced mitochondrial activity even in the liver (Fig. 3i,j). Supporting this observation, we also observed an increase in MitoTracker Red staining in primary and isolated hepatocytes from control and Rab24 knockdown mice (Fig. 3k,l). The induction in OCR was accompanied by a 25% increase in cellular ATP and enhanced reactive oxygen species (ROS) production (Fig. 3m,n). Despite the observed ROS elevation, we did not detect an increase in carbonylated proteins (Fig. 3o), indicating that the induced ROS levels upon Rab24 knockdown were not damaging for the hepatocytes. Altogether, our data showed an induction of mitochondrial mass and respiration upon Rab24 depletion, highlighting a regulatory role of Rab24 in energy metabolism.
Reduction of Rab24 leads to enhanced glycolysis. To test whether the increase in mitochondrial respiration was also affecting glycolytic flux, we investigated the effect of Rab24 knockdown on glycolysis by measuring the extracellular acidification rate (ECAR). Interestingly, we observed an increase in ECAR, basal glycolysis and glycolytic capacity in the Rab24 knockdown hepatocytes, suggesting enhanced glycolytic activity (Extended Data Fig. 2a,b). The activation of glycolysis was strongly reduced by treatment with antimycin A and rotenone, indicating the contribution of mitochondrial respiration. However, there was still a notable 30% difference in ECAR between control and Rab24 knockdown cells after antimycin A and rotenone treatment, which was completely abolished by injection of 2-deoxy-d-glucose, suggesting an induction of glycolysis upon Rab24 knockdown. This is in agreement with our proteomics data, where we detected an upregulation of glucokinase, phosphoglucomutase, pyruvate kinase and pyruvate dehydrogenase complex, indicating enhanced glycolysis (Extended Data Fig. 2c). The increase in glycolysis was accompanied by enhanced glucose uptake during the ECAR assay and on extracellular glucose stimulation (Extended Data Fig. 2d). This can be explained by an elevation in GLUT2 (SLC2A2) expression, but not GLUT1 (SLC2A1) in vitro and an increase in GLUT1 (SLC2A1) in vivo upon Rab24 knockdown, suggesting differential activation of the glucose transporters (Extended Data Fig. 2e,f). The increase in ECAR caused a 20% accumulation of lactate in the medium upon Rab24 knockdown, which further confirmed an activation of glycolysis (Extended Data Fig. 2g). Interestingly, the effect of Rab24 was predominantly acting on anabolic glycolysis, since no effect on glucose production via gluconeogenesis was observed after Rab24 knockdown in vivo (Extended Data Fig. 2h). Altogether, these data demonstrate that Rab24 reduction causes a mild increase in glycolysis and a strong activation of mitochondrial respiration.

Rab24 knockdown increases mitochondrial connectivity by reducing fission.
To investigate how Rab24 regulates mitochondrial mass and activity, we first checked whether Rab24 affects mitochondrial biogenesis. Thus, we measured mRNA levels of PGC1A, PGC1B, NRF1 and PPARG in primary hepatocytes in vitro and liver in vivo ( Supplementary Fig. 4a,b) and observed no alterations upon Rab24 knockdown, indicating that Rab24 knockdown does not activate mitochondrial biogenesis. Mitochondria are very dynamic organelles that undergo morphological adaptations to different nutritional conditions to optimize their ATP production depending on the external nutrient cues 20 . Thus, enhanced respiration and bioenergetics are associated with an increased mitochondrial network and elongation, which are induced under prolonged starvation 21 . On the other hand, nutrient overload induces fragmentation of the mitochondrial network and a shift towards nutrient storage 22 . To test whether Rab24 knockdown changed mitochondrial morphology, we employed electron microscopy analysis in liver tissues and primary hepatocytes to determine the density, surface area, perimeter (contour), form factor (complexity and branching of the mitochondria) and circularity (roundness) of the mitochondria 23 . Surprisingly, we observed a 10-30% increase in mitochondrial density, surface area, perimeter and form factor and an approximately equal to 10% decrease in circularity upon Rab24 knockdown in vivo ( Fig. 4a-f) and in vitro (Extended Data Fig. 3a-i), indicating slightly bigger and more connected mitochondria. Importantly, mitochondrial morphology appeared normal with similar levels of cristae structures, demonstrating an increase in healthy mitochondria (Extended Data Fig. 3c,d).
To further confirm an induction in mitochondrial connectivity, we performed mitochondrial network analysis from deconvolved images stained for TOM20 using the skeletonization analysis tool from Fiji (Analyze Skeleton (2D/3D), v.3.3.0). With this, the fluorescence signals of deconvolved images are detected and skeletonized, allowing quantitative measurements of TOM20-labelled mitochondrial outer membrane structures, including individual branches, junctions and the length of mitochondrial branches. Remarkably, Rab24 knockdown induced more connected mitochondria as evident in the zoomed deconvolved images in vivo (Fig. 4g,h) and in vitro (Extended Data Fig. 4a), characterized by a 50-70% increase in the number of branches and junctions per area as well as the length of branches ( Fig. 4i-k and Extended Data Fig. 4b-d).
The accumulation of mitochondrial density with a wider surface and better connectivity, without changing biogenesis, suggested an alteration in the mitochondrial fusion/fission cycle 24 . Our proteomics analysis excluded alterations in the fusion machinery 20 , since mitofusin 1 and optic atrophy 1 protein levels were unchanged (Extended Data Fig. 5a). However, we observed significant reductions in mitochondrial fission regulator 1 and the solute carrier family 25 member 46 (Extended Data Fig. 5a), both regulators of mitochondrial fission [25][26][27] .
Fission is enhanced by endoplasmic reticulum/mitochondria interactions on organelle contact sites 28,29 . Thus, we stained primary mouse hepatocytes with the endoplasmic reticulum marker KDEL and the mitochondrial marker TOM20 and observed enhanced colocalization upon Rab24 knockdown (Extended Data Fig. 5b, left and middle panels), measured by a 50% increase in the Pearson's correlation between KDEL and TOM20 intensities (Extended Data Fig. 5b, right panels, and 5c), suggesting more contact between the endoplasmic reticulum and mitochondria when Rab24 was reduced. Mitochondrial fission is induced by forming the fission complex, which consists of FIS1, mitochondrial fission factor and mitochondrial dynamics proteins 49 and 51; this induces the activation and recruitment of dynamin-related protein 1 (DRP1) to the mitochondrial membrane 30,31 . Therefore, we stained primary hepatocytes for DRP1 and FIS1 and observed a 25% reduction in the mean fluorescence intensity per cell of DRP1 without affecting FIS1 (Fig. 5a-c), indicating inefficient recruitment of DRP1 upon loss of Rab24. Importantly, proteomics analysis revealed no change in the protein levels of DRP1, supporting a defect in subcellular localization Quantifications are derived from n = 33 (Ctrl) and n = 25 (knockdown) (c) and n = 9 (Ctrl) and n = 12 (knockdown) (d) cellular regions. e,f, MitoTracker Green staining (200 nM for 45 min) of primary hepatocytes after 3 d of RNAi in vitro (e) (n = 12 wells per condition) and in hepatocytes isolated after in vivo RNAi for 5 d (f) (n = 32 wells per condition) normalized to DAPI. g-j, Seahorse measurements of the OCR and the corresponding metabolic rates after 3 d of knockdown in primary hepatocytes in vitro (g,h) and hepatocytes isolated after in vivo RNAi for 5 d (i,j) (n = 10 wells per condition for g and i, except n = 7 for control in vivo). h,j, n = 3 time points with 10 wells per time point (basal and ATP production). n = 8 (Ctrl) and n = 9 (knockdown) (h), n = 6 (Ctrl) and n = 10 (knockdown) (j) wells per condition for maximal respiration. The experiment was repeated five times with similar results. k,l, MitoTracker Red staining (250 nM for 45 min) of primary hepatocytes after 3 d of RNAi in vitro (k) (n = 12 wells per condition) and hepatocytes isolated after in vivo RNAi for 5 d (l) (n = 32 wells per condition) normalized to DAPI. m-o, Levels of ATP, ROS and carbonylation of proteins, respectively. n = 18 (m), n = 6 (n) and n = 3 (o) wells per condition. All animals were treated with control and Rab24 knockdown LNPs (0.5 mg kg −1 ). Primary hepatocytes were treated with control or Rab24 knockdown siRNA (40 nM) and measured after 3 d after RNAi (mean ± s.e.m.). *P < 0.05, **P < 0.01, ****P < 0.0001, *****P < 0.00001 by two-tailed unpaired Student's t-test. (Fig. 5d). Indeed, we observed an equal decrease in the colocalization of DRP1 with TOM20 measured using the Pearson's correlation (Fig. 5e,f), indicating reduced DRP1 recruitment to the mitochondria upon Rab24 knockdown. In agreement with the proteomics data, mitofusin 1 and 2 were unaffected, supporting a preferential function of Rab24 on the fission machinery The images are representative of n = 3 (control) and n = 4 (knockdown) independent biological samples, which give rise to the quantifications in i-k. i-k, Quantification of the number of branches, number of junctions per area and mean length of the branches using Fiji. Scale bars, 20 µm. All animals were treated with control and Rab24 knockdown LNPs (0.5 mg kg −1 ). *P < 0.05, ****P < 0.0001, *****P < 0.00001 by two-tailed unpaired Student's t-test.
(Extended Data Fig. 5d-f). In fact, Rab24 has been shown to interact with FIS1 in mammalian cells (BioGrid 3.5, https://thebiogrid. org/119817/summary/homo-sapiens/rab24.html). To directly test this observation in the liver, we performed pulldown experiments of glutathione S-transferase (GST)-tagged Rab24 compared to a control Rab, Rab3a, using 12-h fasted and 12-h fasted plus 2-h refed liver lysates. Strikingly, we found a specific interaction of Rab24 with FIS1 but not with DRP1 in whole liver samples (Fig. 5g), supporting the BioGrid 3.5 interaction data. Importantly, no interaction with Rab3a was observed (Fig. 5g). These data indicate that Rab24 participates in regulating mitochondrial fission by directly interacting with FIS1.
If the connection of Rab24 to FIS1 was crucial for mitochondrial morphology and activity, interfering with FIS1 should mimic the Rab24 knockdown phenotype on mitochondria. Rab24 only induced an approximately 25% reduction in the assembly of the fission machinery; therefore, we performed a knockdown of FIS1 with similar efficiency (Extended Data Fig. 6a). Reduction of FIS1 by 30% caused a rise in basal respiration, ATP production and maximal respiration in primary hepatocytes (Extended Data Fig. 6b-e), induced by an increase in TOM20 intensity and mitochondrial connectivity (Extended Data Fig. 6f-j), supporting reduced fission and enhanced activity under mild FIS1 knockdown conditions. Overall, these data underlined the conclusion that under physiological The images are representative of 3 independent wells of a 24-well plate. The experiment was repeated twice with similar results. d, Abundance of Drp1/Dnml1 in the liver by proteomics from n = 6 animals per condition. e, Representative confocal images (single confocal section) of primary hepatocytes stained with DAPI (blue), TOM20 (green) and DRP1 (red). The images are representative of 3 independent wells of a 24-well plate. The experiment was repeated twice with similar results. f, Pearson's correlation analysed with Fiji representing the overlap between the green and red channel from three wells per condition. All measured after 3 d of RNAi (40 nM) in primary hepatocytes (mean ± s.e.m.). Scale bar, 20 µm. g, Pulldown assay for Fis1 and Drp1 with Rab24 or Rab3a as the control in mice fasted for 12 h or fasted and refed for 2 h. n = 3 mice per condition. The experiment was done twice with similar results. ***P < 0.001, ****P < 0.0001 by two-tailed unpaired Student's t-test. Only data that reached statistical significance are indicated.  (c) and their quantification (d). The red arrows indicate mitophagic events. Scale bar, 2 µm. n = 6 wells per condition. e-g, Representative immunoblots (e) of LC3-II/valosin-containing protein (VCP) levels in primary hepatocytes (in vitro) and liver tissue (in vivo) in fed, starved or chloroquine-treated cells. Cells were kept in full medium (fed) or serum-starved for 12 h (fasted), followed by 20 µM chloroquine treatment for 3 h (fasted and chloroquine). Mice were fed ad libitum, fasted for 12 h or fasted for 12 h, and this was followed by 100 mg kg −1 chloroquine treatment for 3 h (fasted and chloroquine). LC3-II flux (fasted and chloroquine/fasted) and net flux (fasted and chloroquine-fasted) measurements of LC3-II protein levels in primary hepatocytes (f) and liver tissue (g) are shown. The quantifications in f are from 6 independent wells of a 24-well plate pooled into n = 2 replicates. All animals were treated with control and Rab24 knockdown LNPs (0.5 mg kg −1 ) for 5 d (n = 2 animals per condition). Both experiments were done twice with similar results. h, Representative confocal images (three merged middle sections) of primary hepatocytes stained for LC3-II and treated as described in e. Scale bar, 30 µm. The images are representative of 3 independent wells of a 24-well plate. i,j, Quantification of LC3-II flux (i) and net flux (j) in hepatocytes acquired in h from n = 22 (controlflux and net flux), n = 19 (knockdown-flux), n = 20 (knockdown-net flux) cells per condition. Primary hepatocytes were treated with control or Rab24 knockdown siRNA (40 nM) and measured 3 d after RNAi (mean ± s.e.m.). *P < 0.05, **P < 0.01, ****P < 0.0001 by two-tailed unpaired Student's t-test.
conditions Rab24 induces mitochondrial fission by directly interacting with FIS1, and that the inhibition of hepatic Rab24 reprogrammes mitochondrial turnover to boost mitochondrial connectivity and metabolic functions, ultimately leading to improvements in systemic metabolic health.

Rab24 knockdown causes a reduction in mitophagy and increases autophagic flux.
Besides the function of mitochondrial fission in organelle plasticity under nutrient-rich conditions, fission is required for mitochondrial degradation, where damaged parts of the mitochondria are fissioned off and degraded via mitophagy 31,32 . Since Rab24 affects mitochondrial fission, we hypothesized that its reduction would also alter mitophagy. To investigate this, we measured mitophagic flux under stress-induced conditions using the nucleoid depletion assay 33,34 . Carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP) and oligomycin treatment induced a 34% reduction in cytoplasmic mitochondrial DNA particles in control cells, indicative of induced mitophagy (Fig. 6a,b). Interestingly, DNA depletion was reduced by twofold in Rab24 knockdown cells upon FCCP and oligomycin treatment suggesting less mitochondrial degradation (Fig. 6a,b), which is in agreement with reduced mitochondrial fission. Reduction in mitophagy was confirmed by another assay that measures the accumulation of mitophagic vesicles upon FCCP treatment with and without chloroquine, where we observed a 20% reduction upon Rab24 knockdown (Extended Data Fig. 7a). This was in agreement with electron microscopy data, where we detected already in the basal state fewer mitochondria engulfed by double membranes designated to fuse with lysosomes ( Fig. 6c,d). Indeed, the colocalization of mitochondria labelled by TOM20 and lysosomes visualized by lysosome-associated membrane glycoprotein 1 (Lamp1) was significantly reduced when measured by Pearson's correlation analysis (Extended Data Fig. 7b,c), supporting reduced mitophagy upon Rab24 knockdown, most probably due to a decrease in mitochondrial fission.
Rab24 has been shown to be involved in the autophagic pathway 35 and to be required for autophagosome clearance by interacting with lysosomes under nutrient-rich conditions in cultured cancer cells 12 . To investigate whether Rab24 knockdown also alters autophagic flux in primary hepatocytes in vitro and liver in vivo, we compared the levels of the lipid-modified form of autophagy-related protein LC3 (LC3-II) and the autophagy receptor sequestosome-1 (p62) in fed, starved and starved plus chloroquine conditions (to prevent lysosomal degradation) by immunoblotting and IF. Starvation is a strong inducer of macroautophagy in the liver 36 , as evidenced by the increase in LC3-II between fasting and feeding in control cells and liver tissue ( Fig. 6e and Extended Data Fig. 8a,b). Chloroquine treatment induced additional accumulation of LC3-II, indicating increased degradation upon starvation in vitro and in vivo ( Fig. 6e and Extended Data Fig. 8a,b). Upon Rab24 reduction, we observed no change in the fed state for LC3-II by immunofluorescence and immunoblot in vitro and by immunoblot in vivo ( Fig. 6e and Extended Data Fig. 8a-c) but a further increase of LC3-II upon chloroquine stimulation ( Fig. 6e and Extended Data Fig. 8a,b), suggesting enhanced autophagy. Changes in autophagic flux can be measured by calculating the ratio of LC3-II levels in fasting plus chloroquine and fasting only conditions ('fasted plus cholorquine' divided by 'fasted') or by subtracting LC3-II levels in fasting from the fasting plus chloroquine conditions ('fasted plus chloroqine' minus 'fasted'), resulting in net LC3-II flux between samples (control versus Rab24 knockdown) 37 . Using this assay, we measured a strong increase in LC3-II levels upon knockdown of Rab24 in vitro by immunoblot and immunofluorescence (Fig. 6f,h-j) and in vivo (Fig. 6g), demonstrating enhanced LC3 flux and net flux in the absence of Rab24.
The levels of p62 showed only a small increase in fasted control cells and tissues upon chloroquine treatment, suggesting minimal degradation of p62 under fasting conditions (Extended Data Fig. 8d-h). This is in agreement with p62's function in selective autophagy 38 , which is not activated under fasting 36 . Therefore, p62 flux during starvation is low and does not contribute to the induction of bulk macroautophagy. The reduction of p62 in Rab24 knockdown conditions (Extended Data Fig. 8d-h) is surprising and could have potential other reasons, such as transcriptional regulation 39 ; however, it has no consequence on the activation of autophagy under starvation in Rab24 knockdown conditions. The induction of autophagic flux was accompanied by an increase of LAMP1 + structures by immunofluorescence (Extended Data Fig. 8i,j), as indicated previously 40 . Our results in primary mouse hepatocytes in vitro and liver in vivo do not agree with the previously observed reduction of autophagy under Rab24 knockdown 40 . However, this was only evident under full medium; upon amino acid starvation, a condition that physiologically activates autophagy, no effect of Rab24 knockdown was observed 40 . This discrepancy indicates that Rab24 might fulfill different functions in autophagy in primary hepatocytes and mouse liver compared to stable-expressing Rab24 cell lines. Altogether, our data demonstrate that Rab24 knockdown increased p62-independent non-selective macroautophagy while decreasing mitochondrial fission and mitophagy, thereby boosting respiration, thus contributing to enhanced nutrient consumption.

Reduction of Rab24 improves glucose and lipid parameters in HFD mice.
To test whether the cellular functions of Rab24 translate into improvements of systemic metabolic health in metabolically impaired conditions, we used a model of diet-induced obesity, feeding mice an HFD, and elucidated the therapeutic possibility of reducing Rab24 levels for glucose and lipid metabolism. HFD-treated mice gained more weight and showed an increase in their fed blood glucose levels compared to low-fat diet (LFD)treated mice (Extended Data Fig. 9a,b), consistent with the occurrence of obesity and hyperglycaemia in this disease mouse model. Interestingly, we observed an induction of Rab24 mRNA levels upon HFD (Extended Data Fig. 9c) in agreement with the human data ( Fig. 1a,b). To examine the therapeutic potential of Rab24, we performed liver-specific knockdown in 13-week-old HFD mice and measured an 80% reduction in Rab24 protein levels after 2 weeks of RNA interference (RNAi) without affecting body weight (Extended Data Fig. 9d,e). Interestingly, reduction of Rab24 led to a decrease in serum cholesterol, LDL, Apo B (Fig. 7a-c) and triglyceride (P = 0.06) (Extended Data Fig. 9f) levels. Note that Rab24 knockdown for two weeks did not completely reverse hypercholesterolaemia in HFD mice compared to controls, but contributed very significantly to an improvement in serum lipid parameters. In fact, serum alanine aminotransferase (ALT) levels were completely restored (Fig. 7d) and liver lipid content strongly reduced (Fig. 7e-g and Extended Data Fig. 9g-j), highlighting the beneficial effect of Rab24 knockdown on overall liver health under an HFD. Interestingly, loss of Rab24 led to a decrease in the liver-to-body weight ratio (Fig. 7h), as observed in wild-type mice (Supplementary Fig. 1e). The improvement in liver and serum lipids was associated partly with lower fasting blood glucose levels in the HFD mice (P = 0.07) (Fig. 7i) and was accompanied by amelioration in their GTT and AUC after 4 weeks of LNP treatment (Fig. 7j,k). Importantly, Rab24 reduction in control LFD mice showed similar beneficial effects on serum lipid parameters without changing blood glucose levels, further strengthening the positive effects of diminishing Rab24 activity (Extended Data Fig. 10a-h).
with the fatty acids oleate and palmitate for 3 d exhibited increased lipid droplet formation as evidenced by perilipin-2 staining compared to BSA alone, which was decreased upon Rab24 reduction (Extended Data Fig. 9k). Interestingly, Rab24 knockdown hepatocytes treated with fatty acids exhibited an increase in OCR, basal respiration and ATP production (Extended Data Fig. 9i and Fig. 7n), underscoring the beneficial effect of loss of Rab24 for liver lipid homeostasis and mitochondrial respiration. Altogether, our data emphasized a potential therapeutic role of Rab24 in liver steatosis, glucose homeostasis and serum cholesterol levels in a model of diet-induced obesity. a,d,f-j, n = 6 animals per condition. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 by two-tailed unpaired Student's t-test.

Loss of Rab24 ameliorates liver steatosis and inflammation in NASH.
Since Rab24 was also upregulated in NASH patients, we performed knockdown experiments of Rab24 in mice under an MCD diet coupled to an HFD (MCD-HFD) 41 . The MCD-HFD model developed rapid steatosis and inflammation and progressive fibrosis 41 , but the severe weight loss of the usual MCD diet was extenuated 42 . Interestingly, we observed a 30% upregulation of Rab24 expression in MCD-HFD versus control LFD (Fig. 8a), supporting the human data (Fig. 1b). LNP-based Rab24 knockdown induced a 60% reduction in mRNA levels in the MCD-HFD group, supporting efficient knockdown of Rab24 in this mouse model (Fig. 8a). Interestingly, Rab24 knockdown improved reduction in body weight and blood glucose in MCD-HFD mice, suggesting a healthier metabolic state for these mice (Fig. 8b,c). This was accompanied by a reduction in the liver-to-body weight ratio, which is usually induced due to the strong increase in steatosis in MCD-HFD mice (Fig. 8d). Indeed, reduction of Rab24 decreased liver steatosis and triglyceride content and improved serum ALT levels, indicating an improvement in liver health (Fig. 8e-h). Intriguingly, markers of NASH induction, such as alpha-actin-2 (encoding alpha smooth muscle actin, a marker of activated hepatic stellate cells) and adhesion G protein-coupled receptor E1 (encoding F4/80, a marker of murine macrophages) were also significantly reduced upon Rab24 knockdown (Fig. 8i,j). These data demonstrate a beneficial effect of reducing Rab24 at an early stage of NASH development.

Discussion
The importance of improving mitochondrial activity has been proven to be beneficial under conditions of diet-induced obesity and in diabetic animal models 43,44 , placing mitochondria at the centre of metabolic control. Mitochondrial activity is strongly regulated by mitochondrial plasticity and turnover, which is controlled by nutrient availability. Fasting inhibits mitochondrial fission, and consequently degradation by mitophagy, and induces hyperfused mitochondria to ensure proper use of energy substrates provided by the autophagic pathways, when no external nutrients are available 21 .
On the other hand, under postprandial conditions, mitochondria fragment, mitophagy is induced and consequently mitochondrial respiration decreases 45,46 .
In this study, we present data that strongly support Rab24 as a regulator of mitochondrial turnover by underlining its role in the assembly of the fission machinery. We show that Rab24 directly interacts with FIS1, thereby ensuring efficient recruitment of DRP1 to mitochondrial membranes to drive the fission process. Reduction of Rab24 causes reduced mitochondrial fission resulting in less mitophagy, increased mitochondrial density and a more branched network capable of higher respiration. At the same time, we observed an induction of autophagic flux under Rab24 knockdown, indicating enhanced energy usage. The induction of macroautophagy combined with enhanced mitochondrial network formation and activity are characteristics of liver starvation 20 . Therefore, we propose that Rab24 knockdown reassembles the fasting state, whereby mitochondria are metabolically reprogrammed towards higher respiration through enhanced connectivity and bioenergetic efficiency. On the other hand, the accumulation of Rab24 in obese patients and patients with NAFLD, and obese mouse models could lead to a situation where autophagy is blocked 47,48 and mitochondrial connectivity is reduced 49,50 , collectively contributing to enhanced energy storage. In fact, reduction in DRP1-mediated fission has been shown to improve mitochondrial fitness in diabetes-related complications 51,52 and in Alzheimer's disease 53 . Clearly, completely reducing fission thus preventing mitophagy is not favourable for maintaining healthy mitochondria and cell survival. However, a slight reduction in fission, as shown for Rab24 knockdown, can shift mitochondria to a more connected and active state and improve their function in diseases with reported mitochondrial dysfunctions.
Based on these data and our previous findings of a regulatory role of Rab5 on gluconeogenesis 14,15 , we propose another level of metabolic control through membrane trafficking regulators, which represents an emerging concept that extends beyond liver metabolism 6 . In fact, the translocation of the insulin-responsive glucose transporter 4 (GLUT4) in fat and muscle is dependent on proper Rab10 function, which is fundamental for regulating glucose uptake in peripheral tissues 54,55 . In addition, clathrin heavy chain 2 variants in humans are associated with altered GLUT4 trafficking and correlate with features of type 2 diabetes 56 . Defective LDL uptake in the liver, due to altered LDL receptor trafficking in patients with mutations in the CCC complex (coiled-coil domain-containing protein 22), highlights the important contribution of trafficking regulators in hypercholesterolaemia 57 and atherosclerosis in humans 58 . Altogether, these data emphasize a thus far rather unexplored connection between membrane transport processes and whole-body energy homeostasis that has to be conceptually exploited for treatment options in type 2 diabetes and its related complications 6 .

Methods
Human samples. In the first cohort, we investigated RAB24 mRNA expression in liver tissue samples obtained from 40 extensively characterized men (n = 23) and women (n = 17) of white ancestry with a wide range of BMI (22.7-45.6 kg m −2 ) who underwent open abdominal surgery for Roux en-Y bypass, sleeve gastrectomy, elective cholecystectomy or explorative laparotomy. BMI was calculated as the weight divided by the squared height. Hip circumference was measured over the buttocks; waist circumference was measured at the midpoint between the lower ribs and iliac crest. Percentage body fat was measured by dual-energy X-ray absorptiometry or bioimpedance analysis. In addition, abdominal visceral and subcutaneous fat areas were calculated using computed tomography or magnetic resonance imaging scans at the level of L4-L5. Using oral glucose tolerance tests, we identified individuals with normal glucose tolerance (n = 40). The methods regarding phenotypic characterization have been described previously 59 . Insulin sensitivity was assessed with the euglycaemic-hyperinsulinaemic clamp method. After an overnight fast and resting for 30 min in a supine position, intravenous catheters were inserted into the antecubital veins of both arms. One line was used for the infusion of insulin and glucose; the other line was used for frequent sampling of arterialized (heating pads) blood. After a priming dose of 1.2 nmol m −2 insulin, infusion with insulin (Actrapid 100 U ml −2 ; Novo Nordisk) was started with a constant infusion rate of 0.28 nmol m −2 body surface per min and continued for at least 120 min. After 3 min, the variable 20% glucose infusion rate was added and adjusted via the clamp to maintain a blood glucose level of 5.5 mmol l −1 (±5%). Bedside blood glucose measurements were carried out every 5 min. The GIR was calculated from the last 45 min of the clamp, where GIR could be kept constant to achieve the target plasma glucose concentration of 5.5 mmol l −1 . Therefore, the duration of the clamp varied between individuals (range, 120-200 min). In premenopausal women, clamp studies were performed during the luteal phase of the menstrual cycle 60 .
All baseline blood samples were collected between 8 and 10 a.m. after an overnight fast. Samples were immediately centrifuged and stored at −80 °C until further analyses were performed. Plasma glucose levels were measured using the hexokinase method. Insulin was measured using the chemiluminescence assay. HDL and LDL cholesterol were measured using enzymatic assays (Cobas; Roche Diagnostics). C-reactive protein was quantified using an Image Automatic Immunoassay System (Beckman Coulter). Circulating levels of high-sensitivity interleukin-6, leptin (R&D Systems) and total adiponectin (ALPCO) were determined in all blood samples with enzyme-linked immunosorbent assays (ELISAs) according to the manufacturer's instructions. The index for homeostatic model assessment of insulin resistance was calculated from fasting plasma insulin and glucose measurements.
All study protocols were approved by the Ethics Committee of the University of Leipzig (nos. 363-10-13122010 and 017-12-230112). All participants gave written informed consent before taking part in the study.
The second cohort comprised 36 individuals who were obese and undergoing bariatric surgery and 8 lean healthy humans (controls) undergoing elective surgery such as cholecystectomy or herniotomy (registered clinical trial no. NCT01477957), part of which was previously reported 16,61 . Individuals who were obese were further classified into individuals who had steatosis (NAFL + ), did not have steatosis (NAFL − ) and had NASH based on liver histology as described in refs. 16,61 . They gave written informed consent before being included in the study, which was approved by the Heinrich Heine University Düsseldorf Institutional Review Board. All participants maintained stable body weight for at least two weeks before surgery and were studied using hyperinsulinaemic-euglycaemic clamps to measure peripheral and hepatic insulin sensitivity and sample blood for routine lab parameters 16 . Participants were asked to refrain from physical activity for 3 d before the clamp test. Volunteers with severe renal, heart or lung disease, acute or chronic inflammatory conditions or any history or signs of liver disease other than NAFLD were excluded from participation. Liver samples used to measure liver fat content from histology, hepatic mitochondrial function and oxidative stress were obtained during surgery as described previously 16 .
Animals. All animal studies were conducted in accordance with German animal welfare legislation. Male C57BL/6N mice obtained from the Charles River Laboratories were maintained in a climate-controlled environment with specific pathogen-free conditions under 12-h dark-light cycles in the animal facility of the Helmholtz Center. Protocols were approved by the institutional animal welfare officer and the necessary licences were obtained from the state ethics committee and government of Upper Bavaria (nos. 55.2-1-55-2532-49-2017 and 55.2-1-54-2532.0-40-15). Mice were fed ad libitum with regular rodent chow. Mice for the HFD studies received an HFD or LFD control from Research Diets for 15 or 17 weeks, starting at the age of 4 weeks, according to the following diet composition: LFD: 16% protein, 73% carbohydrate, 11% fat in kcal; HFD: 16% protein, 25% carbohydrate, 58% fat in kcal. Mice for the NASH studies received an L-amino acid diet with 0.1% methionine and no added choline or LFD control from Research Diets for 4 weeks starting at the age of 6 weeks, according to the following diet composition: LFD: 18% protein, 71% carbohydrate, 10% fat; MCD: 18% protein, 21% carbohydrate, 62% fat in kcal. All experiments were carried out using male mice with littermates as controls.
Antibodies and reagents. The following primary antibodies were purchased: Rab24 (catalogue no. ab154824; Abcam); mitofusin 1 (catalogue no. ab104274; Abcam); TOMM20 (catalogue no. ab78547; Abcam); vinculin (catalogue no. Phosphorothioate linkages were introduced using commercially available 3-((dimethylaminomethylidene)amino)-3H-1,2,4-dithiazole-3-thione (AM Chemicals). Oligonucleotides were synthesized 'DMT off ' and deprotection was carried out according to published procedures 63 . Crude oligonucleotides were purified by anion exchange high-performance LC (HPLC) and analysed by reversed-phase HPLC for purity and electrospray ionization MS for identity. To generate siRNAs from RNA single strands, equimolar amounts of complementary sense and antisense strands were mixed and annealed in a 20 mM NaCl (Sigma-Aldrich), 4 mM sodium phosphate pH 6.8 buffer (Sigma-Aldrich). siRNAs were further characterized by size exclusion HPLC and were stored frozen until use. The nanoparticle formulations were prepared by combining the lipid solution with the siRNA solution at a total lipid-to-siRNA weight ratio of 7:1. The lipid ethanolic solution was rapidly injected into aqueous siRNA solution to produce a suspension containing 33% ethanol. The solutions were injected using a syringe pump (Harvard Pump 33 Dual Syringe Pump; Harvard Apparatus). Subsequently, the formulations were dialysed twice against PBS, pH 7.4 at volumes 200 times of the primary product using a Slide-A-Lyzer Dialysis Cassette (Thermo Fisher Scientific) with an molecular weight cut-off of 10 kDa (regenerated cellulose membrane) to remove ethanol and achieve buffer exchange. The first dialysis was carried at room temperature for 3 h; then, the formulations were dialysed overnight at 4 °C. The resulting nanoparticle suspension was filtered through a 0.2 µm sterile filter (Sarstedt) into glass vials and sealed with a crimp closure.
Particle size and ζ-potential of formulations were determined using a Zetasizer Nano ZS (Malvern Panalytical) in 1× PBS and 15 mM PBS, respectively. The siRNA concentration in the liposomal formulation was measured by ultraviolet-visible spectroscopy. Briefly, 100 µl of the diluted formulation in 1× PBS was added to 900 µl of a 4:1 (v/v) mixture of methanol (Sigma-Aldrich) and chloroform (Sigma-Aldrich). After mixing, the absorbance spectrum of the solution was recorded between 230 nm and 330 nm on a DU 800 Spectrophotometer (Beckman Coulter). The siRNA concentration in the liposomal formulation was calculated based on the extinction coefficient of the siRNA used in the formulation and on the difference between the absorbance at a wavelength of 260 nm and the baseline value at a wavelength of 330 nm.
Encapsulation of siRNA by the nanoparticles was evaluated using the Quant-iT RiboGreen RNA assay (Invitrogen). Briefly, the samples were diluted to a concentration of approximately 5 µg ml −1 in Tris EDTA (TE) buffer (10 mM Tris-HCl, 1 mM EDTA, pH 7.5; Sigma-Aldrich); 50 µl of the diluted samples were transferred to a polystyrene 96-well plate, then either 50 µl of TE buffer or 50 µl of a 2% Triton X-100 (Sigma-Aldrich) solution was added. The plate was incubated at a temperature of 37 °C for 15 min. The RiboGreen reagent was diluted 1:100 in TE buffer and 100 µl of this solution was added to each well. Fluorescence intensity was measured using a fluorescence plate reader (Wallac VICTOR 1420 Multilabel Counter; PerkinElmer) at an excitation wavelength of approximately 480 nm and an emission wavelength of approximately 520 nm. The fluorescence values of the reagent blank were subtracted from that of each of the samples and the percentage of free siRNA was determined by dividing the fluorescence intensity of the intact sample (without the addition of Triton X-100) by the fluorescence value of the disrupted sample (caused by the addition of Triton X-100).
Hepa1-6 cells were obtained from ATCC (in partnership with LGC Standards) and cultured in ATCC-formulated DMEM (ATCC in partnership with LGC Standards) supplemented to contain 10% fetal calf serum (FCS, ultra-low immunoglobulin G) from Thermo Fisher Scientific and 1% penicillin-streptomycin (Biochrom) at 37 °C in an atmosphere with 5% CO 2 in a humidified incubator. To transfect the Hepa1-6 cells with siRNA, cells were seeded at a density of 20,000 cells per well in 96-well regular tissue culture plates. Transfection of siRNA was carried out with Lipofectamine 2000 Transfection Reagent (Thermo Fisher Scientific) according to the manufacturer's instructions. Dose-response experiments were done with final Rab24 siRNA concentrations of 24, 6, 1.5, 0.375, 0.0938, 0.0234, 0.0059, 0.0015, 0.0004 and 0.0001 nM. Control wells were transfected with firefly luciferase, Renilla luciferase or AHA-1 siRNA, or mock-treated. For each siRNA and controls, four wells were transfected in parallel and individual data points were collected from each well; 24 h posttransfection, media were removed and cells were lysed in 150 µl lysis mixture (1 volume lysis mixture, 2 volumes nuclease-free water) and then incubated at 53 °C for 60 min. A branched DNA assay was performed according to the manufacturer's instructions (Thermo Fisher Scientific). Luminescence was read using a Wallac VICTOR 1420 Luminescence Counter (PerkinElmer) following 30 min incubation in the dark. For each well, the Rab24 mRNA level was normalized to the glyceraldehyde 3-phosphate dehydrogenase (GAPDH) mRNA level. The activity of a given Rab24 siRNA was expressed as the percentage Rab24 mRNA concentration (normalized to GAPDH mRNA) in treated cells, relative to the mean Rab24 mRNA concentration (normalized to GAPDH mRNA) averaged across the control wells.
In vivo LNP injections. Eight-week-old male C57BL/6N mice, 36-week-old male FGF21 knockout mice (C57BL/6 background) 64,65 and heterozygous male littermates as control or male C57BL/6N mice fed an HFD for 13 weeks (HFD starting at the age of 4 weeks) received either PBS or siRNA in LNP formulations (either Rab24 or luciferase control) at 0.5 mg kg −1 via tail vein injection as described previously 14,15,66,67 . Intraperitoneal glucose tolerance tests (IPGTTs), pyruvate tolerance tests (PTTs) and insulin tolerance tests (ITTs) were performed in C57BL/6N mice 5 d post-injection using 2 g kg −1 glucose, 2 g kg −1 pyruvate or 0.75 U kg −1 insulin (Lilly) after starvation of 6 h for GTT and ITT or 16 h for PTT. On day 6 post-injection, after 6 h fasting, mice were killed using cervical dislocation and serum and tissue were collected and snap-frozen in liquid nitrogen. HFD mice were treated for 2 (for lipid parameters) or 4 (for IPGTT) weeks with weekly injection of LNPs starting at 13 weeks of an HFD. Male C57BL/6N Mice on an MCD diet were injected weekly contemporaneously to the diet starting at 6 weeks of age.
To study insulin sensitivity of the liver and peripheral tissue, mice starved for 6 h were injected with 0.75 U kg −1 insulin or PBS and killed after 7 min by freeze clamping. Liver, gastrocnemius muscle and epididymal fat were collected for immunoblot analysis.
For all studies, LNPs at 0.5 mg kg −1 were injected through the tail vein.

Proteomics and bioinformatics.
For the proteomics analysis, liver tissue from luciferase control and Rab24 knockdown mice, starved for 6 h, was obtained 5 d after LNP injection and snap-frozen in liquid nitrogen; 50 µg protein were solubilized in 3× volume of lysis buffer (4% sodium deoxycholate, 100 mM Tris pH 8.5, heated for 5 min at 95 °C and sonicated (Branson probe sonifier, output 3-4, 50% duty cycle, 3 × 30 s)). Protein was reduced and alkylated for 15 min at room temperature with 10 mM tris(2-carboxyethyl)phosphine and 40 mM 2-chloroacetamide and digested with LysC and trypsin 1:50 (protein:enzyme) overnight at 37 °C. The digested peptides were acidified to a final concentration of 1% trifluoroacetic acid (TFA). The peptide solution was cleared by centrifugation and loaded onto activated (30% methanol, 1% TFA) double-layer styrenedivinylbenzene-reversed-phase sulphonated STAGE tips (3 M Empore) 64 . The STAGE tips were first washed with 200 µl 0.2% TFA, then with 200 µl 0.2% TFA and 5% ACN. The peptides were eluted with 60 µl styrene-divinylbenzene-reversedphase sulphonated elution buffer (80% ACN, 5% NH 4 OH) for single-shot analysis. For MS analysis, 2 µg peptides were loaded onto a 50-cm column with a 75 µM inner diameter, packed in-house with 1.9 µM C18 ReproSil particles (Dr. Maisch) at 60 °C. The peptides were separated by reversed-phase chromatography using a binary buffer system consisting of 0.1% formic acid (buffer A) and 80% ACN in 0.1% formic acid (buffer B). Peptides were separated on a 120 min gradient (5-30% buffer B over 95 min; 30-60% buffer B over 5 min) at a flow rate of 300 nl on an EASY-nLC 1200 system (Thermo Fisher Scientific). MS data were acquired using a data-dependent top-15 method with maximum injection time of 20 ms, a scan range of 300-1650 Thomson (Th) and an automatic gain control target of 3 × 10 6 . Sequencing was performed via higherenergy collisional dissociation fragmentation with a target value of 1 × 10 5 and a window of 1.4 Th. Survey scans were acquired at a resolution of 60,000. The resolution for the higher-energy collisional dissociation spectra was set to 15,000 with a maximum ion injection time of 28 ms and an underfill ratio of either 20 or 40%. Dynamic exclusion was set to 30 s.
Raw MS data were processed with MaxQuant v.1.5.6.4 using default settings unless otherwise stated. The FDR at the protein, peptide and modification level was set to 0.01. Oxidized methionine and acetylation (N-term protein) were selected as variable modifications, and carbamidomethyl as fixed modification. Three missed cleavages for protein analysis and five for phosphorylation analysis were allowed. Label-free quantitation and 'Match between runs' were enabled. Proteins and peptides were identified with a target-decoy approach in reversed mode, using the Andromeda peptide search engine integrated into the MaxQuant environment. Searches were performed against the mouse UniProt Hierarchical clustering, one-dimensional annotation enrichment and Fisher's exact test were performed in Perseus.
Pulldown assay. Rosetta (DE3) Escherichia coli competent cells were transformed with pET-60-DEST (Novagen). Expression of GST-Rab24 and GST-Rab3a was induced by 1 mM isopropylthiogalactoside at an optical density (600 nm) of 0.5 and bacteria were incubated for 4 h at 37 °C. Bacterial pellets were resuspended in lysis buffer (20 mM HEPES, pH 7.5, 100 mM NaCl, 2 mM MgCl 2 , 100 µg ml −1 lysozyme) followed by incubation for 30 min at 4 °C. After adding 0.1 mM phenylmethylsulfonyl fluoride and 1 mM dithiothreitol, bacterial lysis was obtained by sonication at an amplitude of 50% for 5 min (30 s sonication and 30 s break). Batch purification of GST-tagged proteins was performed on Glutathione Sepharose 4B (GE Healthcare Life Sciences). Beads were equilibrated with purification buffer (20 mM HEPES, pH 7.5, 100 mM NaCl, 2 mM MgCl 2 ) and bacterial lysates were incubated on beads (50:1) overnight at 4 °C with end-overend rotation. Loaded beads were then washed 5× in purification buffer. The purity of GST-Rab24 and GST-Rab3a was analysed by MS.
Nucleotide exchange was obtained by incubating mouse liver lysates with 1 mM EDTA for 10 min at 37 °C followed by 10 min at 37 °C in presence of 5 mM MgCl 2 and 1 mM 5′-guanylylimidodiphosphate (GMP-PNP) or guanosine 5′-O-(2thiodiphosphate) (GDP-β-S). Nucleotide exchange for immobilized GST-Rab24 and GST-Rab3a was performed as described by Vitale et al. 65 . Briefly, recombinant Rab proteins were incubated with nucleotide buffer (50 mM Tris/HCl, pH 8.0, 150 mM NaCl, 5 mM MgCl 2 , 1 mM EDTA, 1 mM GMP-PNP or GDP-β-S) for 1 h at room temperature followed by incubation with stabilization buffer (50 mM Tris/HCl, pH 8.0, 150 mM NaCl, 2 mM MgCl 2 , 1 mM GMP-PNP or GDP-β-S) for 10 min at room temperature with rotation. For protein binding, prepared mouse liver lysates were supplemented with 1% glycerol and then incubated on GST-Rab24 or GST-Rab3a beads for 1 h at room temperature under rotation with each batch having the corresponding nucleotide state. Beads were washed three times in wash buffer (50 mM Tris/HCl, pH 8.0, 150 mM NaCl, 2 mM MgCl 2 , 10 µM GMP-PNP or GDP-β-S, protease and phosphatase inhibitor). Proteins bound to GST-Rab24 or GST-Rab3a were detected by SDS-polyacrylamide gel electrophoresis (PAGE) and subsequent immunoblotting using the FIS1 and DRP1 antibodies.

Triglyceride concentration.
To measure the amount of triglycerides stored in the liver of luciferase and Rab24 knockdown mice, liver tissue was snap-frozen in liquid nitrogen 5 d after LNP injection and the triglyceride level quantified with the triglyceride assay kit from Abcam, according to the manufacturer's instructions.
Human RAB24 mRNA expression studies. Human RAB24 mRNA expression was measured by quantitative reverse transcription PCR (RT-PCR) in a fluorescent temperature cycler using the TaqMan assay; fluorescence was detected on an ABI PRISM 7000 sequence detector (Applied Biosystems). Total RNA was isolated using TRIzol (Thermo Fisher Scientific). Quantity and integrity of RNA was monitored with a NanoVue Plus Spectrophotometer (GE Healthcare Life Sciences); 1 µg total RNA from liver tissue samples was reverse-transcribed with standard reagents (Thermo Fisher Scientific). Complementary DNA (cDNA) was then processed for the TaqMan probe-based quantitative RT-PCR using the QuantStudio 6 Flex Real-Time PCR System (Thermo Fisher Scientific). mRNA expression was measured by quantitative RT-PCR using the Hs01585710_g1 probe for RAB24. Fluorescence emissions were monitored after each cycle. Human RAB24 mRNA expression was calculated relative to the mRNA expression of 18S ribosomal RNA (Hs99999901_s1).
Quantitative RT-PCR analysis using SYBR Green. Liver tissue or cultured primary hepatocytes were lysed using TRIzol to extract RNA. cDNA was transcribed using the SuperScript III Reverse Transcriptase Kit from Thermo Fisher Scientific. RT-PCR was conducted in 10 µl of total reaction volume containing SYBR Green (Thermo Fisher Scientific), 200 nM of forward and reverse primers (Supplementary Table 3) and 24 ng of total cDNA. The reaction was performed using the QuantStudio6 system (Thermo Fisher Scientific) with the following thermal cycling conditions: 50 °C for 2 min; 95 °C for 10 min; 95 °C for 15 s; and 60 °C for 1 min (40 cycles). Relative mRNA levels were quantified by calculating the comparative 2 -ΔΔCt method.
Histology. Liver samples of luciferase control and Rab24 knockdown mice were collected 5 d after LNP administration and snap-frozen in liquid nitrogen. Excised specimens were fixed in 4% (w/v) neutrally buffered formalin (Sigma-Aldrich), embedded in paraffin (SAV LP) and cut into 3-µm slices for hematoxylin and eosin (H&E) staining or for immunohistochemistry. Immunohistochemical staining was performed under standardized conditions on a Discovery XT automated stainer (Ventana Medical Systems) using rabbit anti-prohibitin (1:200, catalogue no. ab28172; Abcam) as a primary antibody and Discovery Universal (Ventana Medical Systems) as secondary antibody. Signal detection was conducted using the Discovery DAB Map Kit (Ventana Medical Systems). The stained tissue sections were scanned with an Axio Scan.Z1 digital slide scanner (ZEISS) equipped with a 20× magnification objective. Images were evaluated with the commercially available image analysis software Definiens Developer XD2 (Definiens) following a previously published procedure 66 . The calculated parameter was the mean brown staining intensity of the stained tissue.
For the cryostat sections, control and Rab24 knockdown livers were defrosted in 4% paraformaldehyde (PFA), rotated in 4% PFA at 4 °C overnight and placed in PBS until further use. Before embedding them in optimal cutting temperature compound, liver specimens were transferred to 30% sucrose for 3 d. Then, 5-7 µm cryostat sections were cut and collected on Superfrost Plus-treated slides. Cryosections were immunolabelled with first and secondary antibodies incubated in PBS BSA (3%) for 2 h and 1 h, respectively, at room temperature. Then, sections were mounted with MOWIOL 4-88 reagent (Merck Millipore) onto coverslips.
Hepatocyte isolation and transfection. Primary hepatocytes were isolated via collagenase perfusion from 8-12-week-old male C57BL/6N mice 67 . Briefly, mice were anaesthetized, both abdominal walls were opened and the liver was perfused through the venae cavae with EGTA-containing HEPES/KH buffer for 10 min, followed by a collagenase-containing HEPES/KH buffer for 10-12 min until liver digestion was visible. The perfused liver was cut out and placed into a suspension buffer-containing dish and hepatocytes were gently washed out. After filtering the cells through a 100-nm pore mesh, cells were centrifuged and washed twice and resuspended in suspension buffer. For a detailed isolation protocol including pictures, please see Godoy et al. 69 .
Two hundred thousand cells per well were plated in collagen-coated 24-well plates (Thermo Fisher Scientific) in William's Medium E (PAN-Biotech) containing 10% FCS (PAN-Biotech), 5% penicillin-streptomycin (Thermo Fisher Scientific) and 100 nM dexamethasone (Sigma-Aldrich) and maintained at 37 °C and 5% CO 2 . After 1 h, cells were washed with PBS (Thermo Fisher Scientific) and incubated with 40 nM siRNA (Rab24 and ubiquitin-like modifier-activating enzyme ATG7) or 0.1 nM FIS1 (Rab24 obtained from Axolabs; FIS1 and ubiquitin-like modifieractivating enzyme ATG7 from Dharmacon) and interferin (1.2 µl well −1 ) (Biomol) in William's Medium E. After 6 h of incubation, cells were washed with PBS and a second layer of collagen was added to maintain cells in a sandwich culture. William's Medium E was changed twice a day 15,18 .
HFD in vitro conditions were obtained by supplementing William's Medium E with 100 µM palmitate and 400 µM oleate (both conjugated with BSA 1:6) (Sigma-Aldrich) or the respective amount of BSA.
Electron microscopy. Electron microscopy was performed at the CCMA EM Core Facility (Université de Nice Sophia Antipolis). Primary hepatocytes silenced or not silenced for Rab24, were fixed in 2.5% glutaraldehyde (Electron Microscopy Sciences) overnight at room temperature. Then, cells were post-fixed in potassium ferrocyanide-reduced osmium tetroxide (Electron Microscopy Sciences) for 1 h and dehydrated in a bath with increasing concentrations of ethanol (VWR). Cells were incubated with a mixture of epon:ethanol (1:2 and 2:1) and finally embedded with pure epon (Electron Microscopy Sciences) 14,17 . Sections of 70 nm were obtained using a Leica ultramicrotome and counter-stained with uranyl acetate and lead citrate. Images of at least 30 cells per conditions were acquired.
The numbers of mitochondria were counted manually for each image and the areas were analysed with Fiji software (ImageJ, v.2.0.0-rc-69/1.52p). For the morphological description of the mitochondria, the area, perimeter, width, height and Feret's diameter were extracted from Fiji by manually segmenting the mitochondria. Then, these values were used to calculate the mitochondrial form factor (perimeter 2 /4π × area), the mitochondrial aspect ratio (width/height), mitochondrial circularity (4π × area/perimeter 2 ) and mitochondrial roundness (4 × area/π × width) 70 . The number of mitophagy events, easily recognizable by morphology, were counted manually on the same set of images and reported to the area analysed.
Cells in monolayer culture were fixed for 15 min in PFA, washed twice for 5 min in PBS and permeabilized in 0.1% Triton X-100 in PBS for 10 min at room temperature. After two more washes for 5 min in PBS, cells were blocked in 10% horse serum for 10 min at room temperature and subsequently treated with primary antibodies in 5% horse serum for 1 h. Cells were washed 3 times for 5 min in PBS and incubated for 1 h with secondary antibodies (1:1,000) at room temperature. Subsequently, cells were washed twice with PBS and stained with DAPI, then mounted onto glass slides with 0.1 g ml −1 MOWIOL 4-88 reagent.
For the LDL uptake assay, cells were starved for 2 h in William's Medium E without serum or dexamethasone. Continuous uptake of DiI-LDL (approximately 2.5 µg ml −1 ) for several time points was performed, cells were washed with cold PBS and fixed in 4% PFA 14 . Cells were stained with DAPI and mounted with MOWIOL 4-88 reagent.
Confocal microscopy and analysis. Immunofluorescent samples were analysed using a laser scanning confocal microscope (Olympus FluoView 1200; Olympus Corporation) equipped with an Olympus UPlanSAPO ×60 1.35 and an UPlanSAPO ×40 1.25 solid immersion lens oil immersion objective (Olympus) at a resolution of approximately 100 µm pixel −1 (×60) and 600 nm step size. Quantification was performed for individual images after background subtraction with a minimum of 30 cells using ImageJ. For particle quantification using the ImageJ plug-in, fluorescent dots with a pixel 2 from 0.1 to 10 and circularity from 0.0 to 1.0 were included in the analysis. The mean dot fluorescence per cell was calculated by dividing the overall intensity of particles with the cellular area in the same field.

Pearson's and Spearman's correlation coefficients analysis and statistics.
To quantify the colocalization between signals from TOM20/KDEL, TOM20/ LAMP1 and DRP1/TOM20, Pearson's and Spearman's correlation coefficients were calculated by using the Pearson's and Spearman's correlation colocalization plug-in of ImageJ v.1.52e (NIH) according to French et al. 71 . At least 10 individual cells in 3 independent images and a minimum of 500 signals were considered for every analysis. Calculated Pearson's and Spearman's correlation coefficient values were between −1 (negative correlation) and +1 (positive correlation); the threshold level, under which pixels were treated as background noise, was set at 10. The results were presented as scatterplots with Pearson's and Spearman's coefficients (r p and r s ).
Mitochondrial three-dimensional (3D) reconstruction and morphometric analysis. For mitochondrial 3D reconstruction, images were deconvolved using the Fiji plug-in's point spread function (PSF) generator 72 and DeconvolutionLab (v.2.1.2) 73 (EPFL; http://bigwww.epfl.ch/). The Z-step was set to 0.6 μm and a PSF algorithm (Born & Wolf 3D optical model) was used for PSF generation. The generated PSF and a 3D deconvolution algorithm (Richardson-Lucy with TV regularization) were applied to the microscopic images using DeconvolutionLab. From the deconvolved two-dimensional (2D) and 3D binary images (8-bit images), the mitochondrial network was determined by generating a skeleton of the images using the Fiji plug-in Skeletonize3D (v.2.1.1) and analysed using the plug-in AnalyzeSkeleton (2D/3D) (v.3.3.0). This plug-in tags all pixel/voxels in a skeleton image, counts the junctions and branches of the mitochondrial network and measures their average length. For mitochondrial network analysis, at least 20 cells were analysed.
Seahorse assays. Mitochondrial respiration and glycolysis were measured using the Seahorse XFe 24 Analyzer (Agilent Technologies). Therefore, 30,000 cells per well were plated as a monolayer culture in Seahorse cell plates precoated with collagen. Oxidative phosphorylation and glycolysis analysis were conducted 2 d after siRNA knockdown of luciferase control or Rab24. To measure in vivo mitochondrial respiration, primary hepatocytes were isolated from control or Rab24 knockdown mice after 5 d of RNAi treatment and directly measured. A mitochondrial stress test was performed by injection of 2 µM oligomycin (Sigma-Aldrich), 1 µM FCCP (Sigma-Aldrich) and 1 µM antimycin A (Sigma-Aldrich) + 1 µM rotenone (Sigma-Aldrich). To measure oxidative phosphorylation after blocking autophagy, cells were incubated 1 h with 25 µM chloroquine before the Seahorse assay. Glycolysis was measured by additional injection of 100 mM 2-deoxy-d-glucose (Sigma-Aldrich).
Cholesterol, bile acid, lactate and FGF21 secretion. Dexamethasone-free primary hepatocytes in the collagen sandwich were washed and starved for 5 h in William's Medium E without serum 3 d after RNAi. Medium was collected and the amounts of cholesterol, bile acids and FGF21 were determined according to the manufacturer's instructions using the Amplex Red Cholesterol Assay Kit (Thermo Fisher Scientific), the Total Bile Acid Kit from Cell Biolabs and the Mouse/Rat FGF-21 Quantikine ELISA Kit from R & D Systems, respectively. Lactate concentration was measured with the Lactate Assay Kit from Sigma-Aldrich, according to the manufacturer's instructions. The amounts of cholesterol, bile acids, lactate and FGF21 were normalized to the total protein levels of the cells. Protein levels were obtained using the DC Protein Assay kit (Bio-Rad Laboratories).
Mitochondrial assays. Primary hepatocytes cultured for 3 d after RNAi treatment or isolated hepatocytes from control or Rab24 knockdown mice (5 d post-RNAi) were treated for 30 min with 250 nM MitoTracker Red and 200 nM MitoTracker Green to determine mitochondrial mass and function, normalized to cell number measured with DAPI fluorescence.
Mitochondrial degradation was assessed using the nucleoid depletion assay 33 . To stimulate mitophagy, 3 d after siRNA treatment primary hepatocytes were incubated with 20 µM FCCP for 1 h. After fixation, total DNA and nuclear DNA (DAPI) were stained to determine cytosolic DNA.
Mitophagy was measured by incubating the cells (30,000 cells per well in a 96-well plate) with 20 µM FCCP alone or with 20 µM chloroquine for 2 h. Accumulation of autophagosomes was measured with the CYTO-ID Autophagy detection kit (Enzo).

Autophagy flux.
To determine autophagy flux in vivo, control and Rab24 knockdown mice were either fed ad libitum, starved for 12 h or starved for 12 h + 3 h chloroquine treatment (100 mg kg −1 ). Chloroquine or PBS as control (fed and fasted group) were injected intraperitoneally. Mice were killed simultaneously; liver specimens snap-frozen in liquid nitrogen and the concentration of the autophagy markers LC3 and p62 were measured via immunoblot.
Autophagy flux in vitro was measured in primary hepatocytes cultured in the collagen sandwich 4 d after siRNA treatment. Cells were either kept in full medium, serum and dexamethasone-starved for 12 h, or starved 12 h + chloroquine treatment (20 µM) for 1 h. Cells were either lysed for immunoblot analysis or fixed in PFA for immunofluorescence.

ATP, ROS and carbonylation assay.
To measure ATP, ROS production and carbonylation of proteins, 30,000 cells per well were cultured in a 96-well plate. Three days after, RNAi concentrations of ATP, ROS and carbonylation were detected using the respective kits according to the manufacturers' instructions: CellTiter-Fluor Cell Viability Assay (Promega Corporation); DCFDA Cellular ROS Detection Assay Kit (Abcam); and Protein Carbonyl Content Assay Kit (Abcam). All were normalized to cell number measured with DAPI fluorescence.
Glucose uptake assay. Glucose uptake was measured in a 96-well plate containing 30,000 cells per well incubated with 10 µM 2-NBDG (Thermo Fisher Scientific) for 1 h to cells in either basal conditions or supplemented with 2 µM oligomycin, 1 µM antimycin + 1 µM rotenone or 25 mM glucose. Cells were washed twice with PBS and the fluorescence of 2-NBDG was measured. 2-NBDG values were normalized to DAPI fluorescence.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The proteomics data generated or analysed during this study are included in this published article (and its Supplementary Information files). Additional data that support the findings of this study are available from the corresponding authors upon reasonable request. Source data for Figs Fig. 1 | RAB24 expression is positively correlated with fat accumulations in humans. (a, b) Spearman's correlation analysis of RAB24 in the liver of a cohort of obese patients versus healthy controls for BMI and Clamp glucose infusion rate (GIR). (N=33 patients) (c, d) Spearman's correlation analysis of RAB24 in liver of patients with NALFD w/ and w/o steatosis and NASH patients versus healthy controls for M-value and liver fat (HCL) levels (N=44 patients). P-value: two-tailed paired Student's t-test. Fig. 3 | Rab24 reduction led to an increase in mitochondrial area. Electron micrographs of mitochondria in primary hepatocytes of control (a, c) and Rab24KD (b, d) cells; higher magnifications shown in c and d. The images are representative of 12 independent wells of a 24-well plate. Scale bar 5 µm. Zoom scale bar 500 nm. Quantification of mitochondria morphology by morpho-EM for mitochondria density (e), surface area (f), perimeter or contour (g), form factor or complexity (h) and circularity (i) of control and Rab24KD cells. N=7 (CTR) and N=12 (KD) cells quantified in (e). N=329 (CTR) and N=453 (KD) in (f), N=108 (CTR) and N=(209) in (g, h, i) mitochondria analyzed. All measured after 3 days of RNAi (40 nM) in primary hepatocytes (mean +/-SEM). *P<0.05, ***P<0.001, by two-tailed unpaired Student's t-test. Fig. 4 | Rab24KD enhanced mitochondrial connectivity. (a) Representative confocal images (single confocal section) of cultured primary hepatocytes after 3 days of RNAi (40 nM) stained with dapi (blue), phalloidin (green) and Tom20 (grey) as single section, deconvolved, zoomed and skeletonized with Fiji. The images are representative of three independent wells of a 24-well plate. (b-d) Quantification of number of branches, number of junctions per area and mean length of the branches of (a) with Fiji from N=10 (CTR) and N=9 (KD) cellular regions. The experiment was done twice with similar results. (e) Representative confocal images (single confocal section) of HFD liver sections stained with dapi (blue), phalloidin (green) and Tom20 (grey) as single section, deconvolved, zoomed and skeletonized with Fiji. The images are representative of four independent biological samples, which give rise to the quantifications in f-h. (f-h) Quantification of number of branches, number of junctions per area and mean length of the branches of (e) using Fiji from N=4 animals per condition. Measured after 14 days of KD with LNPs (weekly injection; 0.5 mg/kg) and 6 h starvation in 15-week HFD mice. Scale bar 20 µm (mean +/-SEM). **P<0.01, ***P<0.001, ****P<0.0001 by two-tailed unpaired Student's t-test. Fig. 5 | Rab24KD enhanced mitochondrial connectivity without affecting the fusion machinery. (a) Abundance of fusion and fission regulators by proteomics. N=6 animals per condition. (b) Representative confocal images (single middle confocal sections) of primary polarized hepatocytes in collagen sandwich stained for Tom20 (green), KDEL (red), actin (grey) and dapi. Higher zoom shown in the middle insert to better observe co-localization. The images are representative of three independent wells of a 24-well plate. Person's and Spearman's correlation analyzed with Fiji representing overlap between the green and red channel on the right and quantification of PSC in (c) analyzed from N=3 wells per condition. The experiment was done twice with similar results. (d, e) Representative confocal images (single confocal sections) of primary hepatocytes stained for Mfn1 (d) and Mfn2 (e). The images are representative of three independent wells of a 24-well plate. (f) Quantification of the mean fluorescent intensity per cell from (d, e) using Fiji from N=22 (CTR) and N=27 (KD) (Mfn1) and N=24 (CTR) and N=29 (KD) (Mfn2) cells per condition. Scale bar = 20 μm. All measured after 3 days of RNAi (40 nM) in primary hepatocytes (mean +/-SEM). *P<0.05, *****P<0.00001 by two-tailed unpaired Student's t-test. Only data that reached statistical significance are indicated. Fig. 7 | Rab24 depletion reduced fluorescence overlap between mitochondria and lysosomes. (a) Mitophagy flux assay in primary hepatocytes incubated for 2 h with 20 µM FCCP w/ or w/o 20 µM chloroquine. N=6 wells per condition. (b) Representative confocal images (single middle confocal sections) of primary polarized hepatocytes in collagen sandwich stained for Tom20 (green), Lamp1 (red) and dapi. Higher zoom shown as insert. Scale bar 20 µm. The images are representative of three independent wells of a 24-well plates. Person's and Spearman's correlation analyzed with Fiji representing overlap between the green and red channel in (c) analyzed from N=19 (CTR) and N=16 (KD) cellular regions (multiple cells per region). The experiment was done twice with similar results. All measured after 3 days of RNAi (40 nM) in primary hepatocytes (mean +/-SEM). *P<0.05, **P<0.01 by two-tailed unpaired Student's t-test.