Endosomal RFFL ubiquitin ligase regulates mitochondrial morphology by targeting mitofusin 2 Open Access

Mitochondria are motile and dynamic organelles that play crucial roles in diverse cellular processes, including ATP generation, metabolism and Ca²⁺ signaling (Cosentino and García-Sáez, 2014; McBride et al., 2006; Nunnari and Suomalainen, 2012). The dynamics of these organelles encompass changes in their morphology, degradation, fission and fusion (Dorn, 2019; Quintana-Cabrera and Scorrano, 2023). Although distinct signaling events contribute to the dynamics, dysregulation of the same has been reported in numerous pathological conditions in humans (Knott et al., 2008; Youle and van der Bliek, 2012). Whereas diseases that are associated with defects in mitochondrial dynamics range from neurodegenerative disease to cancer, primary mitochondrial dysfunction typically affects the brain, heart and skeletal muscles (Knott et al., 2008; Ng and Turnbull, 2016).

One of the major proteins involved in mitochondrial dynamics is mitofusin 2 (MFN2), an outer membrane mitochondrial GTPase. Mitofusins present on opposing mitochondria via trans interactions (homo or hetero oligomeric complex formation) are believed to mediate tethering between the organelle during fusion (Chen and Chan, 2005). The 757-amino-acid long MFN2 consists of a GTPase domain, two coiled-coil domains (HR1 and HR2) and a proline-rich domain (PR) between the HR1 and transmembrane domains (TM) (Filadi et al., 2018). The GTPase and HR1 domains are in the cytosol, whereas the TM spans the outer mitochondrial membrane, and HR2 and C-terminal reside within the intermembrane space (Mattie et al., 2017). Although MFN2 knockout in mice is embryonic lethal, mutations in MFN2 are associated with multiple disorders, including Charcot–Marie–Tooth disease type 2A (CMT2A), a peripheral neuropathy disease (Cartoni et al., 2010; Chen et al., 2003; Sandoval et al., 2014), along with atherosclerosis and hypertension (Chen et al., 2004; Guo et al., 2007; Shen et al., 2007), exemplifying the importance of MFN2 for cellular homeostasis.

Diverse molecules involved in mitochondrial dynamics are regulated by post-translational modifications, such as phosphorylation, acetylation and ubiquitylation (Chen et al., 2023), including MFN2. Phosphorylation of MFN2 at T111 and S442 by PINK1 facilitates PRKN binding and subsequent ubiquitylation of MFN2 (Chen and Dorn, 2013). MFN2 is also reported to be ubiquitylated by other E3 ligases (Escobar-Henriques and Joaquim, 2019) apart from PRKN (Gegg et al., 2010; Poole et al., 2010; Tanaka et al., 2010; Ziviani et al., 2010), like membrane associated RING-CH-type finger 5 (MARCHF5; also known as MARCH5 and MITOL) and mitochondrial E3 ubiquitin protein ligase 1 (MUL1) (Kim et al., 2015; Li et al., 2008; Sugiura et al., 2013; Tang et al., 2015; Yun et al., 2014). Although PRKN is a cytosolic ubiquitin ligase, MARCHF5 and MUL1 are mitochondrial E3 enzymes. Apart from these, two more cytosolic E3 ligases, RANBP2-type and C3HC4-type zinc finger containing 1 (RBCK1; also known as HOIL1) and HECT, UBA and WWE domain containing e3 ubiquitin protein ligase 1 (HUWE1) have been shown to ubiquitylate MFN2 (Di Rita et al., 2018; Leboucher et al., 2012; Michel et al., 2017; Su et al., 2024). Furthermore, AMFR (also known as GP78), an E3 ligase associated with the ER-associated degradation pathway, also increases MFN2 ubiquitylation and degradation, although direct interaction between these molecules has yet to be demonstrated (Fu et al., 2013; Shankar et al., 2013). Given the links of mitofusins and post-translational modifications with diverse cellular processes, including mitophagy, hypoxic and genotoxic stress, Ca²⁺ signaling, and mitochondrial transport, it is understandable that regulatory pathways of mitofusins by more E3s localized to different compartments are waiting to be discovered.

RFFL (also known as CARP2, RNF189, RNF34L, RIFIFYLIN and Sakura) was initially identified as a palmitoylated E3 ligase containing a RING finger domain in the brain, testes, and endocrine organs alongside its homolog RNF34 (also known as CARP1) (Araki et al., 2003). This discovery was concurrently complemented by the work of McDonald and El-Deiry, who identified RFFL and RNF34 as interacting partners of the death effector domain (DED) of caspase-8 and -10 in yeast two-hybrid screen (McDonald and El-Deiry, 2004). Both RFFL and RNF34 also contain an FYVE-like domain (a cysteine-rich zinc finger found in Fab1, YOTB, Vac1, and EEA1 proteins), which confers the ability to bind phosphatidylinositol 3-phosphate in conjunction with modifications like palmitoylation (Ravindran et al., 2022; Tibbetts et al., 2004). RFFL has been shown to localize to intracellular vesicles, such as Rab5B- and Rab7A-containing endosomes, the plasma membrane and the Golgi (Coumailleau et al., 2004; Liao et al., 2008; Okiyoneda et al., 2018; Ravindran et al., 2022; Sakai et al., 2019). Diverse signaling molecules like the receptor-interacting serine/threonine kinase 1 (RIPK1), tumor protein p53 (TP53), caspase 8 and 10, proline rich 5 like (PRR5L), STUB1, and K⁺ voltage-gated channel subfamily H member 2 (KCNH2; also known as hERG) are targeted by RFFL for ubiquitylation (Gan et al., 2012; Liao et al., 2008; McDonald and El-Deiry, 2004; Roder et al., 2019; Sharma et al., 2023; Yang et al., 2007). Additionally, RFFL is involved in protein quality control, specifically targeting misfolded mutant cystic fibrosis transmembrane conductance regulator (CFTR) protein for ubiquitylation and proteasomal degradation, as shown in research by Okiyoneda et al. (2018).

Earlier studies from our group have demonstrated that RFFL-positive endosomes are associated with damaged mitochondria prior to PRKN recruitment and facilitate the clearance of mitochondrial proteins. RFFL-knockout cells exhibited a delay in the translocation of PRKN to fragmented mitochondria, suggesting an important role for RFFL in maintaining mitochondrial homeostasis (Ravindran et al., 2022). However, no information on the morphology of mitochondria in RFFL KO cells and whether RFFL targets any mitochondrial proteins is available. In this paper, we include evidence from electron micrograph images that show the enlargement of mitochondria in RFFL KO cells. We further demonstrate that RFFL interacts with and ubiquitylates MFN2, resulting in its degradation in cells. Our study, for the first time, shows direct ubiquitylation of MFN2 by RFFL in vitro, using recombinant proteins. Importantly, our findings also unravel hitherto unknown cross-talk between endosomal ubiquitin ligases and MFN2 and further MFN2 biology.

The delayed PRKN recruitment to mitochondria and reduced elimination of mitochondrial proteins observed in RFFL-knockout cells (Ravindran et al., 2022) led us to closely evaluate mitochondrial networks in these cells using confocal microscopy after immunostaining with anti-TOMM20 (one of the outer mitochondrial proteins) antibody and transmission electron microscopy (TEM) because a similar phenotype has been reported in cells with hyperfused mitochondria (Gomes et al., 2011; Rambold et al., 2011). Deconvoluted three-dimensional (3D) images revealed more enlarged or fused mitochondrial networks in A549 RFFL-knockout (KO) cells than in wild-type (WT) cells. Importantly, this phenotype was reversed when RFFL KO cells stably expressed RFFL (reconstituted cells) (Fig. 1A; Movies 1–4). Expression levels of RFFL protein in these cells are shown in Fig. S1A. To evaluate these differences, we measured different parameters, such as the number of mitochondria per cell, the mean volume of mitochondria, and the length and width of mitochondria using methods described previously (Chaudhry et al., 2020; Das et al., 2022a,b; Triolo et al., 2023). Data from these measurements (Fig. 1B–E) confirmed significant differences in mitochondria in cells with or without RFFL expression. Furthermore, we also measured the average length of mitochondria in each cell and categorized them as fragmented, normal or interconnected/hyperfused mitochondria as described previously (Das et al., 2022a; Parra et al., 2008) and plotted the results (Fig. 1F). Again, the data showed a significant increase in interconnected mitochondria in cells that did not express RFFL compared to in the RFFL-expressing cells (Fig. 1F).

We further examined mitochondria in these cells using transmission electron microscopy (TEM) (Fig. 1G). Consistent with the observations from the confocal microscopy images, in RFFL KO cells, mitochondria appeared to be enlarged compared to those in control WT cells. This phenotype was the result of RFFL loss, as mitochondria from KO cells stably reconstituted with RFFL showed morphology similar to that observed in WT cells (Fig. 1G). Conversely, the mitochondria in A549 cells stably expressing RFFL–EGFP appeared to be shorter in length. When the mitochondrial area, perimeter and Feret's diameter from these images were measured, they were found to be significantly higher in RFFL KO cells compared to in WT cells and the KO cells reconstituted with RFFL. Consistent with this, the same parameters of mitochondria in cells stably expressing RFFL–GFP were also significantly lower than that in WT cells alone, suggesting RFFL expression influences the extent of defect in mitochondrial morphology assessed by various parameters measured (Fig. 1H–J; Fig. S1A). These various mitochondrial parameters quantified fall within the ranges reported previously, indicating the robustness of the image analysis (Faitg et al., 2020; Han et al., 2021; Parra et al., 2013; Sidarala et al., 2022). The marking of the mitochondria used for measurements is shown in Fig. S1B. Together, the data shows that mitochondrial morphology is influenced by RFFL.

Given that mitochondria appear different in RFFL KO cells, we hypothesized that proteins involved in mitochondrial dynamics could be potential substrates for RFFL. A number of proteins have been reported to mediate mitochondrial fusion and fission (Chen et al., 2023). To predict potential substrates for RFFL among these proteins, we applied a previously reported protein language model. This model ranked the proteins based on their likelihood of being RFFL substrates by evaluating the masked language modeling (MLM) loss for each protein pair. A lower MLM loss when analyzing protein pairs indicates that the model identifies a coherent or complementary context between them, suggesting a possible interaction (Lin et al., 2023; Lupo et al., 2024). Among the tested proteins, the model ranked MFN2 and MFN1 as the most likely interacting partners of RFFL (Fig. 2A). Interestingly, the hyper-fused mitochondria observed in RFFL KO cells appeared to be very similar to the previously reported phenotype of mitochondria observed in midbrain neurons and brain tissues of Mfn2 transgenic mice (Han et al., 2021; Zhao et al., 2017). This led us to first test whether MFN2 could be a potential substrate for RFFL. The role of MFN2 is well-established in mitochondrial biology, and the localization of MFN2 to mitochondria facilitates fusion (Dorn, 2020; Koshiba et al., 2004). Hence, we evaluated MFN2 localization to mitochondria with and without RFFL in cells by immunostaining with anti-MFN2 antibody and anti-TOMM20 (for mitochondria). Confocal imaging and Pearson's correlation coefficient analyses revealed increased MFN2 colocalization with TOMM20 in RFFL KO cells (Fig. 2B,C). Furthermore, we evaluated the dynamics of RFFL, MFN2 and mitochondria in live cells using confocal imaging. For this, we expressed MFN2–iRFP670 in A549 cells stably expressing RFFL–EGFP and stained the mitochondria with MitoTracker Red CMXRos. As expected, the clustering of mitochondria as a result of MFN2 overexpression was observed (Fig. 2D; Fig. S2A) (Huang et al., 2007). Monitoring of live cells showed an association of RFFL-positive vesicles with MFN2 and the mitochondrial population (Fig. 2D; Fig. S2A, Movies 5, 6). Interestingly, RFFL vesicle–mitochondria contact sites were also found to be positive for MFN2. This association, although transient, appears consistently in various mitochondria (Movies 5 and 6). These observations demonstrate the presence of RFFL vesicles in the close vicinity of MFN2-labeled mitochondria, even under constitutive conditions. Given that RFFL vesicles appear to come into contact with MFN2 on mitochondria, we hypothesized that RFFL can target MFN2 for ubiquitylation. As MFN2 is known to undergo rapid ubiquitylation during PRKN-mediated mitophagy triggered by mitochondrial depolarization (McLelland et al., 2018), we first investigated the effect of RFFL on this ubiquitylation process.

For this, we transfected HEK293T cells stably expressing empty vector (EV) or RFFL with HA–ubiquitin (Ub) and treated the cells with mitochondrial depolarizing reagent carbonyl cyanide 3-chlorophenylhydrazone (CCCP) or DMSO. The endogenous MFN2 was immunoprecipitated under denaturing conditions using an anti-MFN2 antibody, and the precipitates were immunoblotted with an anti-HA antibody. As expected, upon mitochondrial damage conditions, a ubiquitin smear on MFN2 was observed. Importantly, an increased ubiquitin signal was noted in extracts from cells expressing RFFL exogenously compared to controls (Fig. 2E). The extent of ubiquitin modification of MFN2 was calculated and plotted (Fig. 2F). These results indicate that RFFL contributes to MFN2 ubiquitylation in cells. We also checked MFN2 ubiquitylation in the presence of PRKN and found that RFFL was able to increase the ubiquitylation of MFN2 in the presence of PRKN as well (Fig. S2B,C). Next, we investigated whether endogenous RFFL could also contribute to the modification of endogenous MFN2, which has been reported to occur upon treatment with CCCP (McLelland et al., 2018). For this, we assessed the extent of MFN2 modification in A549 WT and RFFL KO cells. As expected, treatment of cells with CCCP resulted in substantial modification of MFN2, appearing as a high molecular mass smear within 1 h of treatment. Importantly, much less MFN2 smear was observed in RFFL KO cells than in WT cells, demonstrating that RFFL contributes significantly to MFN2 modification (Fig. 2G,H). To further explore whether MFN2 forms a complex with RFFL in cells, we immunoprecipitated endogenous RFFL from extracts of WT or KO cells and immunoblotted for endogenous MFN2 (Fig. 2I). Consistent with the changed mitochondrial morphology observed in WT versus KO cells without any treatment (Fig. 1), an RFFL–MFN2 complex was noted in cells treated with DMSO, indicating an interaction between RFFL and MFN2 even under constitutive conditions. A slight reduction in MFN2 associated with RFFL after CCCP treatment was noted, suggesting perhaps enhanced modification and rapid turnover of MFN2 under stress conditions (Fig. 2I) (McLelland et al., 2018). These results collectively demonstrate that RFFL forms a complex with MFN2 and contributes to MFN2 ubiquitylation.

Given that endogenous RFFL interacts with endogenous MFN2 (Fig. 2I) and the presence of MFN2 on RFFL-positive vesicles (Fig. 2D), we investigated the plausibility of a direct interaction between recombinant RFFL and MFN2. We co-expressed full-length RFFL with a GST tag and MFN2 with either a His or His–Sumo tag in Escherichia coli cells and used the extracts from these cells for pulldown using glutathione–Sepharose beads. Pulldown of GST–RFFL revealed the presence of His–MFN2, whereas no MFN2 was observed with pulldown of either GST alone or with beads (Fig. 3A). These results confirm that RFFL interacts directly with full-length MFN2. To identify regions in MFN2 interacting with RFFL, we co-expressed GST-tagged RFFL with truncated MFN2 (amino acids 1–357) with either His or His–Sumo tags (Fig. 3B). The binding of RFFL was still observed when the MFN2 (1–357) variant was used, suggesting that RFFL recognizes the cytosolic N-terminal of MFN2, which contains the GTPase domain. Whether His or His–Sumo was used as the tag made little difference between the interaction of tagged MFN2 (1–357) and GST–RFFL (Fig. 3C), with GST alone showing no interaction. Given that GST–RFFL interacts with the MFN2 (1–357) variant and there was not much difference in this interaction with His alone or His–Sumo tag, we used His–Sumo tagged MFN2 (1–357) for most of the other experiments, considering its better solubility. Next, we assessed the interaction of MFN2 with different variants of RFFL. The ubiquitin ligase activity of the RING domain-containing E3s is mediated by the coordination of a zinc ion by conserved cysteine and histidine residues. Substitution of any of these amino acids, for example, H333A in the case of RFFL, results in loss of the E3 activity (Araki et al., 2003; Freemont, 2000; Liao et al., 2008; Okiyoneda et al., 2018). Another variant, where the amino acids cysteine at 5 and 6 of RFFL (presumably involved in palmitoylation) are replaced by alanine, changes RFFL localization from intracellular vesicles to cytosolic (Araki et al., 2003; Okiyoneda et al., 2018; Ravindran et al., 2022). We found that both of these variants of RFFL interacted well with MFN2 (1–357) (Fig. 3D), suggesting that these mutations have no effect on MFN2 interaction in vitro. Next, we examined various constructs of RFFL with individual domains deleted. The deletion variants tested showed variations in the extent of binding, suggesting that MFN2 associates with RFFL through more than one domain (Fig. 3E,F), apparently with different strengths.

Given that recombinant RFFL protein was found to directly interact with MFN2, we tested whether this interaction could lead to ubiquitylation of MFN2 in vitro. For this, we conducted a ubiquitylation assay using recombinant RFFL and MFN2. We co-purified MFN2 and RFFL and set up an in vitro ubiquitylation reaction on beads. As E2 enzymes also play a crucial role in substrate ubiquitylation along with E3s, we first evaluated the MFN2 ubiquitylation in the presence of different E2 enzymes and RFFL as E3 ligase. The results showed that only in the presence of the E2s UBE2W, UBE2D1 and UBE2D2 (also known as UBC16, UBCH5A and UBCH5B, respectively) ubiquitin modification on MFN2 by RFFL was observed (Fig. 4A). Whereas UBE2D1 and UBE2D2 promoted autoubiquitylation of RFFL as expected (Liao et al., 2008), the reaction with UBE2W did not result in RFFL modification. As expected, all these three E2s promoted MFN2 ubiquitylation in the presence of ATP only. None of the other eight E2s showed any appreciable modification of either RFFL or MFN2 (Fig. 4A). To our knowledge, this is the first report of MFN2 ubiquitylation by any of the E3s in vitro. We also found that full-length MFN2 is also ubiquitylated in vitro by RFFL, albeit to a lesser degree compared to the MFN2 (1–357) variant (Fig. 4B). To further validate the ubiquitylation of MFN2, we repeated the experiments with His–Sumo constructs, the results of which clearly demonstrated modification and a high molecular mass smear representing polyubiquitylation (Fig. 4C; Fig. S3A). Moreover, no such modification or smear was noted when E3 inactive RFFL was used (Fig. 4D), suggesting that the modification observed in the in vitro reaction was indeed a result of the ubiquitin ligase activity of RFFL.

Given that the ubiquitylation data from the in vitro assay showed the presence of multiple high molecular mass bands with MFN2-specific antibodies, we evaluated whether these modifications involve polyubiquitylation. To assess this, we used the ubiquitin variants Me-Ub (free amino groups of lysine residues are chemically modified by reductive methylation and hence cannot participate in further ubiquitin chain formation) (Kirisako et al., 2006) or Ub-K0 (all lysine mutated to arginine residues) along with GST–Ub (N-terminus of Ub is fused with GST tag) and untagged Ub (Ub–WT). Consistent with the idea that RFFL promotes polyubiquitylation, Ub-K0, and Me-Ub showed very few multiple bands of MFN2 compared to untagged WT Ub or GST–Ub in the presence of ATP (Fig. 4E, marked by arrows). The data collectively suggest RFFL targets MFN2 for polyubiquitylation, which might lead to its proteasomal or lysosomal degradation in cells.

Given that we found direct interactions between RFFL and MFN2, which leads to the ubiquitylation of MFN2 in vitro, we looked at whether this ubiquitylation leads to the degradation of MFN2. We looked at the endogenous MFN2 in the lysate of A549 cells stably expressing EV or RFFL variants treated with DMSO or CCCP. A reduction in endogenous MFN2 was noted in cells stably expressing RFFL WT compared to EV upon CCCP treatment (Fig. 5A,B). However, this reduction was not observed in the presence of ligase inactive RFFL (H333A) (Fig. 5A,B). Given that endocytic vesicle-associated RFFL appeared to target MFN2 (Fig. 2D; Movies 5, 6), we wondered whether the expression of the endosomal-association-defective RFFL variant (RFFL C5, 6A) could also target MFN2 for degradation (Ravindran et al., 2022). When the RFFL C5, 6A variant was expressed along with MFN2–Myc, unlike WT, variant (C5, 6A) showed no effect on MFN2 protein level (Fig. 5C). As the C5, 6A variant of RFFL could still interact with MFN2 (Fig. 3D), the data suggests that the endosome association of RFFL is important for the degradation of MFN2 by RFFL, further highlighting the importance of endosome–mitochondria contacts. As expected, the reduction in MFN2 levels was not observed with the inactive RFFL E3 mutant (H333A), further indicating the involvement of the ubiquitin ligase activity of RFFL in the reduction of MFN2 protein (Fig. 5C). Given that we observed a robust reduction in MFN2 levels by RFFL under overexpression conditions, we next investigated whether RFFL targets other proteins involved in mitochondrial fusion and fission. Specifically, we examined MFN1, a key mitochondrial fusion protein closely related to MFN2 and ranked second in our MLM score-based ranking (Fig. 2A), as well as DRP1 (also known as DNM1L), a mitochondrial fission factor, to assess the effect of RFFL (Dorn, 2019). For this, we used MFN2 as positive control and co-transfected EGFP–MFN1 or mCherry–DRP1 along with RFFL in HEK293T cells. Not surprisingly, RFFL also targeted MFN1 for degradation to a similar extent as MFN2 (Fig. 5D,E). However, RFFL expression was unable to reduce levels of DRP1 protein, suggesting the specificity of RFFL towards MFN1 and MFN2 (Fig. 5F,G).

We then co-transfected HEK293T cells with MFN2–Myc along with EV or untagged RFFL WT and treated the cells with proteasome inhibitor MG132 (Tsubuki et al., 1996) to assess whether the reduction in MFN2 was the result of proteasomal degradation. Upon MG132 treatment, the protein levels of MFN2–Myc moderately increased not only in EV control but also in cells expressing RFFL, suggesting the contribution of RFFL to MFN2 degradation by proteasomal machinery (Fig. 5H,I). Because RFFL appears to target MFN2, and lack of RFFL results in hyperfused mitochondria, we wondered about the effect of RFFL on MFN2 variants that are known to cause hyperfusion of mitochondria. Two MFN2 mutations have been reported to be associated with CMT2A, namely, MFN2 R364W and T206I, and are known to cause mitochondrial hyperfusion (Das et al., 2022a,b; Das et al., 2024; El Fissi et al., 2018; Stuppia et al., 2015; Verhoeven et al., 2006). Interestingly, the pathological and hyperfusion mutant variants of MFN2 showed increased susceptibility to degradation by RFFL than WT (Fig. 5J,K). Given that these MFN2 mutants are reported to cause hyperfusion of mitochondria, we wondered whether RFFL could rescue the effect of the CMT2A-associated mutants. We overexpressed these mutants in HeLa cells and observed mitochondrial morphology using confocal microscopy. As reported, we saw hyperfused mitochondria in the presence of T206I and R364W mutants of MFN2. Importantly, co-transfection with RFFL decreased this hyperfusion, bringing back the mitochondria morphology to normal (EV) levels (Fig. 5L,M; Fig. S3B).

Given that endosome–mitochondria association facilitates lipid exchange (Ouyang et al., 2023), we investigated the effect of RFFL expression on the accumulation of lipid droplets (LDs), considered as energy storage units in cells (Enkler and Spang, 2024; Kumar et al., 2024; Mallick et al., 2024). For this, we stained A549 cells stably expressing RFFL–EGFP or EGFP alone with HCS LipidTOX Deep Red, a dye that stains neutral lipids. Confocal images of these cells showed differences in lipid staining between the cells with or without RFFL overexpression (Fig. 6A). Quantification of LDs showed that there were more LDs in RFFL-expressing cells with a substantial increase in LDs area than in control cells (Fig. 6B). Interestingly, the colocalization between LDs and mitochondria (stained with MitoTracker Red CMXRos), measured by different parameters, were also found to be less in RFFL-expressing cells (Fig. 6C). These results are consistent with observations reported earlier in cells with reduced MFN2 (Cai et al., 2024; Hu et al., 2024). These results indicate a hitherto unknown role for RFFL in lipid homeostasis.

Cellular homeostasis is maintained by mitochondrial dynamics established through numerous contacts with diverse organelles like the ER, endosomes and lysosomes, in addition to mitochondria themselves. These continuous crosstalks affect mitochondrial morphology and diverse signaling like lipid homeostasis, synaptic plasticity, cell death and autophagy by aiding the exchange of mtDNA, ions, lipids and metabolites (Collier et al., 2023; Harper et al., 2020). Recently, research on mitochondria–endosome crosstalk (MECS) has been gaining momentum (Müntjes et al., 2021; Todkar et al., 2019). MECS has been reported to aid iron and cholesterol transfer between endosomes and mitochondria (Charman et al., 2010; Das et al., 2016; Hamdi et al., 2016; Nara et al., 2023; Satoh et al., 2023; Sheftel et al., 2007; Wang et al., 2020). Here, we show that RFFL endosomes come into close contact with mitochondria and alter mitochondrial morphology by specifically targeting outer mitochondrial membrane (OMM) proteins involved in mitochondrial fusion (MFN1 and MFN2) but not fission-associated protein DRP1. Moreover, we demonstrate that exogenous expression of RFFL results in an increased lipid content.

Lipid droplets (LDs) are primarily made up of neutral lipids, such as triglycerides and cholesterol. They form highly dynamic connections with various organelles, including the ER, endosomes and mitochondria (Enkler and Spang, 2024; Peng et al., 2025). Although many aspects of lipid–organelle crosstalk are not yet understood, LDs are at the center of lipid and energy homeostasis in cells, and it is increasingly believed that their dynamic association with organelle are coupled to the cycles of lipid droplet expansion and shrinkage (Enkler and Spang, 2024; Zadoorian et al., 2023). Our findings that RFFL expression leads to not only a reduction in colocalization of LDs with mitochondria but an increase in LD numbers and area suggest that these changes could be the consequence of RFFL-mediated changes in MFN2 protein levels. MFN2 is reported to connect mitochondria to LDs via its association with perilipin 1 (PLIN1), a member of the perilipin family on LDs, and suppress lipid accumulation in cells (Boutant et al., 2017; Cai et al., 2024; Mahdaviani et al., 2017). Interestingly, PRKN is also known to be involved in lipid remodeling during mitophagy (Lin et al., 2022; Tang et al., 2023). Combined with our earlier observation of delayed PRKN recruitment to damaged mitochondria in RFFL KO cells, this study establishes a novel role for RFFL and endosomes in mitochondrial dynamics, mitophagy and lipid homeostasis.

MFN2 is known to be ubiquitylated in cells upon mitochondrial depolarization (Gegg et al., 2010; McLelland et al., 2018). Although several E3 ligases, namely MARCHF5 and MUL1 (mitochondria), PRKN, RBCK1 and HUWE1 (cytosolic), and AMFR (ER) are reported to contribute to MFN2 ubiquitylation, none of these E3 ligases has been shown to directly modify MFN2 nor associate with endosomes. In this study, we demonstrate that RFFL interacts and ubiquitylates MFN2 in vivo and in vitro, and loss of RFFL results in reduced MFN2 burst upon CCCP treatment. These results clearly indicate that RFFL is an E3 for MFN2. Furthermore, for the first time, we demonstrate the ubiquitylation of MFN2 by any E3 using recombinant proteins. Interestingly, unlike UBE2D1 and UBE2D2, E2s that facilitate ubiquitylation of both MFN2 and autoubiquitylation of RFFL, UBE2W modifies MFN2 but not RFFL, suggesting differential RFFL activity with different E2s. These differences can be of significance in the physiological context, given the reported diverse tissue-specific roles for MFN2 (Chandhok et al., 2018). MFN2 splice variants dubbed ER mitofusin 2 (ERMIN2) and ER mitofusin 2 tether (ERMIT2) have been reported. The former is involved in shaping the ER, and the latter in ER–mitochondria tethering (Naón et al., 2023). It will be interesting to explore the effect of RFFL on these isoforms. Given that no endosomal-associated E3 ligase is known to control MFN2 turnover and mitochondrial morphology so far, identifying RFFL as an E3 ligase for MFN2 opens a new avenue to explore endosome mitochondria crosstalk. Consistent with such a possibility, a recent unbiased analysis of the endogenous interactome of MFNs has identified a number of endosomal-associated proteins (Gordaliza-Alaguero et al., 2025).

Given that our data demonstrate that RFFL binds and ubiquitylates the cytosol facing N-terminus of MFN2 (1-357 amino acids, containing GTPase domain) like the full-length MFN2, it raises the possibility that under physiological conditions, RFFL can influence the GTPase activity and subsequent MFN2 dimerization and membrane fusion. It is interesting to note that the majority of mutations of MFN2 reported in individuals with CMT2A are concentrated in the N terminus of MFN2 (GTPase domain) (Cartoni and Martinou, 2009; Filadi et al., 2018; Züchner et al., 2004). Exogenous expression of the CMT2A pathogenic variants of MFN2 (T206I and R364W) are known to promote hyperfusion of mitochondria, reportedly by stabilizing the mutated protein level (Das et al., 2022b, 2024). Importantly, we show that co-expression with RFFL not only reduced the protein levels of both the pathogenic variants but also rescued the hyperfusion phenotype of mitochondria. Restoring impaired mitochondrial dynamics in CMT2A disease is an active avenue of research (Franco et al., 2016; Rocha et al., 2018). Hence, the observation that there is a reduction in the levels of pathogenic MFN2 variants mediated by RFFL can open up a potential therapeutic option by modulating RFFL activity. Interestingly, mutations in not only MFN2 but also endosomal Rab7 proteins have been implicated in the development of Charcot–Marie–Tooth neuropathy diseases (Meggouh et al., 2006; Ponomareva et al., 2016). It is also possible that the RFFL–MFN2 crosstalk affects signaling pathways other than mitochondrial dynamics. For example, RFFL positively regulates cell migration and proliferation (Dong et al., 2013; Gan et al., 2012, 2013; Sharma et al., 2024), with MFN2 controlling negatively (Chen et al., 2004, 2014; Guo et al., 2007; Shen et al., 2007), making it another avenue for exploration.

Overall, the findings of this study provide novel insights into the intricate interplay between RFFL, mitochondrial dynamics and the regulation of MFN2 turnover. By elucidating the role of RFFL in modulating mitochondrial morphology and its direct interaction with MFN2, this research lays the groundwork for further exploration of the underlying molecular mechanisms and the potential implications for cellular homeostasis and disease pathogenesis. These findings have the potential to pave the way for the development of novel therapeutic strategies targeting RFFL-mediated regulation of mitochondrial dynamics and lipid homeostasis in various pathological conditions associated with mitochondrial dysfunction.

A549 (ATCC, #CRM-CCL-185), HeLa (ATCC, #CRM-CCL-2) and HEK293T cells (ATCC, #CRL-3216) were grown in Dulbecco's modified Eagle's medium (Thermo Fisher Scientific, 10569-010) supplemented with 10% heat-inactivated fetal bovine serum (Thermo Fisher Scientific, 10270-106) and 100 U/ml penicillin as well as 100 µg/ml streptomycin (Thermo Fisher Scientific, 15140122) at 37°C in a humidified chamber with 5% CO₂ as described previously (Ravindran et al., 2022).

Transfection of cells with plasmid DNA was undertaken using Lipofectamine 3000 reagent (Thermo Fisher Scientific, L3000-015) according to the manufacturer's instructions. The total amount of DNA in all wells was kept constant using empty vector DNA in all experiments. Cells stably expressing different constructs were generated as described previously (Ravindran et al., 2022). A549 RFFL KO cells were generated as described previously (Ravindran et al., 2022).

Human CARP2 cDNA (NP_001017368.1) and its variants were cloned to pMYs-IP (Kitamura et al., 2003) for stable cell generation (Ravindran et al., 2022). RFFL variants were cloned into pGEX-4T-2 (GE Healthcare Life Sciences, #27458101) for bacterial expression and purification and pcDNA 3.1 vector (Invitrogen, #V790-20) for transfection in mammalian cells. MFN2–YFP was a generous gift from Atsushi Tanaka (Biochemistry Section, National Institutes of Health, USA; Tanaka et al., 2010). MFN2 cDNA (NP_001121132.1) was cloned into piRFP670-N1 for imaging studies. piRFP670-N1 was Addgene plasmid #45457, deposited by Vladislav Verkhusha (Shcherbakova and Verkhusha, 2013). WT MFN2–Myc, T206I MFN2–Myc and R364W MFN2–Myc constructs were kind gifts from Prof. Oishee Chakrabarti (Biophysics and Structural Genomics Division, Saha Institute of Nuclear Physics, India; Das et al., 2022a,b, 2024). pERS28a was a kind gift from Dr Ramanathan Natesh (School of Biology, Indian Institute of Science Education and Research Thiruvananthapuram, India; Maddi and Natesh, 2021). MFN2 was cloned to the pERS28a or pET-28a vector (EMD Biosciences, #69864-3) for expression and purification from bacteria. YFP–Parkin was Addgene plasmid #23955, deposited by Richard Youle (Narendra et al., 2008). pRK5-HA-Ubiquitin-WT was Addgene plasmid #17608, deposited by Ted Dawson (Lim et al., 2005). pRK5-HA-Ub-K0 was a kind gift from Prof. Subramaniam Ganesh (Biological Sciences & Bioengineering, Indian Institute of Technology Kanpur, India). mCherry–DRP1 was Addgene plasmid #49152, deposited by Gia Voeltz (Friedman et al., 2011). EGFP-MFN1 was generated by cloning MFN1 cDNA into pEGFP-C1 (Mukherjee and Chakrabarti, 2016). Ub cDNA (untagged or fused with GST) was subcloned to pET-3A (Novagen, #69418) for protein purification. All plasmids contain cDNA of human origin.

Antibodies against the following proteins were used in this study; TOMM20 (1:1000, Santa Cruz Biotechnology, sc-17764), MFN2 (1:2000, Proteintech, 12186-1-AP), HA (1:2000, Santa Cruz Biotechnology, sc-7392), ubiquitin (1:2000, Santa Cruz Biotechnology, sc-8017), GST (1:2000, Sigma-Aldrich, G7781), Myc (1:500, DSHB, 9E 10), RFFL (Ravindran et al., 2022), GAPDH (1:2000, Abgenex, 10-10011), tubulin (1:5000, Sigma-Aldrich, T6199), UQCRC1 (1:2000, Invitrogen, 16D10AD9AH5), GFP (1:1000, Mouse Living Colors Clontech, 632375 or SCBT, sc-9996), RFP (1:2000, ChromoTek, 5f8-100) and parkin (1:2000, Santa Cruz Biotechnology, sc-32282). Anti-mouse-IgG HRP (1:5000, Invitrogen, 61-6520/Jackson ImmunoResearch, 115-035-174) or anti-rabbit-IgG HRP (1:5000, Invitrogen, 65-6120/Jackson ImmunoResearch, 211-032-171) were used to visualize the immunoblots with chemiluminescence.

For mitochondrial staining, cells were incubated with 10 nM MitoTracker Red CMXRos (M7512, Gibco-Thermo Fisher Scientific) for 30 min in medium. For lipid staining, HCS LipidTOX™ Deep Red Neutral Lipid Stain (H34477, Invitrogen, Carlsbad, CA, USA) was added at 1:5000 dilution for 25 min after MitoTracker Red CMXRos staining. DAPI (D8417, Sigma-Aldrich) staining was performed as described previously (Ravindran et al., 2022). The following chemicals were used: MG132 (M7449, SigmaAldrich), Bafilomycin A1 (19-148, Merck, Darmstadt, Germany), and CCCP (C-2759, Sigma-Aldrich).

Immunoprecipitation experiments (both native and denaturation conditions) were performed as described previously (Ravindran et al., 2022). Western blotting experiments were done using standard protocols. To assess MFN2–Myc degradation, HEK293T cells were co-transfected with 400 ng of untagged RFFL and 1000 ng MFN2–Myc constructs in a 35 mm dish. After 36–48 h of transfection, cells were harvested (500 g for 2 min) and suspended in SNTBS buffer (50 mM Tris-HCl pH 7, 150 mM NaCl, 2% SDS, 1% NP-40) with 1 mM phenylmethylsulfonyl fluoride (Sigma-Aldrich, P7626) and 1× protease inhibitor cocktail (Sigma-Aldrich, P8340). The extracts were sonicated (Sha et al., 2019) and incubated at 37°C for 30 min before loading on SDS-PAGE gels. Unless otherwise specified, for MFN2 detection, samples were heated at 37°C for 30 min with Laemmli buffer (Schägger, 2006; Yamano et al., 2020). Quantification of bands was done from western blot images taken without a saturating signal.

Cells were seeded in a T75 flask for 24 h and collected at ∼90% confluency. Cells were washed with 100 mM phosphate buffer and fixed with fixative (2.5% glutaraldehyde plus 2.0% paraformaldehyde in 0.1 M phosphate buffer) for 2 h at room temperature. Cells were scrapped after fixation and washed three times with phosphate buffer for 15 min each and post-fixed with 1% osmium tetroxide for 1 h at 4°C. After fixation, cells were dehydrated in a graded series of ethanol (30%, 50%, 60%, 70%, 80%, 90% and 100%). These samples were infiltrated with toluene/resin and embedded in Epon 812 resin. The resin-embedded samples were polymerized at 65°C for 48 h. The ultrathin sections of 70 nm were cut using Leica UC7 ultramicrotome and mounted on the carbon-coated 200 mesh copper grid. The sections were heavy metal stained using uranyl acetate and lead citrate, then viewed under a transmission electron microscope Talos F200S (Thermo Fisher Scientific).

Individual mitochondria in each cell were marked as regions of interest (ROI) for quantification, and images were calibrated. Then, the area and perimeter were measured using ImageJ software (ImageJ 1.53t) (Schneider et al., 2012). Feret's diameter is defined as the longest distance between any two points along the selection boundary and was also measured using ImageJ. Mitochondria marking and measurement were done by a person who was unaware of the experimental conditions.

For fixed-cell imaging, cells were grown on coverslips and immunostained as described previously (Ravindran et al., 2022). Images were acquired in Zeiss LSM 880 confocal microscope (Carl Zeiss, Germany) with either a Plan-Apochromatic 63×/1.4 or a 100×/1.4 oil immersion objective. For live-cell imaging, cells were cultured in a glass bottom dish (Cell Vis, D35-20-1.5-N) and imaged using the FV3000 confocal microscope (Olympus, Japan) with an UPLXAPO 60×/1.42 oil immersion objective lens. The FV3000 system had an attached STX stage top incubator (TOKAI HIT, Japan), humidified with 5% CO₂ at 37°C. 2D deconvolution was performed on all time-lapse movies for better visibility using Olympus cellSens Dimension Desktop 3.2. Image contrast was adjusted using the lookup table for better visibility of representative Figures. For colocalization analysis, BIO-jacop plugin was used after auto thresholding (https://github.com/BIOP/ijp-jacop-b; Bolte and Cordelières, 2006). Lipid-mitochondria analysis was performed using the analyze particle function and colocalization plugin after auto thresholding (https://imagej.net/ij/plugins/colocalization.html). All quantifications were performed from raw confocal images unless otherwise indicated.

A549 cells were fixed and immunostained against endogenous TOMM20 as described above. Fixed cells were imaged using Zeiss LSM 880 with Plan-Apochromat 100×/1.40 oil objective. The pinhole was set to 0.75 airy unit (AU). Around 20–35 Z-sections were taken for each image with a 250 nm step size. Bidirectional scanning was enabled for fast capture. These images were deconvoluted using the Fiji (ImageJ 1.53t) plugin theoretical PSF generator (Kirshner et al., 2013) and deconvolution lab2 (Version 2.1.2). PSF was calculated using the Born and Wolf 3D Optical Model with the best accuracy. Deconvolution was performed using the Richardson–Lucy algorithm with ten iterations. Deconvoluted images were saved, calibrated and converted into 8-bit images. These images were thresholded first using mitochondria analyzer plugin (Chaudhry et al., 2020) with pre-processing commands subtract background (rolling 1.25), sigma filter plus (radius 0.6), enhance local contrast (max slope 1.4), adjust gamma (0.9), local threshold (weighted mean), C value 3.5, block size 1.25 µm and post-processing commands, Despeckle, Remove outliers (radius 0.6 pixels) and Fill 3D holes commands were enabled. Analysis was performed on these thresholded images after drawing a region of interest on single cells and clearing out the rest of the field. The 3D view was generated using a 3D viewer plugin available in Fiji (Schindelin et al., 2012).

For calculating the average area of mitochondria per cell, live imaging was performed and 2D images of mitochondria stained with MitoTracker Red CMXRos were captured. The pinhole was set to 0.75 airy units (AU). These images were deconvoluted using Olympus cellSens Dimension Desktop 3.2. The analyze particle function in Fiji software was used to remove pixels less than 0.4 µm. Then, the mitochondria analyzer plugin (Chaudhry et al., 2020) was used as described above to segment mitochondria, followed by area measurement.

Manual measurement of mitochondrial length and width was measured as described previously (Lam et al., 2021). Briefly, for mitochondrial length, the freehand line tool in ImageJ was used. Ten random positions per cell where mitochondria were not clustered were measured using this tool. Similarly, the width of the mitochondria was measured using the straight line tool. The percentage of cells with fragmented, normal and interconnected mitochondria was calculated based on the length. Measurements were done by a person who was unaware of the experimental conditions. Mitochondria with a length of <2 μm considered as fragmented, 2–5 μm as normal (tubular), and >5 μm as interconnected (filamentous or hyperfused) (Das et al., 2022a).

For in vitro interaction experiments, His-tagged MFN2 constructs and GST-tagged RFFL were co-transformed in Arctic Express E. coli cells, and transformants with both constructs were selected on agar plates with antibiotics: 10 μg/ml gentamicin (Sigma-Aldrich, #G1272), 20 μg/ml kanamycin (Himedia, CMS7172) and 100 μg/ml ampicillin (Himedia, #MB104). The bacterial culture was grown from a single colony at 37°C with rotation at 220 r.p.m. until the optical density at 600 nm (OD600) reached 0.4. Then 150 µM of IPTG was added, and the culture was shifted to 10°C for 24 h at 220 r.p.m. The culture was pelleted (3000 g for 10 min), and the cell pellet was lysed in buffer A (50 mM NaH₂PO₄ pH 8.0, 15 mM 2-mercaptoethanol, 500 mM NaCl, 25 mM trehalose, 5 mM MgCl₂, 0.025 mM MnCl₂, 5% glycerol, 0.5% Triton-X, 0.714 mg/ml lysozyme, 2 mM PMSF and 1× protease inhibitor (Sigma-Aldrich, cOmplete Mini Protease Inhibitor Cocktail). Cells were then lysed by sonication and centrifuged at 18,000 g for 10 mins at 4°C. The supernatant was collected and incubated with pre-washed Glutathione beads (GE Healthcare, Glutathione Sepharose 4 Fast Flow) for 4 h at 4°C. Beads were then washed three times with buffer B (50 mM NaH₂PO₄ pH 8.0, 15 mM 2-mercaptoethanol, 500 mM NaCl, 5 mM MgCl₂ and 5% glycerol) and then resuspended in 2X Laemmli buffer and heated for 30 min at 37°C.

For in vitro ubiquitylation experiments, RFFL and MFN2 were co-expressed, and GST pulldown was performed as described above. Then GST-bound Glutathione beads were washed in Buffer C (50 mM Tris-HCl pH 7.4 and 5 mM MgCl₂) and re-suspended in 30 µl Buffer C with an E1 (10 nM UBA1; R&D Systems, E-304) and E2 (0.2 µM; see below) enzyme, Ub (0.2 ug), 2 mM DTT and 2 mM ATP. The samples were incubated at 30°C for 90 min with rotation at 600 r.p.m., and the reaction was stopped by adding the final 2× Laemmli buffer and heated for 30 min at 37°C. The following E2 enzymes were used in this study: UBE2W (E2-725) UBE2H (K-980B) UBE2R1 (K-980B) UBE2D1 (K-980B) UBE2D2 (K-980B) UBE2E1 (K-980B) UBE2E3 (K-980B) UBE2L3 (K-980B) UBE2C (K-980B) UBE2N/Uev1a (E2-664) UBE2G2 (E2-680), from R&D Systems, USA. If the E2 is not specified, UBE2D1 was used. Me-Ub was from R&D Systems, USA (U-501).

Recombinant ubiquitin was produced through overexpression in E. coli BL21 (DE3) competent cells. A singular colony was introduced into LB broth supplemented with 100 µg/ml ampicillin and cultured at 37°C for 12 h. The resulting primary culture was transferred to 1 liter of LB broth containing 100 µg/ml ampicillin and maintained at 37°C until the OD600 reached was 0.4–0.6. Subsequently, cells were induced with 1 mM IPTG and further incubated at 37°C for 5 h. Upon completion of the induction period, the culture was centrifuged at 9820 g for 10 min at 4°C and stored at −80°C. The thawed cell pellet was suspended in a 25 ml lysis buffer (50 mM sodium acetate and 5 mM EDTA, pH 5.1, 2.87 mM PMSF, and one cOmplete, mini, EDTA free protease inhibitor cocktail tablet; Roche) and subjected to sonication for 3 min (pulse: 20 s on, 30 s off; amplitude, 55%) on ice. The lysed cells were then centrifuged at 101,633 g for 90 min at 4°C. The resulting supernatant was applied to an SP Sepharose column, pre-equilibrated with sodium acetate buffer (matching the lysis buffer but without PMSF and protease inhibitor), and left overnight in a cold room. After washing the beads with sodium acetate buffer, protein elution was achieved using a sodium chloride gradient ranging from 10 to 250 mM in sodium acetate buffer. The eluted fractions, assessed by 15% SDS-PAGE and Coomassie staining, were pooled based on comparable purity. Subsequent dialysis (50 mM Tris-HCl pH 7.4, 5 mM MgCl₂ and 2 mM DTT in deionized water) was conducted for 8 h at 4°C, with buffer changes occurring at four-hour intervals. The dialyzed protein was concentrated further using Amicon Ultra-4 centrifugal filters with a 3 K regenerated cellulose membrane. Protein concentration was determined, and aliquots were preserved at −80°C.

To predict how likely a pair of proteins bind to each other, we used protein language model ESM-2 (Lin et al., 2023). The list of prey proteins was concatenated in pairs with bait protein, and Masked Language Model loss (MLM loss) was calculated using the masked language modeling capabilities of ESM-2 and randomly mask residues (Lupo et al., 2024). We used esm2_t30_150M_UR50D with 150 million parameters for this study.

Data were collected from at least three independent experiments. Significance levels were set at 0.05, and are marked as *P<0.05, **P<0.01, and ***P<0.001. All statistical analyses were performed in GraphPad Prism (Version 8.3.1). All graphs were generated in GraphPad Prism. Figures were arranged using Inkscape 1.2 (https://inkscape.org/). For protein domain illustrations, we used DOG 2.0 and modified (Ren et al., 2009).

The authors would like to acknowledge Prof. M. K. Mathew, Prof Dipshikha Chakravortty, Prof. Maneesha S. Inamdar, Prof. Oishee Chakrabarti, and Dr Parameswaran Ramakrishnan for their helpful discussions. We thank Athul Krishnan R. and Minnu Manoj for their technical help. The authors also thank Prof. Oishee Chakrabarti, Prof Ravi Manjithaya, and Prof. Chandan Goswami for sharing plasmids.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.263830.reviewer-comments.pdf