The coronavirus disease-19 (COVID-19) pandemic is caused by the positive-sense single-stranded RNA coronavirus, SARS-CoV-2, that has infected, in a year, more than 200 million people and has killed almost 4.5 million people worldwide ¹ since it has no definitive and effective treatment until today. This infection affects mainly certain groups of people that have high susceptibility to present severe COVID-19 due to comorbidities that include cardiovascular disease, chronic kidney, respiratory or liver disease, severe obesity, or hypertension among others. In these patients, the cytokine storm induced by the virus poses a serious death risk for these patients due to the systemic inflammation and multiorgan failure ^{2– 6} . Moreover, the so-called long COVID-19 comprises a series of symptoms that may remain in some patients for months after infection that further compromises their health, even after non-severe COVID-19 ⁷ . Despite huge efforts to stop infections and deaths worldwide, only a few treatments have been demonstrated to ameliorate severe COVID-19, and different vaccine strategies are currently under investigation in clinical phase III and/or IV trials. Therefore, in this scenario, a deeper understanding of the cellular mechanisms exploited by SARS-CoV-2 for cell infection could undoubtedly provide new unforeseen strategies to tackle this pandemic.

Several recent reports have shown that mitochondria play a relevant role during SARS-CoV-2 infection ^{8– 10} . These findings and previously published results of SARS-CoV-2, SARS-CoV and other coronavirus biology allow us to hypothesize that mitochondria could be responsible for the assembly of double-membrane vesicles (DMV). These are membrane modifications induced by SARS-CoV-2 and it is where viralRNA (vRNA) replication occurs in the infected cell, that are believed to be derived from the endoplasmic reticulum (ER) or other mechanisms, such as autophagy ^{11– 13} . However, some published literature supports that double-membrane mitochondria-derived vesicles (MDV), discovered some years ago ¹⁴ , could be the precursors or relatives of DMV. This hypothesis of mitochondria role in DMV assembly and the involvement of MDV has been suggested previously ^{8, 15, 16} . Indeed, specialized replication organelles (RO) at mitochondria outer membrane (MOM) have been observed in FHV ¹⁷ , whereas HIV RNA is known to locate in the mitochondria ¹⁸ . Here, a brief review of the evidence that supports this notion is presented and integrated into the current model ¹⁹ , with the intention to provide a novel framework that could open possibilities to tackle the SARS-CoV-2 pandemic.

DMV along with other membrane modifications are part of the RO induced by SARS-CoV-2 that also includes convoluted membranes (CM), zippered ER (zER), vesicle packets (VP), and double-membrane spherules (DMS) ¹⁹ ( Figure 1). RO are induced by SARS-CoV-2 in infected cells, as a variety of RO are induced by viruses including other nidovirus and picornavirus among others ^{11, 13} . DMV assembly is induced by viral proteins but seems to also require other viral or host factors because cell plasmid transfection of transmembrane-containing non-structural proteins (nsp) 3-, 4-, and 6-induced membrane arrangements that resemble DMV but with smaller size ²⁰ . These nsp are part of the complex involved in vRNA replication, together with nsp12, the RNA-dependent RNA polymerase, and other nsp. Consistently, nsp4 mutation alters the assembly of DMVs ¹⁵ and abolishes viral replication ²¹ . Nsp4, 3, and the nuclear (N) protein are located at the DMV ^{16, 19} , where nsp3 has been shown to form a pore complex that communicates DMV interior with the cytoplasm, that was elusive for some time, pore that could also involve host factors and/or other viral proteins ²² .

Most SARS-CoV-2-induced membrane modifications are derived from ER membranes and are interconnected. In the case of DMV, this origin was in part assumed because such mechanism is presumed to mediate RO assembly in other positive sense RNA viruses ²³ ; because they have contacts with other membrane modifications of the RO and the ER, because ribosomes have been observed on DMV surface, and because DMS and CM were proposed to be precursors of DMV ^{11, 16, 19, 20, 23} . DMV sizes range from 150 to 300 nm, but they grow through infection, and it has been established that SARS-CoV-2 DMV are the location where vRNA synthesis occurs ^{16, 19, 24} . Importantly, DMV are early (1–2 h post-infection [p.i.]) observed in the cell cytoplasm after coronavirus infection in vitro, and their number increase through time reaching a maximum in 6–8 h p.i. ^{16, 25} . There are currently two models for DMV assembly from the ER. In the case of coronavirus, DMV are thought to be assembled from zippered ER that folds and closes in response to vRNA, as observed with IBV ^{11, 23} , with CM and DMS as putative intermediate precursors ^{16, 19} . However, major challenges remain for this model, because no intermediate structures have been recognized between DMV and DMS or CM, and because CM and DMS, both derived from the ER, have no relationship with DMV beyond their membrane contacts ^{16, 19} . DMV do not have ER (nor ERGIC, Golgi, or endosomal/lysosomal) markers as it would be expected if they were assembled from the ER, and most importantly, it has been shown that DMV assembly (1–2 h p.i.) precedes CM assembly (~3 h p.i.) ²⁵ . Furthermore, CM and DMS do not carry out vRNA synthesis as DMV do ¹⁹ . Thus, CM and DMS are not DMV precursors ¹⁹ . On the other hand, the high energetic and complex topological requirements assumed to occur for DMV assembly through the zippered ER−CM−DMS model, given their numbers after a few hours post-infection, further complicate this notion. Indeed, it has been suggested that DMV could have another origin than the ER ^{8, 11, 19} .

Given the recent findings that relate SARS-CoV-2 with mitochondria, it is possible that MDVs are the origin of DMV. MDV were discovered more than a decade ago and they comprise different vesicles shed from mitochondria involved in its quality control ^{14, 26} . Interestingly, some MDV have double-membrane with a size of 60–150 nm and are shed independently of drp1, mitochondria fission, and autophagy ( Figure 2A) ^{26, 27} . Interestingly, several coronavirus proteins have been shown to down- or up-regulate drp1 (N/envelope [E], nsp3, nsp4a, nsp4b, and ofr9b) ¹⁰ . MDV are generated in steady conditions and have been observed in vivo ²⁸ , but their number increases after mitochondrial stress or higher respiratory activity ^{26, 27} . In this regard, it has been found that SARS-CoV-2-infected monocytes have compromised mitochondrial function and energy deficit ^{30, 31} . This could be the long-term result of viral infection in which mitochondria shedding of MDV/DMV, triggered initially by a burst of metabolic activity or stress, leads to mitochondria function impairment.

MDV not only look like DMV for their double-membrane, they also transport selective cargo to peroxisomes and the endolysosomal system after mitochondrial stress, depending on Vps35 and syntaxin-17/SNAP29/VAMP7, respectively ^{26, 27, 32, 33} . This offers a pathway that could be involved in the intracellular transport of viral components to secondary vesicular structures ( i.e. ERGIC, lysosomes, multivesicular bodies) where viral particles are assembled and set ready for exocytosis. Interestingly, PINK1/Parkin and the mitochondrial ubiquitin ligase MULAN1 (MAPL) have been involved in MDV shedding from mitochondria ^{27, 34} . Notably, MULAN1 is known to be involved in the antiviral response of mitochondria ³⁵ . In addition, some MDV contain mitochondrial molecules, including the translocase complex (specifically TOM 20), components of the OXPHOS complexes, the VDAC, and/or pyruvate dehydrogenase ^{26, 29} . This opens the possibility that other mitochondrial components and resources for vRNA replication are transported or generated within MDV, such as proteins involved in the translocation of metabolites and solutes since MDV are like “chunks” of mitochondria. In this regard, it was recently shown that DMV have pores that span the double-membrane that mediate the export of vRNA from DMV and could also mediate the exchange of molecules between DMV and the cytoplasm ²² . These authors also showed that once vRNA is exported, it complexes with viral N protein, association that is known to increase vRNA translation in trans ³⁶ , which could probably occur at ribosomes located at the external membrane of DMV ¹⁶ . It is important to note that MDV have a very short shedding dynamic (~30 s), and peak at 2 h after stimulation ^{26, 34} , this could explain why the shedding step is not frequently spotted by TEM. Interestingly, in a recent paper, an alternative mechanism of mitochondria fission has been described that occurs under stress and high energy demand, that depends upon the MOM molecule Fis1, which yields small mitochondria destined for mitophagy ³⁷ . A common feature of RO are nearby mitochondria, which may show signs of cisterna swelling and disorganization, similar to mitochondria with induced MDV shedding ^{26, 27, 29, 32} , or membrane continuity with DMV ^{16, 19, 23– 25} . Nevertheless, some TEM images have shown budding of what could be DMV from mitochondria ( Figure 2B). In some cases, in closely located DMV and mitochondria, vRNA signal can be observed within both, apposed to each other ( Figure 2C). Moreover, in nsp4 temperature-sensitive mutants, at a non-permissive temperature, there is an increase in mitochondria size, with enlarged cisternae, and increased localization of nsp4 and nsp3 at mitochondria, perhaps resulting from the reduction of MDV shedding, that in turn results in a reduced number of DMV ¹⁵ . Interestingly, electron tomography has shown what could be an intermediate between DMV and MDV, a vesicle tethered to the ER with a double-membrane that contains a cisterna-like arrangement of the inner membrane ( Figure 2D) ²³ . Further support for the notion that mitochondria could be targets of SARS-CoV-2 vRNA infection that leads to DMV assembly, comes from the observation that shows mitochondria containing newly synthesized vRNA ¹⁹ . This suggests that mitochondria somehow get vRNA that could induce stress and therefore shedding of MDV (/DMV), where the vRNA replication machinery and newly synthesized vRNA are mostly located ¹⁹ . Interestingly, it was recently found that Fe-S cofactors, of which biosynthesis initiates at the mitochondria, are involved in SARS-CoV-2 RdRp function ³⁸ . Moreover, a recent in silico analysis predicted SARS-CoV-2 RNA localization to host mitochondria and nucleolus, further supporting this idea ³⁹ . In addition, it is possible that the abundance of vRNA and viral proteins within mitochondria are under tight control through the shedding of MDV, and thus, that only a few vRNA are found within mitochondria at a given moment. Interestingly, some images have shown that vRNA located inside of mitochondria is polarized towards DMV pools ( Figure 2C and E) ¹⁹ .

Supporting also the role of mitochondria for DMV assembly is the unexpected identification of several mitochondria molecules involved in different mechanisms of its physiology (electron transport, metabolism, mitochondria ribosomes, RNA maturation, and cellular immune signaling) as interactors of viral proteins ⁴⁰ . Some of the putative relevant interactions that this work identified is that of nsp4 with the inner mitochondria membrane translocase (TIMM) complex, the interaction of ORF9b with TOMM70, and the interaction of nsp6 and ORF9c with the Sigma receptor. This receptor has been involved in several mitochondria functions, related to its location at the mitochondria-associated ER membranes (MAM) ⁴¹ , enriched with interactors of nsp2 and 4. Additional intriguing, unexpected interactions were those of SARS CoV-2 membrane (M) protein with FASTKD5, involved in mitochondrial RNA maturation, and that of nsp8 with different mitochondria ribosomal proteins (MRP). Strikingly, interactions of ORF3a and M protein with relatives of known partners of MULAN1 (REEP and TRIM) were also identified. In a different study, nsp2 was found to interact with VDAC2 ⁴² , the mitochondrial porine, whereas the mitochondria antiviral-signaling protein (MAVS) has also been identified as a target of SARS-CoV-2 infection ¹⁰ . Together, these interactions of viral proteins with the host support that mitochondrial function is very relevant for SARS-CoV-2 infection. Furthermore, since the ribosome, mitochondrial RNA maturation and translocation mechanisms are targets of viral proteins, that according to the current model of infection are unexpected, these findings also hint that SARS-CoV-2 infects mitochondria as part of its replication cycle, rather than only taking control of this organelle through viral proteins synthesized elsewhere. The down-regulation of mtDNA encoded genes and mitochondrial RNA in patient autopsies by SARS-CoV-2 also supports this notion ⁴³ . Other relevant interactions of the viral proteome with mitochondrial proteins have been analyzed by others ^{8, 10, 42} . A key question in this scenario is which are the steps that mediate the shedding and transformation of MDV into DMV, and how viral proteins are involved ( Figure 1 question mark 1).

How MDV–DMV are induced by SARS-CoV-2? This question has no answer yet; however, there are at least two main possibilities that are non-self-exclusive: one that is consistent with the current paradigm is that viral proteins synthesized at the ER and/or its membrane modifications in the RO somehow reach mitochondria, modulate its physiology and induce DMV, in 1–2 hours. In this regard, there are some viral proteins with mitochondria localization sequences such as 3b ⁴⁴ , or that target proteins at the inner mitochondria membrane (IMM) ( i.e., TIMM, electron transport proteins, and MRPS), at the MOM (TOM), or at the mitochondria matrix (FASTKD5). Alternatively, the mitochondria could start viral protein synthesis with ribosomes located at the MOM ⁴⁵ , and/or uptake vRNA from the cytoplasm after virus entry and vRNA release into the cytoplasm. In this regard, it is known that mitochondria are capable of importing RNA from the cytoplasm through a pathway that involves the TOM/TIMM complex ⁴⁶ , and SARS-CoV-2 RNA is predicted to locate at this organelle ³⁹ . On the other hand, a tantalizing possibility is that vRNA accesses mitochondria directly from vesicles shed from the plasma membrane (PM) in which SARS-CoV-2 is endocytosed. This PM–mitochondria pathway mediates caveolin transport to mitochondria in myocytes after stress ⁴⁷ ( Figure 3A), and it could be an early step of what we called plasma membrane-mitochondria bridges, which we recently described in astrocytes ( Figure 3B) ⁴⁸ , involved in the emerging pathway of PM–mitochondria interactions ⁴⁹ . These PM-mitochondria bridges contain vesicles, most probably caveolae and mediate mass transfer from PM to mitochondria in minutes. Given that DMV are induced early by coronavirus (1–2 h p.i.), direct access of vRNA to mitochondria seems plausible, providing also the possibility to synthesize some viral proteins at IMM tethered mitoribosomes, with which nsp8 interacts ⁴⁰ . Notably, mitoribosomes synthesize most exclusively membrane proteins that are co-translationally inserted into the membrane with the participation of OXA1 ^{50, 51} , as it is the case of transmembrane-containing nsp3, 4 and 6, involved in vRNA replication, of which nsp3 and 4 have been located at DMV and colocalize with nsp2, 5, 8, 12, 13 and 15 ^{16, 19, 22, 24, 25} . Interestingly, the M protein could optimize vRNA translation through its interaction with mitochondrial FASTKD5 protein ⁴⁰ , involved in mitochondrial RNA maturation ⁵⁰ , and at the same time, viral replication could profit the mitochondria synthesized Fe-S cofactors required for RdRp function ³⁸ . This scenario could provide the advantage of the protected environment of mitochondria matrix, rich in ATP, avoiding the requirement of large amounts of protein to be transported from ER to mitochondria that would require energy and time.

However, the main concern against the idea that mitochondria can directly uptake SARS-CoV-2 from PM caveola comes from one study suggesting that SARS-CoV endocytosis is caveolin-independent ⁵² . This finding is based on the observation that cholesterol sequestration (one of the main components of lipid rafts that in turn is endocytosed by caveolae, ⁵⁰ ) with nystatin and filipin did not block pseudovirus entry. Indeed, nystatin enhanced it, whereas another cholesterol sequestering molecule, MβCD, did block it, therefore raising doubts about how cholesterol is involved. As matter of fact, different mechanisms of endocytosis have been found to mediate SARS-CoV-2 internalization ^{52, 53– 55} . In addition, in this study and a different one with CoV NL63, a major lack of colocalization between viral spike protein and caveolin-1 at 20 or 60 min p.i. was also considered as evidence for a caveolin-independent mechanism. However, the putative fast dynamic nature of this interaction (since we found that mass is transferred from PM to mitochondria in ~2 min) nor extracellular conditions were considered in these approaches. Extracellular conditions are expected to be acidified in the inflammatory setting and the extracellular acidification rate (ECAR) has been found increased in SARS-CoV-2 infected monocytes ³¹ . Importantly, extracellular acidification is known to induce transfer of caveolae to mitochondria ( Reviewed in 47 ). In addition, some evidence supports a role of the caveolae pathway in SARS-CoV-2 endocytosis: a) lipid rafts integrity is required for SARS-CoV entry and ACE2 is localized into lipid rafts, that are endocytosed through caveolae, well-known signaling hubs ^{54, 57} ; b) the S protein co-fractionates with caveolin-1 after binding to ACE2 ⁵⁴ ; c) an in silico approach found that SARS-CoV-2 proteins S, M, orf3, and replicase 1AB have putative caveolin binding motifs ⁵⁸ ; and d) orf3a protein binding to caveolin has been demonstrated experimentally ⁵⁹ . Thus, the precise role of cholesterol, caveolae, and caveolin for SARS-CoV-2 infection requires further investigation, because direct viral targeting to mitochondria could be of great relevance for SARS-CoV-2 infection. Interestingly, cholesterol-bound RNA probes are targeted to mitochondria ⁶⁰ . Furthermore, several alternative entry factors to ACE2 and facilitators capable to mediate SARS-CoV-2 infection have been identified ^{61– 64} , and they could mediate SARS-CoV-2 caveolae-mediated endocytosis. Taken together, it is possible that the diversity of receptors and entry pathways exploited by SARS-CoV-2, together with the fast dynamics of PM–mitochondria communication can obscure the caveolae role.

CM-convoluted membranes; DMS-double membrane spherules; DMV-double membrane vesicels; ER-endoplasmic reticulum; ERGIC-ER-Golgi intermediate compartment; IMM-inner mitochondria membrane; M-mitochondria; MAM-mitochondria associated membrane; MAVS-mitochondria antiviral-signaling protein; MOM-mitochondria outer membrane; nsp-non-structural protein; nsp-non-structural protein; PM-plasma membrane; TIMM-translocase of the mitochondrial inner membrane; TOM-translocase of the mitochondrial outer membrane; VDAC-Voltage-dependent anion channel; VP-vesicle packet (not shown); vRNA-viral RNA; zER-zippered ER.

All data underlying the results are available as part of the article and no additional source data are required.

PMOB did all work related with this manuscript.