PNPase but not SUV3 is required for the export of mtdsRNAs to the cytosol

In a previous collaborative study, BAK and BAX participated in mtdsRNA export (Dhir et al, 2018), akin to the release of mtDNA during apoptosis (Rongvaux et al, 2014; White et al, 2014; McArthur et al, 2018). However, as BAK and BAX are typically associated with apoptosis and patients with PNPase mutations survive until late in life (Vedrenne et al, 2012; von Ameln et al, 2012; Eaton et al, 2018; Sato et al, 2018; Rius et al, 2019), activation of a classic apoptotic pathway is unlikely. We sought to characterize the mtdsRNA export pathway in detail. We used immunofluorescence microscopy with a monoclonal antibody (J2) that is specific for dsRNA and is widely used to detect viral dsRNA in plant and animal cells (Weber et al, 2006). The monoclonal J2 antibody was developed against the dsRNA killer virus of Saccharomyces cerevisiae and recognizes dsRNAs that are ∼40–50 nucleotides or longer (Schonborn et al, 1991).

We first confirmed a linear pathway in which SUV3 knockdown (KD) followed by PNPase KD was required for release of mtdsRNAs to the cytosol (Dhir et al, 2018). An RNAi KD strategy was used instead of genomic deletion for PNPase (Fig S1A) because cells that lack PNPase lose mtDNA and become rho null (ρ⁰) (Shimada et al, 2018). SUV3 was also knocked down with RNAi (Fig S1B). Previous studies have suggested that PNPase and SUV3 form a complex (Borowski et al, 2013), suggesting that their expression may be co-dependent However, we show that PNPase and SUV3 protein levels are not dependent upon each other (Fig S1A and B), indicating that these proteins are not bona fide partners. As a negative control (Ctl), a non-targeting RNAi construct did not affect the expression of PNPase or SUV3.

We investigated the subcellular localization of mtdsRNAs when PNPase and/or SUV3 were knocked down in HeLa cells. With the non-targeting RNAi construct (Ctl), mtdsRNAs (marked by red fluorescence) were essentially not detected in mitochondria, which were marked with TOMM40 (green fluorescence) (Fig 1). SUV3 KD resulted in accumulation of mtdsRNAs in mitochondria in small puncta that overlapped with the mitochondrial marker TOMM40. However, a fraction of mtdsRNAs was detected in puncta in the cytosol when PNPase was knocked down (Fig 1). When both PNPase and SUV3 were knocked down, mtdsRNAs mostly remained in mitochondria, indicating that SUV3 is required before PNPase for mtdsRNA release. Pearson’s correlation coefficient (PCC) in ImageJ software was used to quantify co-localization of mitochondria and mtdsRNA (Dunn et al, 2011). PCC values of 0.73, 0.69, and 0.55 were calculated for KD of SUV3, SUV3/PNPase, and PNPase, respectively. Portions of the images were also magnified to show localization of mtdsRNA with respect to mitochondria. Thus, the mtdsRNA seems to specifically show decreased co-localization with mitochondria when PNPase is knocked down.

To determine the time frame in which mtdsRNAs were released, a time-course study was done when PNPase levels were reduced by RNAi (Fig S2). Over 96 h, cells remained healthy. After ∼48 h, mtdsRNAs (marked by red fluorescence) began to move out of mitochondria (marked with green fluorescence) and localize to the cytosol (PCC = 0.59), and at ∼60 h, the mtdsRNAs were found in cytosolic puncta. Based on this time course, 60–72 h post-RNAi treatment (PCC = 0.56) was selected for studies to evaluate mtdsRNA localization. At 96 h, mtdsRNAs were in red puncta and the mitochondrial network appeared largely fragmented (PCC = 0.41). Because we have previously shown that cells without PNPase lose mtDNA, cells will not continue to grow unless adapted to glycolytic media supplemented with uridine (Shimada et al, 2018). In summary, mtdsRNAs were released to the cytosol in a controlled manner.

The mtdsRNAs partially co-localize with stress granule protein TIA-1, and the type 1 IFN stress response is induced

To independently demonstrate that the dsRNAs detected above originated from mitochondria, we performed immunoprecipitation with the J2 antibody from a total cell lysate followed by Nanopore long-read sequencing. Briefly, the immunoprecipitated RNAs were polyadenylated in vitro and then sequenced using the Oxford Nanopore platform (Liu et al, 2022). After alignment with the mitochondrial genome, the RNA-sequencing profile for cells with PNPase KD showed abundant reads that mapped to both the heavy (H)- and light (L)-strands of the mitochondrial genome, implying the presence of intermolecular dsRNA (Fig S3). Most abundant mtdsRNAs aligned to the ribosomal (r)RNAs and displacement (D)-loop regions. In addition, reads were also detected that aligned with the coding regions of the mitochondrial genome, particularly Cox1 and Cox2. For SUV3 KD, mtdsRNAs were also detected for the rRNAs and D-loop, whereas mtdsRNAs were barely detected for the control cells. Because the L-strand for the rRNAs was abundant in PNPase KD, a transcript for the entire L-strand was generated from the origin of L-strand transcription.

To confirm the specificity of the dsRNAs, we also queried the sequencing reads with the nuclear genome. We only detected dsRNAs that aligned with the coding and non-coding strands for the 18S rRNA when PNPase or SUV3 was knocked down (Fig S4). Previously, dsRNAs for the 18S rRNA have not been reported, but a class of antisense ribosomal siRNAs (risiRNAs) has been identified that functions in ribosomal biogenesis (Zhu et al, 2018). The dsRNA for the 18S rRNA may have a novel role in attenuating cytosolic translation during the IFN-1 response (Wan et al, 2021). In sum, the sequencing analysis confirms that the dsRNAs detected by immunofluorescence were of mitochondrial origin.

MDA5 and RIG-I are dsRNA sensors in the cytosol that recognize long (>2.0 kb) and short (<2.0 kb) dsRNAs, respectively, and relay the IFN-1 response (Kato et al, 2008; Peisley et al, 2011). Our previous study indicated that knockdown of MDA5 in PNPase-depleted cells mainly blocked the IFN-1 response, but RIG-I also did to a lesser extent (Dhir et al, 2018). We determined whether MDA5 co-localized with mtdsRNA in cells that were depleted for PNPase using immunofluorescence microscopy (Fig 2A). In control cells, mtdsRNA was not detected and MDA5 (green) levels were low. However, PNPase KD resulted in increased MDA5 expression throughout the cell, but it failed to co-localize with mtdsRNA (red). It may not be surprising that MDA5 did not markedly co-localize with cytosolic mtdsRNAs, because published studies have noted that MDA5 did not bind to viral RNAs and the viral RNAs seemed to localize to stress granules (Langereis et al, 2013). Thus, MDA5 likely binds to mtdsRNAs transiently when initiating the IFN-1 response.

Given that mtdsRNAs are present in large structures that are reminiscent of cytosolic stress granules, we tested whether mtdsRNAs co-localized with the RNA binding protein TIA-1, which promotes the assembly of stress granules (Gilks et al, 2004; Asadi et al, 2021). A subset of TIA-1 (green) seemed to co-localize with mtdsRNAs (red) in large puncta at 72 h post-PNPase KD (Fig 2B), suggesting the mtdsRNAs are moved to stress granules.

The presence mtdsRNAs in the cytosol mimics a viral invasion and triggers an IFN-1 response (Dhir et al, 2018). Thus, measuring the IFN-1 response is a more specific method to assess the IFN-1 response than quantitating the amount of mtdsRNA that localizes to the cytosol versus the mitochondria. We used qRT–PCR to quantitate the expression of a set of genes for the IFN response including IFNB1 for IFN-β and the IFN-stimulated genes (ISGs) IFI44 for antiviral activity and ISG15 for signaling (Dzimianski et al, 2019). PNPase KD but not SUV3 KD resulted in the increased expression of the IFNB1 and ISGs, indicating that the mtdsRNA indeed reaches the cytosol to induce the IFN-1 response (Fig 2C). Because MDA5 specifically binds to mtdsRNA (Dhir et al, 2018), we investigated RNA levels of IFIH1 (gene for MDA5) with qRT–PCR 72 h after various RNAi treatments (Fig 2D). When PNPase was knocked down but not SUV3 or SUV3 and PNPase, IFIH1 transcript levels increased markedly. Because MDA5 response with PNPase KD seemed very specific, we used IFIH1 transcript analysis for subsequent studies. Thus, the IFN-1 pathway is specifically activated when PNPase is knocked down, confirming localization of mtdsRNA to the cytosol.

PNPase functions in the intermembrane space (IMS) and the matrix

Published studies report that PNPase localizes to the IMS and matrix (Chen et al, 2006; Rainey et al, 2006; Borowski et al, 2013). To determine the compartment in which PNPase functions, we replaced the endogenous mitochondrial targeting sequence (MTS), which consists of amino acids 1–37. To target to the IMS, a construct (designated IMS-PNPase) with the MTS (amino acids 1–68) from serine β-lactamase–like protein LACTB was generated (Polianskyte et al, 2009; Rhee et al, 2013). To target to the matrix, a construct (designated Mat-PNPase) with the MTS (amino acids 1–69) of N. crassa F_o-ATPase subunit 9 was generated (Pfanner et al, 1987). The constructs were integrated into a HeLa T-Rex FLP-In cell line; expression was induced with 500 ng/ml doxycycline, and endogenous PNPase was simultaneously silenced (Fig S5A). Mat-PNPase and IMS-PNPase contained a C-terminal FLAG tag for detection, and the empty vector was also included as a control. Experiments in which the mitochondrial outer membrane was ruptured by incubation in hypotonic buffer but the inner membrane remained intact indicated that the constructs were targeted correctly (Fig S5B). Abundant mtdsRNAs (marked by red fluorescence) in the cytosol were detected upon knockdown of endogenous PNPase (Fig 3). However, induction of PNPase targeted to the matrix or IMS resulted in markedly reduced mtdsRNA puncta (Fig 3). Taken together, these results indicate that PNPase surprisingly functions in both the IMS and matrix.

Outer membrane components VDAC and BAK/BAX mediate transport of mtdsRNAs

In collaborative studies, we previously showed that BAK or BAX KD in combination with PNPase KD blocked the IFN-1 response (Dhir et al, 2018), suggesting that mtdsRNAs were not exported from mitochondria to the cytosol. Thus, we investigated the localization of mtdsRNA in the MEF cell line in which BAX and BAK were both knocked out (Bak/Bak DKO) (Takeuchi et al, 2005). RNAi constructs specific for mouse PNPase or a non-targeting RNAi construct (Ctl) were used for KD. The mtdsRNA (red) localization was determined by immunofluorescence, and mitochondria were marked with an antibody against TOMM40 (green) (Fig 4A). In WT MEFs, mtdsRNAs localized to the cytosol when PNPase was deleted (data not shown). However, in the MEFS that lacked BAK and BAX with PNPase KD, mtdsRNAs co-localized with mitochondria (PCC = 0.79) and were not released into large cytosolic puncta (Fig 4A). Instead, the mitochondrial morphology seems to be altered in that an enlarged mitochondrial network surrounds the mtdsRNA.

The BAX inhibitor peptide V5, which blocks BAX translocation to mitochondria (Gomez et al, 2007), was also used to determine whether mtdsRNA export was attenuated (Fig 4B). HeLa cells were treated with 100 μM V5 peptide (Gomez et al, 2007), and PNPase was simultaneously knocked down followed by immunofluorescence microscopy. Whereas mtdsRNA (green) localized to the cytosol in PNPase KD (PCC = 0.50), the addition of V5 peptide resulted in mtdsRNA that showed increased co-localization with mitochondria (red, PCC = 0.74). Taken together, BAK and BAX are important for mtdsRNA localization to the cytosol.

Because VDAC mediates the import of RNAs into mitochondria, as well as the export of mtDNA from mitochondria (Kim et al, 2019), we also considered it as a candidate for mtdsRNA export. Of the three VDAC genes, VDAC1 is expressed most abundantly, followed by VDAC2, and finally, VDAC3 is least expressed (Yamamoto et al, 2006). VDAC2 through association with BAK inhibits apoptosis (Cheng et al, 2003), suggesting the VDAC2 may have a unique function. We investigated a combination of VDAC1 or VDAC2 KD with PNPase KD (Fig 5). The mtdsRNA (green) localized to the cytosol when PNPase alone was knocked down (PCC = 0.51), but KD combinations of VDAC1 (PCC = 0.81) or VDAC2 (PCC = 0.68) with PNPase resulted in mtdsRNA that markedly co-localized with mitochondria (red). Of note, the mitochondrial network seems more fragmented with VDAC KD. MDA5 expression also was not induced markedly when by VDAC1 or VDAC2 KD alone or in combination with PNPase KD (Fig 2D), supporting that mtdsRNA was not being exported to the cytosol.

To complement the VDAC knockdown studies, we used the chemical 4,4′-diisothiocyanatostilbene-2,2′-disulfonate (DIDS) that inhibits the VDAC channel activity (Ben-Hail & Shoshan-Barmatz, 2016). DIDS is routinely used as an inhibitor for VDAC by blocking the channel and/or altering oligomerization (Ben-Hail & Shoshan-Barmatz, 2016). A titration curve with DIDS was conducted over 72 h, and a concentration of 50 μM, which did not inhibit cell growth or viability and was similar to the published literature (Kim et al, 2019), was selected. PNPase KD or the control RNAi construct was added to cells, and 24 h post-transfection, 50 μM DIDS was added. At 72 h post-transfection, cells were imaged for mtdsRNA and mitochondria (Fig S6). In cells treated with DIDS that lacked PNPase, mtdsRNA (red) co-localized to mitochondria (green). In contrast, mtdsRNA showed less co-localization with mitochondria as expected when cells were not treated with DIDS. Thus, treatment with DIDS and knockdown of VDAC resulted in mtdsRNAs remaining in mitochondria. In sum, outer membrane components VDAC, BAK, and BAX seem to play a role in mtdsRNA export to the cytosol.

Prohibitin 1 and 2 are required for mtdsRNA export

The mechanism by which mtDNA or mtdsRNA crosses the mitochondrial inner membrane is not clear. Published studies suggest that mtDNA might be released by herniation of the inner membrane (McArthur et al, 2018; Riley et al, 2018) or by mitochondrial-derived vesicles (Cadete et al, 2016). Given that mitochondrial physiology was not severely compromised in a conditional mouse model that lacked PNPase in the liver (Wang et al, 2010), we reasoned that the inner membrane might have a regulated mechanism for mtdsRNA release that does not severely compromise overall mitochondrial function. In contrast, the release of mtDNA from mitochondria might be expected to result in apoptosis, because mtDNA is essential for viability. From mass spectrometry experiments and the published literature, we considered (1) the anionic inner membrane channel CLIC5, (2) the BAK/BAX-interacting inner membrane protein GHITM, and (3) inner membrane proteins that may regulate inner membrane dynamics, prohibitin 1 and 2 (PHB1/2), as potential candidates. Although the interactions with PNPase in mass spectrometry studies were not robust (Yoshinaka et al, 2019), PHB1/2 and CLIC5 are also implicated in the IFN-1 response (Langlais et al, 2016; Yoshinaka et al, 2019; Refolo et al, 2020), which is similar to PNPase. The KD of CLIC5 or GHITM in combination with PNPase KD did not block the release of mtdsRNAs (red) from mitochondria (green) (Fig S7A and B), thereby eliminating them as potential candidates. Because PHB1 and PHB2 co-assemble, the proteins were knocked down individually (Nijtmans et al, 2000; Tatsuta et al, 2005). When PHB1 (Fig 6) or PHB2 (data not shown) was knocked down singly, mtdsRNA was not detected, in contrast to KD of PNPase (PCC = 0.42). The KD of PHB1 (PCC = 0.74) or PHB2 (PCC = 0.73) in combination with PNPase KD blocked the release of mtdsRNAs (red) to the cytosol, and the mtdsRNA remained in mitochondria (green) (Fig 6). In addition, the abundance of mtdsRNAs seemed to be somewhat decreased. To confirm that the mtdsRNA did not induce the IFN-1 response, MDA5 expression was not markedly elevated when PNPase was knocked down in combination with PHB1 or PHB2 (Fig 2D).

We also tested the PHB1/2 inhibitor, rocaglamide (Roc-A), which has been shown to inhibit PHB1/2 assembly (Polier et al, 2012). Roc-A was titrated over a 72-h period, and a concentration of 25–50 nM did not seem to impair cell growth. Cells were knocked down with PNPase or the control RNAi construct, and 25 nM Roc-A was added after 24 h followed by imaging at 72 h (Fig S8). The mtdsRNA (red) pool again co-localized with mitochondria (green), indicating that PHB1/2 seem to play a role in the export of mtdsRNAs to the cytosol. Thus, a candidate for release of mtdsRNAs from mitochondria is the PHB1/2 complex.

Cytosolic mtdsRNAs are detected in a subset of lung cancers

Release of mtDNAs has been reported in different cancer cell lines (Wu et al, 2021), particularly in non–small-cell lung cancer (NSCLC) cell lines where mtDNA release induces the STING signaling pathway (Kitajima et al, 2019). In addition, Roc-A treatment was shown to inhibit proliferation, migration, and anchorage-independent growth in a subset of KRAS-mutated NSCLC cell lines (Yurugi et al, 2017), suggesting mtdsRNA may be an important DAMP. In previous studies, we have characterized NSCLC cell lines and found they can be separated by preference for oxidative (OXPHOS^HI) versus glycolytic (OXPHOS^LOW) metabolism (Momcilovic et al, 2019). We reasoned that, in addition to mtDNA export, mtdsRNAs might also be exported in a subset of NSCLCs, particularly those that are glycolytic (Shimada et al, 2018). Analysis of The Cancer Genome Atlas (TCGA) database showed that PNPase (PNPT1) gene expression was up-regulated significantly in a subset of lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) lines (Fig S9) (Momcilovic et al, 2018). In addition, PNPase expression was up-regulated in numerous tumor types, compared with controls (Fig S9). Because PNPase is an IFN-1–induced gene (Leszczyniecka et al, 2003), increased PNPT1 expression in tumors may reflect those that have an increased IFN-1 response.

With well-characterized NSCLC lines, we investigated whether mtdsRNAs were detected using immunofluorescence. Barbie and colleagues previously reported in the LUAD cell line H1792 (co-mutations in KRAS and TP53, designated KP) that mtDNA was not detected in the cytosol (Kitajima et al, 2019); this cell line has an OXPHOS^HI metabolic profile (Chen et al, 2019). Similarly, immunofluorescence studies showed that mtdsRNAs were not detected (Fig 7). In addition, the cell line H2009, which has an OXPHOS^HI metabolic profile as well, lacked mtdsRNA. In contrast, abundant mtdsRNAs were present in LUAD cell lines, HCC44 and H23, with co-mutations in KRAS and STK11/LKB1 (designated KL) (Fig 7). These cell lines have an OXPHOS^LOW, glycolytic, metabolic profile (Meijer et al, 2018; Majem et al, 2020). In the cell line H23, a cohort of the mtdsRNA seemed to localize to the cytosol in larger puncta when compared to the mtdsRNA cohort in HCC44 cells. Thus, in a subset of NSCLC cell lines, mtdsRNA may be exploited as a DAMP for signaling mitochondrial stress.

mtdsRNAs in a subset of NSCLC lines are detected by sequencing

To confirm that the dsRNAs are derived from mitochondria, we sequenced mtdsRNA from the NSCLC cell lines (Dhir et al, 2018). Briefly, a cell lysate was treated with DNase I to degrade DNA, then treated with S1 RNase to remove single-stranded RNA. To the remaining dsRNA, adapters were ligated followed by strand-specific RNA sequencing. The reads were aligned against the mitochondrial genome (Fig 8A). As a negative control, a cell line (1B-S) in which mtdsRNA was not detected by immunofluorescence was included (Fig 8B); a small amount of mitochondrial ribosomal DNA was detected, which probably formed secondary structures (Fig 8B). Similarly, a low amount of mtdsRNA was detected in the H2009 cell line. In contrast, a large amount of mtdsRNA was detected along the entire mitochondrial chromosome in the H23 and HCC44 cell lines. Interestingly, the H23 cell line has large amounts of rRNA, whereas the HCC44 cell line had greater reads in the coding region. Thus, mtdsRNA was most abundant in cell lines HCC44 and H23 as shown in the bottom panel (Fig 8B), confirming that mitochondria are the source of the dsRNA. Studies in NSCLC cell lines support that mtdsRNA may be a novel DAMP in lung cancer, supporting a physiologic role of this pathway in cancer.

PNPase seems to function in both the matrix and IMS with PHB1/2

PNPase has been placed in numerous compartments. Originally, PNPase was identified as a cytosolic protein that played a role in terminal differentiation and cellular senescence (Leszczyniecka et al, 2002). However, PNPase has an N-terminal targeting sequence that directs its import into the mitochondrial IMS in a pathway that depends on Yme1 in vitro (Rainey et al, 2006). Submitochondrial fractionation also showed that PNPase resided primarily in the IMS (Chen et al, 2006). Additional studies suggested that a fraction resides in the matrix and binds to SUV3 (Borowski et al, 2013), which aligns with its role in RNA degradation. Finally, PNPase has been reported to have a role in decay of cytosolic poly(A) RNAs when released during apoptosis (Liu et al, 2018), but late release of PNPase does likely not support degradation of cytosolic RNA as its primary function (Chen et al, 2006).

Given that loss of PNPase results in the export of mtdsRNAs to the cytosol, we took advantage of this activity to target PNPase to the IMS or matrix by appending a different targeting sequence. The IMS and matrix targeting sequences localized PNPase to the IMS or matrix, respectively. When endogenous PNPase was knocked down by RNAi in combination with the induced expression of the matrix- and IMS-forms of PNPase, the mtdsRNA levels were drastically reduced. These results support that PNPase seems to function in both the IMS and matrix. Mitochondrial proteins that are dual-localized to the matrix and IMS are rare. Potentially, PNPase is part of a larger export complex or channel that may contain PHB1/2, allowing PNPase to function on either side of the IM.

We used a candidate approach to determine potential components that function at the IM to release mtdsRNAs. Our criteria were IM proteins that had potential channel properties and induced the IFN-1 response, as well as interactions with PNPase (albeit weak). Previously, we have tried to identify partner proteins for PNPase but have not identified robust partners, and PNPase in yeast migrates in the same-sized complex as in mouse mitochondria at the inner membrane (Chen et al, 2006). Thus, PNPase seemed to lack partner proteins and may instead transiently associate with other partners.

We considered GHITM and CLIC5, but mtdsRNAs were exported when PNPase was knocked down. Instead, reduction of PHB1/2 by knockdown or inhibition with the small molecule inhibitor Roc-A seemed to prevent the export of mtdsRNAs to the cytosol. The abundance of mtdsRNAs also seemed to be somewhat decreased. Given that PHB1/2 and PNPase expression is increased by the IFN-1 response (Yoshinaka et al, 2019), they seem to align well for export of mtdsRNA from the IM. Early cryo-EM studies showed that PHB1/2 can form a large ring (Tatsuta et al, 2005), but its structure in the mitochondrial inner membrane is not known. PHB1/2 is structurally related to the bacterial HflKC complex that binds to the AAA protease FtsH (Qiao et al, 2022). A cryo-EM structural analysis showed that 12 copies of HflK and HflC form a large cage in the membrane and four FtsH hexamers with periplasmic domains and transmembrane helices enclosed inside the cage and cytoplasmic domains situated at the base of the cage (Qiao et al, 2022). This large complex has a pore that is the size of the nascent polypeptide exit channel of the ribosome; however, given the intricacy, the complex could be dynamic in accommodating the transport of cargos such as mtdsRNA. Whether the mitochondrial AAA proteases YmeL1 and Afg3L2 function with PHB1/2 in export of mtdsRNAs is not known. Interestingly, YmeL1 has a role in mtDNA escape (Thorsness et al, 1993; Sprenger et al, 2021). Alternatively, the heptad repeat region of PHB2 has been suggested to form a scaffold with CLPB, ATAD3A, and AKAP1 and invoke the immune response from the mitochondrial signalosome (Yoshinaka et al, 2019). It is also likely that other proteins may participate and different pathways may exist for mtdsRNA export. A recent work by Wredenberg and colleagues showed that mtdsRNAs escape from mitochondria under conditions in which mitochondrial mRNA polyadenylation or degradation was disrupted in the fly model (Pajak et al, 2019). Finally, protein kinase RNA-activated recognizes short mtdsRNAs and can move in and out of mitochondria (Kim et al, 2018).

Mitochondrial outer membrane permeabilization is important for release of mtdsRNAs at the outer membrane

The release of mtDNA from mitochondria at the outer membrane via mitochondrial outer membrane permeabilization with subsequent activation of the cGAS/STING pathway has been well established (McArthur et al, 2018; Riley et al, 2018; Xian et al, 2022). Here, we show that VDAC and BAK/BAX also play a role in the release of mtdsRNAs. In MEF cells that are knocked out for both BAK and BAX, mtdsRNA co-localized with mitochondria. In addition, VDAC also seems to play a role. This may not be so surprising as mtDNA relies on the same proteins; however, the specific mechanism and similarities between mtDNA and mtdsRNA export are not known. A recent study suggests that the assembly rate of BAX and BAK in large channels is an important determinant of the kinetics of mtDNA release (Cosentino et al, 2022). The release of mtDNA is often associated with apoptosis (Cosentino et al, 2022); however, mtdsRNA release did not result in apoptosis, suggesting the export pathways may differ. VDAC may be an important player for differentiating between mtDNA and mtdsRNA. Alternatively, mtdsRNA may be released under low-stress conditions, with mtDNA being released under high-stress conditions when apoptosis is triggered.

In cancer, mtdsRNA as a new DAMP

A recent publication by Barbie and colleagues (Kitajima et al, 2019) dissected the underlying mechanism of immune resistance in NSCLC cell lines and found that LKB1 loss resulted in suppression of the IFN genes (STING) and insensitivity to cytoplasmic double-stranded DNA detection. They showed that mtDNA was released from mitochondria to the cytosol and activated STING and the innate immune response (Kitajima et al, 2019). Thus, mtDNA release from mitochondria represents an unexpected intrinsic tumor-sensing mechanism. We also find in a subset of NSCLC cell lines that mtdsRNA could be detected in LUADs characterized by an OXPHOS^LOW, glycolytic, metabolic profile (Shackelford & Shaw, 2009; Shackelford et al, 2013; Momcilovic et al, 2019). We confirmed that the dsRNA was derived from the mitochondria by sequencing. Indeed, most of the mitochondrial coding regions seemed to be detected, with increases in certain parts of the mitochondrial genome. In addition, the distribution of mtdsRNA in mitochondria versus cytosol seemed to differ between lung cancer cell lines with KP and KL mutations, suggesting that NSCLC cells with loss-of-function mutations in the LKB1 (STK11) gene may have stronger activation of the IFN-1 pathway. Given that mtDNA release is likely associated with apoptosis and innate immune responses, mtdsRNA may represent a new marker for characterizing NSCLC types and other cancers. Moreover, this pathway might be targeted for therapeutic strategies. For example, Roc-A is a member of the flavagline family that has antitumor properties and sorafenib suppresses the both DNA- and RNA-sensing–mediated IFN-1 pathway in hepatocellular carcinoma (Huang et al, 2022). Thus, the pathway by which mtdsRNAs are generated represents a new pathway to characterize broadly in cancer. In tumors that have an IFN-1 signature, mtdsRNA may be an activator (Budhwani et al, 2018). Alternatively, in tumors that become resistant to IFN signaling through cGAS/STING (Benci et al, 2016; Jiang et al, 2020), activation of mtdsRNA release may provide a new pathway to reactivate the immune system, particularly if mtDNA release was an initial activator (Kitajima et al, 2019). Additional published studies support that mtdsRNA is detected in numerous tumors (Killarney et al, 2023).

Cell lines and culture media

HeLa (American Type Culture Collection [ATCC]), HeLa T-REx Flp-In (Life Technologies), MEF Bax Bak DKO SVO (ATCC), and NSCLC cell lines H23 (ATCC), HCC44 (DSMZ), H2009 (ATCC), and H1792 (ATCC) were maintained at 37°C in a humidified incubator with 5% CO₂. Cells were grown in high-glucose DMEM supplemented with 10% FBS, 100 U/ml penicillin, and 100 μg/ml streptomycin, with the exception of the MEF cell line that was supplemented with 5% FBS. Transfections were performed using BioT transfection reagent (Bioland) or Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer’s instructions. All cell lines were routinely tested and confirmed to be free of mycoplasma using the MycoAlert mycoplasma detection kit (Lonza).

Chemicals

Chemicals were dissolved in DMSO, stored at −80°C, and used at the indicated concentrations. DMSO at 0.1% served as the vehicle control. The following chemical inhibitors were used: Bax-inhibiting peptide V5 (#196810; Sigma-Aldrich), prohibitin-binding rocaglamide-A (#S7428; Selleck Chemicals), and VDAC inhibitor 2,2'-(1,2-ethenediyl)bis(5-isothiocyanato-benzenesulfonic acid) (DIDS) (#16125; Cayman Chemical).

Constructs

Constructs to target PNPase specifically to the IMS (IMS-PNPase) or matrix (Mat-PNPase) were constructed by removing the MTS (amino acids 1–37) from PNPase and replacing it with the MTS (amino acids 1–68) from serine β-lactamase–like protein LACTB (IMS) (Polianskyte et al, 2009; Rhee et al, 2013) or the MTS (amino acids 1–69) of N. crassa F_o-ATPase subunit 9 (Pfanner et al, 1987). IMS-PNPase and Mat-PNPase contained a C-terminal FLAG tag for detection. The constructs were cloned into pcDNA5/FRT/TO and then integrated into a HeLa T-Rex FLP-In cell line by transfecting cells with 1.8 μg pOG44 (Life Technologies) and 0.2 μg pcDNA5/FRT/TO IMS-PNPase or Mat-PNPase expression construct using BioT transfection reagent in a six-well plate. After 48 h, the cells were transferred to a 10-cm dish and stable cells were selected with 200 μg/ml hygromycin B (Wako Pure Chemical Industries). Stable colonies were selected and maintained in 100 μg/ml hygromycin B. IMS-PNPase and Mat-PNPase expression was induced with 500 ng/ml doxycycline for 48 h.

RNAi knockdown and small molecule modulators

Silencer Select siRNAs as indicated in Table S1 were used for knocking down target genes. Briefly, HeLa cells were transfected with 20 nM gene-specific Silencer Select using Polyplus jetPRIME transfection reagent, and 48–96 h post-transfection, cells were harvested for knockdown confirmation and immunofluorescence studies. The knockdowns were confirmed by Western blot analysis. Bax-inhibiting peptide V5 (100 μM) was added simultaneously with an RNAi construct, whereas rocaglamide-A (25 nM) and DIDS (50 μM) were added 24 h post-RNAi transfection.

Immunofluorescence

Cells were transferred to six-well plates including coverslips. After 24 h, cells were attached to coverslips and were fixed in 4% formaldehyde, then permeabilized in 0.25% Triton X-100. Subsequently, cells were blocked in 0.1% phosphate-buffered saline with 3% BSA and 0.1% Tween-20, followed by incubation with the primary antibody (Table S2) overnight. Cells were washed and incubated with the appropriate secondary antibodies for 1 h. The cells were washed and transferred to mounting media (Antifade GOLD reagent with DAPI; Thermo Fisher Scientific). A 3i Spinning Disk Confocal microscope (Marianas) consisting of CSU-X1 A1 Spinning Disk (Yokogawa) attached to a Zeiss Axio Observer 7 microscope (Zeiss) was used. Standard filter sets were used for imaging. The PCCs were calculated using the JACoP plugin on ImageJ.

Western blotting

Cells were lysed in RIPA buffer (50 mM Tris–HCl, pH 7.5, 150 mM NaCl, 1% NP-40, 0.1% SDS, 0.5% sodium deoxycholate, 1 mM EDTA, 1 mM PMSF [phenylmethylsulfonyl fluoride]) for 30 min on ice. After centrifugation at 21,000g for 10 min, the protein concentration was measured with the BCA protein assay kit (Thermo Fisher Scientific). Equal amounts of protein extracts were separated by SDS–PAGE and transferred to Immobilon PVDF membranes (Millipore). Proteins were detected with the indicated antibodies (Table S2) followed by chemiluminescence and imaged on ChemiDoc MP Imager (Bio-Rad).

Mitochondrial isolation from cultured cells

For large-scale mitochondrial isolation, cells were resuspended in homogenization buffer (20 mM Hepes, pH 7.4, 220 mM mannitol, 70 mM sucrose, 0.5 mM PMSF, 2.5 mM NaF, and 1 mM Na3VO4) and ruptured with 30 strokes in a Teflon-glass homogenizer. The homogenized cellular extract was centrifuged twice at 850g for 10 min to obtain the post-nuclear supernatant. Mitochondria were pelleted at 10,000g for 10 min and washed once in homogenization buffer. For small-scale mitochondrial isolation, cells were resuspended in homogenization buffer, if necessary supplemented with 10 mM NEM and 10 μM MG132, and lysed by passing the cell mix through a 25-gauge needle 20 times with a 1-ml syringe. The extract was centrifuged twice at 770g for 5 min, and the supernatant was pelleted at 10,000g for 10 min to isolate mitochondria. The mitochondria were washed once with homogenization buffer. The yield was determined with the BCA protein assay kit after lysing the mitochondria in 0.1% SDS.

PNPT1 expression analysis in tumors

The gene expression data for PNPT1 from RNA sequencing and corresponding clinical information were retrieved from The Cancer Genome Atlas (https://portal.gdc.cancer.gov/). The expression of PNPT1 was analyzed as previously described (Momcilovic et al, 2018). The statistical tests were calculated using GraphPad Prism.

Immunoprecipitation of dsRNA

Cells were lysed in NP-40 lysis buffer (catalog #: J60766) for 30 min on ice. Protein A/G beads (catalog #: 88803) were incubated with J2 antibody suspended in NP-40 lysis buffer for 1 h at room temperature. The cell lysates were then incubated with the beads for 2 h at 4°C. The beads were washed with a high-salt PBS buffer (1x PBS, 0.1% Tween, 1 M NaCl) followed by two washes with a low-salt PBS buffer (1x PBS, 0.1% Tween, 0.5 M NaCl). The dsRNA from the immunoprecipitation was eluted by treating with 1 mg/ml Proteinase K (catalog #: P8044-1G) in NP-40 lysis buffer.

Sequencing of immunoprecipitated dsRNA with the Oxford Nanopore platform

The immunoprecipitated RNA with the J2 antibody was extracted with phenol–chloroform and then precipitated using ethanol. Sequencing libraries were prepared from the remaining in vitro polyadenylated RNAs (∼4 ng) using the PCR-cDNA barcoding sequencing kit from Oxford Nanopore (ONT, catalog #: SQK-PCB111.24) as per the manufacturer’s instructions. Sequencing was performed using R9.4.1 Flongle cells on a MinION Mk1B device and sequenced for 24 h. Basecalling was performed using Guppy Basecaller (version 6.1.1+1f6bfa7f8). Reads were then mapped to the human mitochondrial genome (see Fig S3) using Minimap 2 (version 2.17-r941). Reads were visualized using IGV (version 2.12.3), and figures were prepared using Inkscape (version 1.1.2). Data are available as FASTQ files and aligned bam files.

dsRNA sequencing and analysis in cancer cell lines

A cell lysate was treated with DNase I to degrade DNA followed by a 1-h incubation at 37°C with Nuclease S1 RNase to remove single-stranded RNA. RNA was purified and incubated with 50% DMSO vol/vol for 1.5 h at 65°C to minimize the secondary structure, then purified with RNeasy MinElute Cleanup Kit (QIAGEN). Strand-specific NGS libraries were constructed from the remaining dsRNA using the KAPA RNA HyperPrep kit (Roche) and sequenced on the Illumina NextSeq using 2 × 100 bp paired-end reads.

The dsRNA-sequencing data were processed using nf-core/rnaseq pipeline, version 3.8 (Ewels et al, 2020), with the following options: reference genome, hg19; adapter and quality trimming, Trim Galore; alignment and quantification route, STAR>Salmon; sort and index alignments, SAMtools; duplicate read marking, Picard MarkDuplicates.

The strandCheckR package (To & Pederson, 2021) was used to quantify the proportion of reads originating from double-stranded RNA. First, a 100-bp window that slides over a bam file (window step = 20 bp) was used to determine the proportion of the genome with reads mapping to both the “+” and “−” strands for each window (i.e., symmetrically aligned reads). Windows that primarily include reads derived from single-stranded DNA (i.e., asymmetrically aligned reads) were eliminated. For each sliding window, the proportion of ± stranded reads was computed. Windows with more than 70% of reads from a single strand (e.g., 25% positive, 75% negative) were classified as having single-strand coverage. The remaining window was deemed coverage for double-strand RNA. Coverage of double-strand RNA was adjusted by total coverage per chromosome. Visualization of the coverage per mitochondrial region was performed using Tableau.

Quantitative PCR analysis for gene expression

Cells were lysed by drawing through a sterile 21-gauge needle and syringe to homogenize the cell. The RNA from the total lysate was isolated using the QIAGEN RNeasy spin column kit according to the manufacturer’s instructions. RNA was reverse-transcribed into cDNA using Bio-Rad iScript cDNA Synthesis Kit (Bio-Rad) in accordance with the manufacturer’s protocols. Reverse transcriptase quantitative PCR was performed with the 2x SYBR Green qRT-PCR kit (Kapa Biosystems) and analyzed with the Roche LightCycler 480 II system (Roche). qRT-PCR primers are listed in Table S3. Samples from three biological replicates were plated for three technical replicates, and the average raw cycle threshold value across the three replicates was calculated.