VDAC oligomers form mitochondrial pores to release mtDNA fragments and promote lupus-like disease

See allHide authors and affiliations

Science  20 Dec 2019:
Vol. 366, Issue 6472, pp. 1531-1536
DOI: 10.1126/science.aav4011

VDACs are MOM's ruin

Mitochondrial DNA (mtDNA) is normally kept within the mitochondria. It can be released into the cytosol in response to stress and thus encounter cytosolic DNA sensors, triggering type I interferon responses. During apoptosis, mtDNA release is mediated by macropores in the mitochondrial outer membrane (MOM) created by oligomerization of the proteins BAX and BAK. Kim et al. found that during oxidative stress, mtDNA escapes instead through macropores formed by oligomerization of voltage-dependent anion channels (VDACs) (see the Perspective by Crow). In a mouse model of lupus, an inhibitor of VDAC oligomerization diminished mtDNA release and downstream signaling events. This treatment reduced lupus-like symptoms in the model, suggesting a potential therapeutic route for conditions mediated by mtDNA release.

Science, this issue p. 1531; see also p. 1445


Mitochondrial stress releases mitochondrial DNA (mtDNA) into the cytosol, thereby triggering the type Ι interferon (IFN) response. Mitochondrial outer membrane permeabilization, which is required for mtDNA release, has been extensively studied in apoptotic cells, but little is known about its role in live cells. We found that oxidatively stressed mitochondria release short mtDNA fragments via pores formed by the voltage-dependent anion channel (VDAC) oligomers in the mitochondrial outer membrane. Furthermore, the positively charged residues in the N-terminal domain of VDAC1 interact with mtDNA, promoting VDAC1 oligomerization. The VDAC oligomerization inhibitor VBIT-4 decreases mtDNA release, IFN signaling, neutrophil extracellular traps, and disease severity in a mouse model of systemic lupus erythematosus. Thus, inhibiting VDAC oligomerization is a potential therapeutic approach for diseases associated with mtDNA release.

Mitochondrial stress, such as that triggered by increased mitochondrial reactive oxygen species (mROS), can release mitochondrial DNA (mtDNA) into the cytosol. There, it interacts with and activates a large number of immunostimulatory DNA sensors such as cyclic guanosine monophosphate–adenosine monophosphate synthase (cGAS) that can trigger autoimmunity, including diseases caused by the type I interferon (IFN) response (1). Mitochondrial outer membrane permeabilization (MOMP) is required for mtDNA release. To date, only BAX/BAK oligomers, which can form extremely large macropores in the mitochondrial outer membrane (MOM), have been shown to mediate mtDNA release (24). However, the formation of the BAX/BAK macropore generally occurs under conditions that activate BAX/BAK, such as apoptosis or treatment with BAX/BAK activators (24). However, the pores that promote MOMP in live cells or in conditions that do not activate BAX/BAK have not been identified.

Because the voltage-dependent anion channel (VDAC) can oligomerize under oxidative stress conditions, and because VDAC oligomers can form large MOM pores (5), we investigated whether VDAC could trigger MOMP in live cells and mediate mtDNA release. VDAC is the most abundant protein in MOM and regulates Ca2+ influx, metabolism, inflammasome activation (6), and cell death (7, 8). Moreover, the expression of both VDAC1 (the most abundant of the three VDAC isoforms) and VDAC3 is increased in the autoimmune disease systemic lupus erythematosus (SLE) (9). In some models of this disease, the trigger is thought to be the release of mtDNA (10, 11).

Previous studies on the mechanism of mtDNA release have used cells that are undergoing apoptosis or have altered mtDNA content (2, 12, 13), hence the results of such studies are difficult to interpret. To avoid these confounding variables, we studied mouse embryo fibroblasts (MEFs) deficient in endonuclease g (Endog), a nuclear-encoded mitochondrial endonuclease (14, 15). Endog–/– MEFs have higher levels of cytosolic mtDNA (cmtDNA) relative to wild-type MEFs (Fig. 1A), despite having similar levels of total mtDNA (Fig. 1B) and cellular growth rates (fig. S1A). Consistent with this finding, Endog–/– MEFs (Fig. 1, C and D, and fig. S1, B and C) and plasmacytoid dendritic cells (fig. S1D) expressed higher levels of IFN-stimulated gene (ISG) mRNA relative to their wild-type counterparts. ISG expression and phosphorylation of TANK-binding kinase 1 (TBK1) and IFN regulatory factor 3 (IRF3) were significantly reduced in ρ0 (mtDNA-deficient) cells derived from Endog–/– MEFs, relative to cells derived from the parental Endog–/– MEFs (Fig. 1, E and F, and fig. S1, E to G). Thus, the cGAS-STING pathway is activated by cmtDNA in Endog–/– MEFs (fig. S2, A to I). Increased mROS appeared to cause mtDNA release in Endog–/– MEFs because mROS was higher in Endog–/– MEFs (fig. S2, J and K). By contrast, the mROS scavenger mito-TEMPO decreased ISG expression in Endog–/– MEFs (Fig. 1G).

Fig. 1 Endog deficiency increases cmtDNA and type Ι IFN signaling.

(A and B) Quantification of cmtDNA (A) and total mtDNA (B) in wild-type (WT) and Endog–/– MEFs. (C and D) Expression levels of ISG, including Usp18 (ubiquitin-specific peptidase 18), Isg15 (interferon-stimulated gene 15), Ifit1 and Ifit3 (interferon-induced protein with tetratricopeptide repeats 1 and 3), Cxcl10 (c-x-c motif chemokine 10), and Ifi44 (interferon-induced protein 44) (C) and heat map analysis of RNA sequencing data (D) in WT and Endog–/– MEFs. (E and F) ISG expression levels in WT and Endog–/– MEFs as well as two independently generated ρ0 MEFs (ρ0 1 and ρ0 2) were determined by immunoblotting (E) and reverse transcription quantitative polymerase chain reaction (RT-qPCR) (F). (G) ISG expression was measured in WT and Endog–/– MEFs after treatment with 10 μM Mito-TEMPO (M-TEMPO) for 48 hours. Data are means ± SEM of at least three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.005 [two-tailed unpaired Student t test in (A) to (C); one-way analysis of variance (ANOVA) with Tukey post hoc test for multiple comparisons in (F) and (G)]; nd, not detected; ns, not significant.

Under physiological conditions, tissue-culture cells comprise two subpopulations: live cells, which make up the majority of the population, and a small minority of cells undergoing spontaneous apoptosis and caspase activation. Prior studies (3, 4, 16) and our data (fig. S3A) demonstrate that BAX/BAK activation with ABT-737 in the presence of caspase inhibition increases ISG expression. However, wild-type and Bax/Bak–/– MEFs had similar cmtDNA levels (fig. S3B), and knocking down Endog robustly induced ISG expression in Bax/Bak–/– MEFs, albeit at levels slightly lower than were observed in wild-type MEFs (fig. S3, C and D). Thus, Endog–/– MOMP can occur in the absence of BAX/BAK macropores. Moreover, mitochondria in Endog–/– MEFs tended to be slightly longer than those in wild-type MEFs rather than being fragmented, as would be expected with BAX/BAK activation (fig. S3E) (2, 17).

We next compared the apoptosis levels in wild-type and Endog–/– MEFs because extremely high levels of mROS can lead to apoptosis, and Endog–/– MEFs have higher mROS levels than wild-type MEFs (fig. S2, J and K). Wild-type and Endog–/– MEFs had similar levels of apoptotic indicators such as caspase activity (fig. S3F), cell viability (fig. S3G), lactate dehydrogenase (LDH) release (fig. S3H), and ethidium homodimer–1 (EthD-1) staining (fig. S3I) either before or after the induction of apoptosis by BAX/BAK. Thus, Endog–/– MEFs have levels of mROS stress high enough to promote MOMP and mtDNA release in a BAX/BAK-independent manner but insufficient to promote apoptosis at the cellular level (fig. S3I).

An alternative mediator of MOMP may be VDAC, and indeed, Vdac1–/–, Vdac3–/–, and Vdac1/3–/– MEFs had lower levels of ISG mRNA and cmtDNA relative to wild-type MEFs (Fig. 2, A and B, and fig. S4, A to C) despite having similar total mtDNA levels (fig. S4D). We excluded VDAC2 from our study because VDAC2 deficiency promotes apoptosis (8). Additionally, knockdown of Endog increased ISG expression in wild-type and Bax/Bak–/– MEFs but not in Vdac1/3–/– MEFs (Fig. 2C and fig. S3C), and treatment with the VDAC inhibitor DIDS (18) decreased ISG expression in Endog–/– MEFs (fig. S4E). In agreement with their reduced type I IFN signaling, Vdac1/3–/– MEFs were less resistant to HSV-1 infection than wild-type MEFs (fig. S4, F to H). Thus, both VDAC1 and VDAC3 contribute to both MOMP and mtDNA release.

Fig. 2 VDAC oligomerization is required for mtDNA fragment release.

(A) Expression levels of Ifna4 and Ifnb (interferon alpha-4 and beta) and expression levels of ISG, including Isg15, Ifit1, Ifi44, and Iigp1 (interferon-inducible GTPase 1), in WT and Vdac1/3–/– MEFs. (B) cmtDNA levels were determined in WT and Vdac1/3–/– MEFs after treatment with 100 μM H2O2 for 18 hours. (C) ISG expression levels were measured in WT and Vdac1/3–/– MEFs after knocking down Endog. (D and E) cmtDNA (D) and expression levels of ISG, including Isg15, Ifit1, Ifi44, Iigp1, and Oasl2 (2′,5′-oligoadenylate synthetase–like 2) (E), which were measured after treatment with 10 μM VBIT-4 in Endog–/– MEFs. (F) VDAC1 oligomerization–dependent release of mtDNA from mtDNA-loaded liposomes and inhibition by VBIT-4. (G) Fragment size distribution of the fimtDNA and cmtDNA. Data are means ± SEM of at least three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.005 [two-tailed unpaired Student t test in (A), (D), (E), and (F); one-way ANOVA with Tukey post hoc test for multiple comparisons in (B) and (C)].

VDAC has been implicated in apoptosis induced by certain stimuli (7). However, Vdac1/3–/– and wild-type MEFs had similar numbers of cells undergoing apoptosis. This was true both in the basal state and in the presence of H2O2 (fig. S5A), which increases cmtDNA (Fig. 2B). In contrast, ABT-737 induced apoptosis in Vdac1/3–/– MEFs but not in Bax/Bak–/– MEFs (fig. S5B). Thus, VDAC is not required for apoptosis, and presumably MOMP, induced by BAX/BAK activation (fig. S5C).

We initially hypothesized that VDAC regulates mtDNA release by promoting Ca2+ influx (19), which opens the mitochondrial permeability transition pore (mPTP) in the mitochondrial inner membrane (MIM) in a manner that does not trigger apoptosis at the whole-cell level (20, 21). This would be possible only if the population of affected mitochondria is very small so that the total caspase activity remains near background levels. Chelation of Ca2+ with BAPTA decreased ISG expression (fig. S6, A to C), and inhibiting mPTP opening with cyclosporin A (CsA) decreased mtDNA release in both Endog–/– MEFs and wild-type mitoplasts as well as ISG mRNA levels in Endog–/– MEFs (fig. S6, D to F). Mitochondria deficient in MICU1 (mitochondrial calcium uptake 1) had elevated mROS levels and IFN responses (fig. S6, G to I) (22). However, mtDNA release from these mitochondria was still inhibited by DIDS in the absence of Ca2+ (fig. S6J). Thus, VDAC may play a Ca2+ flux–independent role in mtDNA release. In agreement with this, the highly potent VDAC1 oligomerization inhibitor VBIT-4 (23) decreased cmtDNA levels and ISG expression in Endog–/– MEFs (Fig. 2, D and E) without inhibiting either Ca2+ uptake (fig. S6K) or mPTP opening (fig. S6L) in the mitochondria. We then loaded mtDNA fragments into liposomes with or without membrane-reconstituted VDAC1. The presence of VDAC1 in liposomes increased mtDNA passage across the lipid membrane (Fig. 2F). However, VBIT-4 decreased VDAC1 oligomerization (fig. S7, A and B) as well as mtDNA efflux (Fig. 2F). Thus, VDAC1 oligomers are sufficient to permeabilize lipid membranes for mtDNA passage.

Because the intact mtDNA is large (16 to 17 kb) and is tethered to the MIM in nucleoid complexes (24), we hypothesized that short and free (untethered) intra-mtDNA fragments (fimtDNA) that can pass through VDAC oligomer pores preexist in live cells (fig. S8A). To investigate this possibility, we treated mitochondria purified from wild-type MEFs with cytoskeleton (CSK) buffer, which gently permeabilizes mitochondrial membranes and releases fimtDNA while leaving mitochondrial nucleoids intact (24). Interestingly, the sequences corresponding to a region within the D-loop in the mitochondrial genome were overrepresented in the fimtDNA pool (fig. S8, A and B). A size-distribution analysis excluding the sequences with 100% homology to both mitochondrial and nuclear genomes indicated that the peak sizes of fimtDNA and cmtDNA are almost identical (~110 bp) (Fig. 2G). Treatment of wild-type MEFs with mito-TEMPO (fig. S8C) or the mTORC1 inhibitor everolimus (fig. S8D), which promotes mitophagy and elimination of damaged mitochondria, decreased fimtDNA. Thus, fimtDNA accumulates preferentially within a subpopulation of mitochondria with elevated mROS, and cmtDNA is derived largely from fimtDNA in live cells (figs. S5C and S8A).

Studies with planar lipid bilayers reconstituted with VDAC1 indicate that the N-terminal domain of VDAC1 can interact directly with mtDNA (fig. S9). The N-terminal domain, which is evolutionarily conserved (fig. S10A), is hydrophilic and is thought to translocate out of the channel when VDAC1 is in an oligomerized state (Fig. 3A) (25). This raises the possibility that the negatively charged backbone of mtDNA may interact with multiple VDAC1 molecules simultaneously and may act as a scaffold to stabilize the oligomers (Fig. 3A). Indeed, mtDNA increased the formation of VDAC1 trimers and higher-order oligomers (Fig. 3, B and C) in vitro. In agreement with increased cmtDNA in Endog–/– MEFs (Fig. 1A), coimmunoprecipitation with antibody to VDAC1 pulled down more mtDNA in Endog–/– MEFs than in wild-type MEFs (Fig. 3D). VDAC1 oligomerization was also increased in Endog–/– MEFs relative to wild-type MEFs, but was reduced by the elimination of mtDNA in Endog–/– MEFs (ρ0) (Fig. 3, E and F). By contrast, BAK oligomerization was not increased in Endog–/– MEFs (Fig. 3G), and BAI, the BAX oligomerization inhibitor (26), did not reduce ISG expression in Endog–/– MEFs (Fig. 3H). Thus, mtDNA may promote further VDAC1 oligomerization, creating a feedforward cycle.

Fig. 3 mtDNA interacts with VDAC1 and stabilizes its oligomeric state.

(A) Schematic diagram of VDAC1 oligomerization accompanied by the N-terminal domain (red) translocation into the large oligomer pore. We could not characterize VDAC3 in vitro because it tends to form aggregates. (B and C) mtDNA-induced oligomerization of purified VDAC1 was visualized by immunoblotting after treatment with the cross-linking reagent EGS to stabilize the oligomers during electrophoresis (B). Quantitative analysis of oligomers is shown (C). (D) mtDNA binding to VDAC1 in WT and Endog–/– MEFs. (E to G) VDAC1 (E) and BAK (G) oligomerization in WT, Endog–/– and Endog–/– ρ0 MEFs was visualized by immunoblotting. The positions of VDAC1 monomers (Mono), dimers (Di), trimers (Tri) and multimers (Multi) are indicated. NSB, nonspecific band. Quantitative analysis of VDAC1 oligomers is shown (F). (H) ISG expression was measured in WT and Endog–/– MEFs after treatment with the BAX oligomerization inhibitor (BAI) (2 μM) for 24 hours. (I) The amino acid sequence of the VDAC1 N-terminal peptide (abbreviations: A, Ala; D, Asp; F, Phe; G, Gly; K, Lys; L, Leu; M, Met; P, Pro; R, Arg; S, Ser; T, Thr; V, Val; Y, Tyr.). The positively charged amino acids were mutated to alanine (red). (J) Direct interaction of mtDNA fragments with WT and 3A N-terminal 26 amino acid peptides. (K) Expression levels of ISG, including Isg15, Ifit1, Ifit3, Ifi44, Oasl2, and Rsad2 (radical S-adenosyl methionine domain 2), were measured by RT-qPCR after treatment with 100 μM H2O2 for 18 hours in MEFs expressing either WT or 3A VDAC1. Data are means ± SEM of at least three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.005 [two-tailed unpaired Student t test in (C), (D), and (K); one-way ANOVA with Tukey post hoc test for multiple comparisons in (F), (H), and (J)].

The N-terminal domain of VDAC1 contains three positively charged residues (Lys12, Arg15, and Lys20) that could potentially interact with the negatively charged backbone of mtDNA (Fig. 3I). Indeed, mtDNA fragments were pulled down by a 26–amino acid VDAC1 N-terminal peptide but not by a VDAC1 N-terminal mutant peptide (3A) in which Lys12, Arg15, and Lys20 residues were replaced by alanine (Fig. 3J). We then examined H2O2-induced expression of ISG in Vdac1/3–/– MEFs with restored expression of wild-type VDAC1, the 3A-mutant VDAC1, or ΔN-terminal VDAC1. Relative to MEFs expressing wild-type VDAC1, ISG expression was reduced in MEFs expressing either the 3A-mutant VDAC1 (Fig. 3K) or ΔN-terminal VDAC1 (fig. S10, B and C). Thus, direct mtDNA-VDAC interactions appear to promote VDAC oligomerization and increase mtDNA release.

An excessive type Ι IFN response is a hallmark of SLE (27). Gene Expression Omnibus (GEO) analysis revealed increased mRNA expression of VDAC1/3 and decreased expression of ENDOG in SLE patients (fig. S11A) (9). However, BAK, BAX, VDAC2, and HSP60 mRNA levels were not changed in SLE patients (fig. S11A). These findings, combined with the observation that type Ι IFN responses in Endog-knockdown MEFs were VDAC1- and VDAC3-dependent (Fig. 2C), suggest that VDAC oligomerization may be associated with SLE. Indeed, splenocytes from MpJ-Faslpr “lupus”-prone mice had more VDAC1 oligomers than did MpJ control mice (Fig. 4A and fig. S11B); the same was true for peripheral blood mononuclear cells (PBMCs) from SLE patients relative to healthy controls (Fig. 4B and fig. S11C). Splenocytes from MpJ-Faslpr mice also had elevated cmtDNA relative to those from MpJ mice, but this was abrogated with VBIT-4 treatment (Fig. 4C).

Fig. 4 VDAC1 oligomerization inhibitor VBIT-4 ameliorates lupus-like disease.

(A) The formation of VDAC1 oligomers in splenocytes of MRL/MpJ-Faslpr lupus-prone mice and MRL/MpJ control mice (see fig. S11B; n = 6 in each group). (B) Oligomeric state of VDAC1 in PBMCs of healthy control and SLE patients (see fig. S11C; n = 6 in each group). (C) cmtDNA levels were measured in splenocytes [see (A)] after treatment with 10 μM VBIT-4. (D) Kidney glomeruli of VBIT-4–treated mice stained with antibodies to complement C3 (green) and IgG (red). Nuclei were stained with Hoechst (blue). Scale bars, 20 μm. (E) Fluorescence intensity of C3 and IgG in (D) (n = 8 in each group). (F to I) Urinary albumin/creatinine ratio (F), serum anti-dsDNA levels (G), ANA levels (H), and IgG levels (I) of VBIT-4–treated mice (n = 10 in each group). (J) Serum cell-free mtDNA levels of VBIT-4–treated mice (n = 5 in each group). (K) A23187 (Ca2+ ionophore)–induced mROS levels were measured by MitoSOX in PBMCs of healthy controls (HC) and SLE patients (n = 3 in each group). (L) A23187-induced NET formation by NDGs from HC or SLE subjects was measured by SYTOX-Pico Green plate assay (n = 3 in each group). Data are means ± SEM. *P < 0.05, **P < 0.01, ***P < 0.005 [Student t test in (A), (B), (E), (F) to (J), and (L); one-way ANOVA with Tukey post hoc test for multiple comparisons in (C); Mann-Whitney U test in (K)].

We next investigated whether VBIT-4 could ameliorate lupus-like symptoms in MpJ-Faslpr mice. VBIT-4 blocked the development of skin lesions and the thickening of the epidermis that accompanies leukocyte infiltration, and suppressed facial and dorsal alopecia without affecting mortality or body weight (fig. S12, A to C). VBIT-4 also decreased spleen and lymph node weights (fig. S12D). Furthermore, ISG mRNA (fig. S12E), renal immune complex deposition (Fig. 4, D and E), proteinuria (Fig. 4F), antibody to double-stranded DNA (dsDNA) (Fig. 4G), antinuclear antibody (ANA) (Fig. 4H), IgG (Fig. 4I), and cell-free mtDNA (Fig. 4J) levels were all reduced by VBIT-4. Cell-free mtDNA plays an immunostimulatory role in human and mouse SLE (10, 11). One potential source of cell-free mtDNA in MpJ-Faslpr mice may be neutrophil extracellular traps (NETs), which are formed in a cell death process termed NETosis (11). mROS is an important trigger for NETosis (11), and VBIT-4 decreased mROS in neutrophils as well as other immune cells from SLE patients (Fig. 4K and fig. S12F). Furthermore, VBIT-4 suppressed NETosis in low-density granulocytes (LDGs), a distinct class of pro-inflammatory and NETosis-prone neutrophils from SLE patients, and normal-density granulocytes (NDGs) isolated from SLE patients and healthy controls (Fig. 4L and fig. S12G). Thus, VDAC oligomerization increases mROS and NETosis, two proposed important triggers of autoimmunity, in both human neutrophils (healthy and SLE) and during lupus-like disease in mice (fig. S12H).

We propose that the MOMP that mediates mtDNA release and type I IFN response depends on the level of mitochondrial stress. In live cells, VDAC1 oligomer pores, and possibly VDAC3 oligomer pores, play a role in moderate stress responses, whereas BAX/BAK macropores feature in extreme stress and/or apoptosis (fig. S13). The small fragment size and the untethered nature of fimtDNA may facilitate its release via VDAC oligomer pores. Notably, there are other pathways of mtDNA release (10, 13) and SLE is a very heterogeneous disease. Nonetheless, inhibiting VDAC oligomerization may be an alternative therapeutic approach for a wide range of diseases, like SLE and Parkinson’s disease (28), that are thought to be associated with mtDNA release.

Supplementary Materials

Materials and Methods

Figs. S1 to S13

Table S1

References (2936)

References and Notes

Acknowledgments: We thank the NHLBI core facilities, including DNA Sequencing and Genomics core, Bioinformatics and Computational Biology core, Flow Cytometry core, Light Microscopy core, Pathology core, and Biochemistry core. We thank the NIH Fellows Editorial Board for manuscript preparation. Funding: Supported by the Intramural Research Program of the National Heart Lung and Blood Institute, National Institute of Arthritis and Musculoskeletal and Skin Diseases, and National Institute of Allergy and Infectious Diseases and by a grant from the National Institute for Biotechnology in the Negev (V.S.-B.) and by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health and Welfare, Republic of Korea (grant number HI14C1176) and a grant from the KRIBB Research Initiative Program (Korean Biomedical Scientist Fellowship Program), Korea Research Institute of Bioscience and Biotechnology, Republic of Korea. Author contributions: J.K. designed and performed the majority of experiments, analyzed the data, interpreted results, and wrote the manuscript; R.G., A.S.-K., and V.S.-B. performed the cross-linking assay, liposome assay, MST, mPTP, VDAC purification, and channel conductance studies and helped write the manuscript; L.P.B., X.W., and M.J.K. performed and analyzed the data for the human sample studies, including NETosis and mROS measurements; K.W. and J.I.C. designed and performed the viral infection study; J.Z. supervised mtDNA sequence analysis; S.Y. and J. Z. performed sequencing and analyzed the data; H.E.Y., M.K., H.K., A.L.B., S.-J.P., X.X., and E.Z.v.R. helped with experiments; M.K.K. and J.H.C. supervised the study, analyzed the data, and wrote the manuscript. Competing interests: None declared. Data and materials availability: All other data and materials are available from the corresponding author upon request. Gene Expression Omnibus (GEO) database of human SLE patient is supported by National Center for Biotechnology Information (GEO accession number: GSE13887, = GSE13887).

Stay Connected to Science

Navigate This Article