Enhanced Immunogenicity of Mitochondrial-Localized Proteins in Cancer Cells

T-cell responses against human cancers contribute to the control of tumor growth, and targeting of CTLA-4 and the PD-1/PD-L1 axis has been very effective in enhancing antitumor immune responses, resulting in clinical objective responses (1, 2), particularly in patients with tumors expressing high mutational burden (TMB; refs. 1, 2). These results underscore an unmet clinical need for many cancer patients with low TMB (i.e., the largest proportion of cancer patients), who could receive greater benefit from immune-checkpoint inhibition treatment, should optimal neoepitopes be identified and used in vaccination strategies.

To this end, new strategies need to be developed to identify the most immunogenic cancer-associated neoepitopes and optimal vaccine platforms to improve immune responses to such predicted epitopes. Current pipelines used for neoantigen prediction have resulted in a low rate of validation, suggesting that the determinants of peptide immunogenicity are still suboptimal (3–5). A greater understanding of the biology of the presentation of the cancer mutanome is thus needed in order to improve such algorithms. Several parameters are currently considered when ranking the immunogenicity of mutations in cancer cells (6–8), but it remains unclear whether the subcellular localization of tumor antigens can modulate the efficiency of priming of antitumor-specific immune responses and whether such parameter should be considered in algorithms ranking immunogenicity of mutated peptides in cancer cells. This represents a critical knowledge gap, as current prognostic scores for responsiveness to immune-checkpoint inhibitors are based mainly on the numbers of mutations, without taking into account whether the subcellular localization of such mutated protein antigens may influence their ability to stimulate an immune response.

Tumor cells fail to directly prime specific immune responses, likely as a result of low costimulation; instead, DCs function simultaneously as both antigen-presenting cells, capable of taking up tumor debris, and IL12-producing cells, in a process referred to as “crosspriming” (9, 10). Evidence that DCs could cross-prime tumor-specific T-cell responses includes the transfer of cellular components from tumor cells to antigen-presenting cells (9, 11). The transfer of mitochondria from tumors to DCs via cytoplasts (11) suggests that mutations in mitochondrial-localized proteins in cancer cells could elicit specific T-cell responses in vivo. Consistent with this possibility, activation of the cGAS–STING pathway in DCs may be driven by tumor mitochondrial DNA, resulting in the induction of a type I IFN response (12). Together, these results suggest that tumor-derived mitochondria and mitochondrial-associated antigens play a role in the generation of tumor-specific immune responses.

Previous studies have described the effect of subcellular localization of protein antigens on direct presentation of T-cell epitopes (13–16). Yamazaki and colleagues have demonstrated that cytoplasmic versus mitochondrial localization of antigenic protein LIVAT-BP determines which portions of the protein are selected as antigenic epitopes for CD8⁺ T-cell recognition (16). Although these results are of interest, this paper falls short of distinguishing whether antigen location modulates: (i) in vivo cross-priming of T-cell responses; (ii) proteasome-dependent degradation of the antigenic protein; and (iii) tumor growth.

Here, we have investigated both in vitro and in vivo the impact of the mitochondrial location of antigenic proteins on direct- and cross-priming, as compared with the immunogencity of the same antigenic proteins expressed in the cytosol. We have extended results obtained in animal models to clinical samples by demonstrating the presence of CD8⁺ T-cell responses specific to mitochondrial-localized neoantigens in a patient with endometrial cancer. Our results demonstrate that mitochondrial-localized proteins induce greater immune responses than cytosol-localized proteins and provide a rationale for including protein localization data to improve prediction algorithms for neoepitopes that are effective in priming immune responses.

STING knockout (17), cGAS knockout (18), MAVS knockout (19), Myd88 knockout, MARCO/SRA knockout, β₂-microglobulin knockout, human leukocyte antigen (HLA)-A2.1 transgenic mice (HHD mice; ref. 20), and C57BL/6 control mice were bred in the local animal facility under specific pathogen–free conditions and used at 6 to 10 weeks of age. Mice were injected subcutaneously (s.c.) with 1 × 10⁶ or 10⁵ cells/mouse for immunogenicity or tumor growth experiments, respectively. Animal studies have been conducted in accordance with, and with the approval of, the United Kingdom Home Office. All procedures were done under the authority of the appropriate personal and project licenses issued by the United Kingdom Home Office License number PBA43A2E4.

A cDNA fragment containing sequences encoding amino acids 1–22 of the human arginase-2 protein followed by OVA amino acids 47–386 (or NY-ESO-1) and the HA tag (YPYDVPDYA) was synthetically made by Thermo Fisher Scientific. This was inserted into plasmid pHR-SIN-MCS-ires-eGFP (5) to create pLenti-mtOva or pLenti mtNY-ESO-1. Using PCR, OVA_47-386HA (or NY-ESO-1-HA) lacking the mitochondrial directing sequence were generated to create pLenti-cytoOVA or pLenti cytoNY-ESO-1 (in pHR-SIN-MCS-ires-eGFP). Using PCR, additional primers allowed us to add other sequences to the N-terminus of OVA_47-386. One such construct, although initially intended to direct OVA to the nucleus (by addition of an N-terminal MPKKKRVGG sequence), failed to do so and instead was shown experimentally, to produce a cytoplasmically located, yet more stable OVA protein. B16 cells transfected with this construct were named B16 mod cytoOVA. A fusion with an N-terminal mCherry protein was generated by inserting the mitochondria targeted OVA_47-386 gene into the plasmid pHR-SIN mCherry (21). The plasmids described above were used to produce lentiviral particles that were then used to transduce CT26 cells, B16F10 cells, and Lewis lung carcinoma cells (LLC). Cells were used between passages 4 and 8 and routinely tested for Mycoplasma contamination. The LLC cell line has been received from the ATCC in 2019, B16F10 was kindly provided by Dr. Michael Palmowsky (University of Oxford, NDM) in 2002, and CT26 from Dr. Jonathan Silk (Adaptimmune) in 2012. The cell lines were not authenticated in the past year.

Whole-cell lysates or subcellular fractions were boiled in sample buffer, separated using SDS/PAGE, transferred to PVDF membranes (Bio-Rad), and blocked with 5% (weight/vol) skimmed milk in 0.5% Tween 20 in TBS for 1 hour at room temperature. Membranes were probed using anti OVA, anti-GAPDH, anti-TOM20, anti-NY-ESO-1, anti-ATPβ, and HRP-conjugated rat anti-mouse or donkey anti-rabbit according to the primary antibody used. Antibody clones are reported in Supplementary Table S1. HRP reaction was developed using Supersignal West Pico kit (Thermo Fisher Scientific).

Cells were stained with Mitotracker Red CMX-Ros (50 nmol/L 15 minutes at 37°C; Thermo Fisher Scientific) and fixed with 3% formaldehyde in PBS (Thermo Fisher Scientific), permeabilized for 15 minutes with PBS Triton-X (Sigma-Aldrich) 1%, and blocked with 2% serum bovine albumin (BSA; Sigma-Aldrich) + 5% fetal calf serum (FCS, Sigma-Aldrich) in PBS for 1 hour at room temperature. Samples were incubated with primary antibodies in blocking buffer for 1 hour at room temperature. After several washing steps, the cells were incubated for 30 minutes with secondary antibodies conjugated to Abberior STAR 600 and/or Abberior STAR Red (Abberior Instruments) diluted 1:250 in 1% BSA in PBS. After several washing steps, the slides were mounted on a drop of Mowiol (Sigma-Aldrich). A pixel-wise Pearson colocalization test to quantify the colocalization of mitotracker and NY-ESO-1 was performed (values of 1 indicate complete colocalization, 0 indicates no colocalization).

B16 cytoOVA, B16 mtOVA, and B16 wild-type (WT) cells were incubated with IFNγ, TNFα, or both (5 ng/mL; BioLegend) for 2 or 5 days and then stained with the H-2K^b-OVA_257-264–specific antibody (22). Stimulated cells were cocultured with OT-I CD8⁺ T cells (ratio 1:40) overnight, and OT-I cells were stained for extracellular markers (CD8α, CD69, and CD25) and viability (antibody clones are reported in Supplementary Table S2). C57BL/6 mice were injected with B16 mtOVA or B16 cytoOVA and B16 WT. Tumors were collected, pierced, and minced in 24-well plates and incubated for 15 minutes at 37°C in 1 mL of digestion buffer (2 mg/mL collagenase D, and 1 mg/mL DNAse I in RPMI-1640, both from Sigma-Aldrich). After the first 15 minutes of incubation, cells were pipetted up and down repeatedly, then returned for a second 15-minute incubation at 37°C. After digestion, B16 cells were stained with H-2K^b-OVA_257-264– and CD105-specific antibodies. Samples were acquired on a FACScanto II (BD Biosciences) flow cytometer, and data were analyzed with FlowJo version 10.4.1. Compensation beads (eBioscience) were used to generate the compensation matrix, and fluorescence minus one (FMO) were used as control.

Splenocytes (2 × 10⁶) were isolated from either naïve or tumor-bearing C57BL/6 mice 7 days after the injection and were cultured in the presence of OVA_257-264 peptide (SIINFEKL; Sigma-Aldrich) or NY-ESO-1_157–165 peptide (SLLMWITQC, Sigma-Aldrich) peptide in complete RPMI-1640 medium supplemented with 10% FBS, 2.1 mmol/L ultra-glutamine in the presence of Brefeldin A (5 μg/ml) and monensin (2 μM; BioLegend). After 5 hours of incubation, cells were stained for extracellular markers (CD8b, B220, CD44, CD69, and CD25) and viability (antibody clones are reported in Supplementary Table S2). Cells were then stained for intracellular IFNγ and TNFα using an Intracellular Fixation and Permeabilization Buffer Set (eBioscience) according to the manufacturer's instructions. Samples were acquired on a FACScanto II (BD Biosciences) flow cytometer, and data were analyzed with FlowJo version 10.4.1. Compensation beads (eBioscience) were used to generate the compensation matrix, and FMOs were used as control.

Mice were injected on days 0 and 2 with 400 μg/mouse of CD8 antibody intraperitoneally (anti-CD8 clone 2.43 or isotype control, both from Invivomab). On day 2, mice were injected with the B16 tumor cells s.c. (1.5 × 10⁵ cells/mouse) and were then injected every 6 days with 200 μg/mouse of CD8 antibody intraperitoneally. To confirm the depletion of CD8⁺ T cells, blood samples were collected from the tail on days 6 and 10 in tubes with heparin. Red blood cells were lysed with red blood cells lysis buffer (Qiagen) following the manufacturer's instruction. After washing, cells were stained for extracellular markers (CD8b, CD3, and B220) and viability (antibody clones are reported in Supplementary Table S2). The same staining was performed on splenocytes on day 14, after which the mice were killed. Samples were acquired on a FACScanto II (BD Biosciences) flow cytometer, and data were analyzed with FlowJo version 10.4.1. Compensation beads (eBioscience) were used to generate the compensation matrix and FMOs were used as control.

Mitochondria from B16 tumor cells or from mouse adult fibroblasts were isolated as previously described (23). Mice were injected with 50 μg of the mitochondrial protein samples. On day 10 after vaccination, mice were challenged with B16 or LLC tumor cell lines and the tumor growth was monitored. Where indicated, CD8⁺ T cells were depleted before tumor injection. CD8⁺ T-cell responses were evaluated 14 days after tumor injection: CD8⁺ T cells were isolated from mice spleens with a CD8 isolation kit (Miltenyi Biotech) and cocultured overnight with B16 cells that had been previously stimulated 48 hours with IFNγ (10 ng/mL). Brefeldin A and monensin were added for the last 5 hours of coculture. Cells were then stained for extracellular markers (CD8b, B220, and CD44) and viability (antibody clones are reported in Supplementary Table S2). Cells were then stained for intracellular IFNγ using an Intracellular Fixation and Permeabilization Buffer Set (eBioscience) according to the manufacturer's instructions. Samples were acquired on a FACScanto II (BD Biosciences) flow cytometer, and data were analyzed with FlowJo version 10.4.1. Compensation beads (eBioscience) were used to generate the compensation matrix and FMOs were used as control.

The somatic mutation data were downloaded from The Cancer Genome Atlas (TCGA) Genomic Data Commons data portal (https://portal.gdc.cancer.gov). The somatic mutation data (version 02-04-2018) included 9,508 samples from 31 cancer types (Supplementary Table S3). The somatic mutations were generated by MuTect2 (24) with the GRCh38 reference genome. To calculate the number of nonsilent somatic mutations locating in mitochondrial genes, we kept the somatic mutations that passed the filter of the mutation calling, located in the genes that belong to Gene Ontology GO:0005739 (cellular component: mitochondrion) and were classified as nonsilent mutations. We used the in-house R script to filter mutations and perform visualization. The R code is available at https://github.com/zhiyhu/mito-mut-pancan.

Endometrial tumor tissue and blood collection were obtained from a patient recruited to the Gynecological Oncology Targeted Therapy Study 01 (GO-Target-01) under research ethics approval number 11-SC-0014, good clinical practice based on the Declaration of Helsinki were used. The patient gave informed consent. Whole-genome sequencing was performed on blood and tumor tissues (BGI Tech Solutions Ltd) as previously described (25). RNA was extracted using the Qiagen RNeasy Mini Kit, and its quality was assessed with the Agilent TapeStation before preparing the sequencing libraries. Two technical replicates were prepared from 400 ng RNA, each using the NEBNext Ultra Directional RNA Library Prep Kit for Illumina (E7420) in combination with the NEBNext Poly(A) mRNA Magnetic Isolation Module (E7490) and NEBNext Q5 Hot Start HiFi PCR Master Mix (M0543). The libraries were indexed and enriched by 14 cycles of amplification, assessed using the Agilent TapeStation and then quantified by Qubit. Multiplexed library pools were quantified with the KAPA Library Quantification Kit (KK4835) and sequenced using 80 bp PE reads on the Illumina NextSeq500 platform. Tumor-specific nonsynonymous mutations were predicted and ranked as previously shown (26). Peptides were synthesized by Pepscan Presto BV.

One hundred twenty peptides were pooled in 6 groups of 20 peptides each. In vitro stimulation of CD8 T cells was done as previously described (27). Briefly, 5 × 10⁶ to 8 × 10⁶ peripheral blood mononuclear cells (PBMC) from the patient were stimulated with each peptide pool at a final concentration of 20 μg/mL (2 μg/mL, each peptide) for 3 days in RH10 [RPMI with 10% heat-inactivated human serum (Sigma), 2 mmol/L L-glutamine, 10 mmol/L HEPES, 1 mmol/L sodium pyruvate, nonessential amino acids (1×), penicillin (25,000 U), streptomycin (25 mg), 50 μmol/L β-mercaptoethanol (Gibco)] supplemented with 25 ng/mL human IL7 (PeproTech). On day 3, half of the media were replaced with RH10-IL2 (RH10 supplemented with 1,000 IU human IL2; Novartis). When confluent, cells were split using RH10-IL2. To further increase cell expansion, 4 weeks later, cells were restimulated with phytohaemagglutinin 1 μg/mL in RH10-IL2 and in the presence of irradiated PBMC feeders and rest for 3 to 4 weeks before screening.

Generation of peptide–MHC class I monomers and tetramerization was performed as previously described (28). Briefly, biotin-tagged HLA-A2 complexes were folded with the UV-sensitive peptide KILGFVFJV. Two micrograms of HLA-A2 complexes was UV exchanged for 1 hour with each screening peptide at a final concentration of 200 μg/mL, in 20 μL. After centrifugation at 2,250 × g, 1.5 μg of complexes (15 μL supernatant) was collected and tetramerized with 1.5 μL of a 1:1 mix streptavidin-APC/streptavidin-PE (eBioscience). Free biotin in the complexes was blocked by adding 20 μL of 50 μmol/L D-biotin. Stimulated PBMCs (1 × 10⁵) were incubated in 50 μL staining buffer (PBS 0.5% BSA) containing 2.5 μL of multimers, for 30 minutes at 37°C. Cells were washed two times with staining buffer and stained with LIVE/DEAD Fixable Aqua (Thermo Fisher Scientific), anti-CD3 FITC (clone SK7, BioLegend), and anti-CD8 PerCP-Cy5.5 (clone SK1, BioLegend). Cells were analyzed in a BD LSRFortessa instrument. PE and APC tetramer-positive cells were sorted in a BD fusion instrument and further expanded for functional assays.

Autologous Epstein–Barr virus-transformed lymphoblastoid cell lines (EBV-LCL) were generated from PBMCs, using supernatant of EBV producing B95-8 marmoset cells and 2 μg/mL cyclosporin A (Sigma). EBV-LCLs used as target cells were loaded with 1 μmol/L peptides, for 1 hour at 37°C. Loaded and nonloaded cells were stained with either CellTrace Far Red or CellTracker Orange CMTMR dyes (Thermo Fisher Scientific) and quenched with FCS. After two wash cycles, loaded and nonloaded targets were mixed in a 1:1 ratio and plated in a 96 U-bottom well plate. Effector T cells, previously incubated overnight in the absence of IL2, were added to the wells at the indicated effector-to-target ratio, in duplicates. Following 4.30 hours of incubation at 37°C, cells were stained with LIVE/DEAD Fixable Aqua (Thermo Fisher) and anti-CD8-FITC (clone SK1, BioLegend). Cells were analyzed in a BD LSRFortessa instrument.

EBV-LCLs were loaded with 1 μmol/L peptide, for 1 hour at 37°C, and washed twice. Loaded cells were plated at 25,000 cells per well in a U-bottom 96-well plate and used to stimulate 2,500 T cells, in duplicates. Following 16 hours of incubation, the production of IFNγ was assessed by ELISA (BD Pharmingen).

To investigate whether mitochondrial proteins contained within tumor cells could be transferred to antigen-presenting cells (APC) during tumor growth in vivo, B16 cells were transduced with lentiviral vectors encoding mitochondrial-localized ovalbumin (OVA) fused with the fluorescent protein mCherry (B16 mtOVA-mCherry), which were then injected subcutaneously in mice. The results of these experiments showed that CD103⁺ and CD11b⁺ migratory DCs had preferentially taken up mCherry protein, compared with CD8α⁺ DCs, CD169⁺ macrophages, and CD11b⁺ resident DCs (Supplementary Fig. S1A). In addition, the uptake of mitochondrial-localized mtOVA-mCherry correlated with enhanced DC maturation, as shown by higher CD86 expression on mtOVA-mCherry⁺ DCs cells compared with mtOVA-mCherry⁻ DCs (Supplementary Fig. S1B and S1C).

The uptake of mitochondrial-localized ovalbumin-mCherry fusion protein by migratory DCs suggested that the phagocytosis of mitochondria by CD103⁺ DCs may elicit the cross-priming of T cells specific for antigenic mitochondrial proteins. To address this, the H-2^d tumor cell line CT26 was transduced with a lentiviral vector encoding either NY-ESO-1 or OVA proteins, which were either targeted to mitochondria (CT26 mtNY-ESO-1 or CT26 mtOVA) or localized in the cytosol (CT26 cytoNY-ESO-1 or CT26 cytoOVA) and then injected into MHC-mismatched mice to assess their ability to induce cross-priming of antigen-specific CD8⁺ T cells. Mitochondrial and cytosolic targeting was confirmed by confocal microscopy (Fig. 1A; Supplementary Fig. S2A and S2B).

We have previously shown that HLA-A2.1 transgenic mice (HHD mice) can generate HLA-A2–restricted CD8⁺ T cells upon vaccine injection (20, 29). We therefore challenged HHD mice with the mismatched H-2^d tumor cell line CT26 mtNY-ESO-1 and CT26 cytoNY-ESO-1 and compared the frequency of HLA-A2–restricted NY-ESO-1_157-165–specific T cells. Because CT26 cells do not express HLA-A2 molecules, priming of HLA-A2–restricted NY-ESO-1–specific responses is dependent on cross-presentation events controlled by HHD-mice resident HLA-A2⁺ DCs.

Injection of CT26 mtNY-ESO-1 cells into HLA-A2⁺ HHD mice resulted in a significantly higher frequency of NY-ESO-1–specific HLA-A2–restricted CD8⁺ T cells than the injection of CT26 cytoNY-ESO-1 cells. CD8⁺ T-cell responses were only marginally higher following injection of CT26 cytoNY-ESO-1 cells than in mice injected with WT CT26 cells, as measured by staining with HLA-A2/NY-ESO-1_157–165 tetramers and intracellular staining for IFNγ and TNFα (Fig. 1B; Supplementary Fig. S2C).

These results were confirmed by injecting mismatched CT26 mtOVA cells into C57BL/6 mice, which generated a significantly higher frequency of H-2K^b/SIINFEKL–specific CD8⁺ T-cell responses compared with the CT26 cytoOVA cell line, independent of the level of expression of the OVA transgene (Fig. 1C; Supplementary Fig. S2E–S2G). CT26 mtOVA was also superior to CT26 cytoOVA at inducing H-2K^b/SIINFEKL–specific CD8⁺ T-cell responses, as measured by induction of proliferation of adoptively transferred OT-I CD8⁺ T cells (Supplementary Fig. S2H).

These findings demonstrate that the cross-priming of CD8⁺ T cells specific for NY-ESO-1 and OVA proteins, in HHD and B6 mice, respectively, is enhanced by the targeting of these antigens to mitochondria.

We next analyzed the steady-state amount of NY-ESO-1 and OVA proteins and their stability in the presence or absence of proteasome inhibitors (Fig. 1D; Supplementary Fig. S2I). The steady-state amount of mitochondrial-localized NY-ESO-1 (Fig. 1D) and OVA (Supplementary Fig. S2I) proteins was preserved in the absence of proteasome inhibitors, compared with cytosolic NY-ESO-1 or OVA proteins, which appear to be degraded in the absence of proteasome inhibitors.

In conclusion, relocation of NY-ESO-1 and OVA proteins from the cytosol to mitochondria increases their stability and steady-state amount, resulting from protection from proteasome degradation, and hence, enhancing their ability to be cross-presented in vivo.

We next assessed whether the enhanced immunogenicity of mitochondrial-localized NY-ESO-1 and OVA proteins could be confirmed in a syngeneic mouse model. B16 melanoma cells were transduced with lentiviral vectors encoding either mitochondrial-targeted OVA (B16 mtOVA) or cytosolic OVA (B16 cytoOVA). In both cell lines, the expression of OVA was linked to the expression of the reporter protein GFP. We observed that cytosolic OVA was rapidly degraded by the proteasome, and that inhibition of proteasome activity with epoxomicin rescued B16 cytoOVA to amounts equivalent to that of B16 mtOVA (Fig. 2A and B). Consistent with equal protein amounts of OVA in B16 mtOVA and B16 cytoOVA cells, OVA mRNA (Supplementary Fig. S3A) and GFP expression was equivalent (Fig. 2A). B16 mtOVA and B16 cytoOVA cells were injected into C57BL/6 mice, and 7 days later, the H-2K^b/SIINFEKL–specific CD8⁺ T-cell response was measured in the spleen. Consistent with previous results using the mismatched CT26 cell lines, B16 mtOVA doubled the frequency of OVA-specific CD8⁺ T cells compared with B16 cytoOVA (Fig. 2C).

In order to assess whether the enhanced priming of the OVA response upon injection of B16 mtOVA cells was due solely to OVA's increased stability or if the location of OVA in mitochondria had an additional role accounting for its enhanced priming ability, we generated a modified cytosolic OVA (hereafter referred to as “mod cytoOVA”) with an N-terminal extension of 9 amino acids. When mod cytoOVA was expressed in B16 cells, it resulted in a significantly higher steady-state amount of OVA compared with WT cytosolic and mitochondrial OVA (Fig. 2A) and was less sensitive to proteasomal degradation (Fig. 2A). Fractionation of B16 mod cytoOVA cell lysates confirmed the enrichment of mod cytoOVA protein in the cytosolic fraction and not mitochondrial, as reference proteins for mitochondrial-localized or cytosolic localized protein TOM20 (the mitochondrial outer membrane translocase) and GAPDH (glyceraldehyde 3-phosphate dehydrogenase) were used, respectively (Fig. 2B). Consistent with the notion that the steady-state amount of antigenic proteins determines the potency of their immunogenicity in vivo (13, 30, 31), we observed that the frequency of OVA-specific CD8⁺ T-cell response following the injection of B16 mod cytoOVA was significantly greater than in B6 mice injected with B16 cytoOVA. However, despite the greater amount of steady-state OVA in B16 mod cytoOVA cells, priming of the OVA-specific CD8⁺ T-cell response in mice injected with B16 mtOVA was more efficient than in mice primed with B16 mod cytoOVA cells (Fig. 2C), suggesting that the localization of OVA into mitochondria confers a distinct priming advantage in addition to its increased steady-state amount.

It is established that the rate of degradation of antigenic proteins directly correlates with the generation of MHC class I epitopes presented on the surface of APCs (32, 33). Consistent with this notion, the reduced stability of OVA in B16 cytoOVA was associated with a greater amount of the H-2K^b/SIINFEKL epitope generated by B16 cytoOVA both in vitro (Fig. 3A) and ex vivo (Fig. 3B; Supplementary Fig. S4A and S4B), as measured by flow cytometry for H-2K^b/SIINFEKL expression (22) and by increased expression of activation markers on OT-I splenoctyes in vitro, suggesting T-cell proliferation (Fig. 3C and D). Direct presentation of the SIINFEKL epitope in vivo on the surface of B16 mtOVA cells was detectable only after 14 days from the injection (Fig. 3B). Upregulation of H-2K^b/SIINFEKL complexes on the surface of B16 cells both in vitro and in vivo was dependent on cytokine stimulation, and in particular on the combined effect of IFNγ and TNFα (Supplementary Fig. S4C). A statistically significant greater amount of the H-2K^b/SIINFEKL complex was also observed on the surface of LLC encoding cytoOVA than on LLC encoding mtOVA in vitro (Supplementary Fig. S4D). These results demonstrated that rapid proteasome-dependent degradation of cytoplasmic OVA in B16 cells was very efficient in directly presenting the H-2 K^b/SIINFEKL epitope, which resulted in greater activation of SIINFEKL-specific CD8⁺ T cells, as compared with mitochondrial-localized OVA and more stable cytosolic OVA in B16 cells.

The findings from the above in vitro experiments (Fig. 3A–D) combined with the results from the in vivo immunogenicity experiments (Fig. 2C), together with previously published studies (30–32), show an inverse relationship between efficient direct antigen presentation of tumor cells and their ability to efficiently prime an in vivo antigen-specific immune response. Because both of these processes are required to generate an efficient tumor-specific immune response capable of controlling tumor growth, we monitored the rate of B16 growth in vivo and demonstrated a slower growth of B16 mtOVA than B16 cytoOVA cells (Fig. 3E), which was dependent on the presence of CD8⁺ T cells (Fig. 3F). These findings were further supported by the observation that B16 mtOVA tumors grew unhindered in β-2 microglobulin knockout mice (Fig. 3G). As a control, we showed that both tumor cell lines had a similar in vitro growth rate (Supplementary Fig. S3B).

In conclusion, these results demonstrate that direct presentation of the H-2K^b/SIINFEKL epitope by B16 cells is more efficient in cells expressing cytoplasmic OVA than mitochondrial OVA, but the amount of the directly presented SIINFEKL peptide on the surface of B16 mtOVA in vivo is sufficient for optimal tumor control.

Previous studies have demonstrated that the efficiency of priming of tumor-specific T-cell responses is enhanced by the release of DNA from tumor cells, which acts as a natural adjuvant to activate the STING pathway and induce a type I IFN response (34, 35). We assessed the role of STING and cGAS in the enhanced immunogenicity of mtOVA B16 cells. The frequency of H-2K^b/SIINFEKL CD8 ⁺ T cells in STING knockout mice injected with B16 mtOVA cells was significantly reduced compared with WT mice (Fig. 4A), whereas no differences were observed after the injection of B16 cytoOVA (Fig. 4B). Similar results were obtained using cGAS knockout mice (Fig. 4C). As a control, vaccination of cGAS and STING knockout mice with adjuvant full-length OVA protein resulted in a similar frequency of H-2K^b/SIINFEKL CD8⁺ T cells as in WT mice (Supplementary Fig. S5A and S5B). The frequency of OVA-specific IFNγ⁺ T cells in mice lacking expression of Myd88, NLRP3, or MAVS was similar to those in WT mice injected with B16 mtOVA cells (Fig. 4D).

In order to further assess the identity of DCs responsible for the cross-priming of OVA-specific CD8⁺ T cells, we investigated the immunogenicity of mitochondrial-localized OVA in BATF3 knockout mice, in which the development of CD103⁺/CD8α cross-presenting DCs is compromised (36). Consistent with our earlier observations demonstrating that mtOVA-mCherry is taken up by CD103⁺ DCs (Supplementary Fig. S1), injection of B16 mtOVA into BATF3 knockout mice failed to elicit detectable H-2K^b/SIINFEKL–specific CD8⁺ T-cell responses as compared with the response seen in WT C57BL/6 mice, suggesting a role for cross-presenting DCs (Fig. 4E). Next, we sought to identify the receptor involved in the uptake of mtOVA cells. In mice lacking the F-actin receptor DNGR-1 (37) and in double-knockout mice lacking the expression of the collagen scavenger receptor MARCO and the scavenger receptor A (SRA; ref. 38), we observed B16 mtOVA elicited H-2K^b/SIINFEKL–specific CD8⁺ T-cell responses comparable with those observed in WT mice (Fig. 4E; Supplementary Fig. S3C).

These results demonstrate that the cGAS–STING pathway contributes to the enhanced immunogenicity of B16 mtOVA cells. These findings also indicate that priming of H-2K^b/SIINFEKL–specific CD8⁺ T cells by the syngeneic B16 mtOVA requires BATF3⁺ DCs, but it is not reduced by the lack of DNGR-1, MARCO, and SRA receptors.

Because mitochondria express more than 1,200 proteins, some of which are known to be mutated from germline sequences (39), we next assessed whether mitochondria purified from WT B16 cells (i.e., not expressing OVA) could prime B16-specific CD8⁺ T-cell responses. Mitochondria were purified from B16 cells or from mouse adult fibroblasts (MAF) from C57BL/6 mice and were then injected SC into WT C57BL/6 mice. Purity of the mitochondrial preparation was assessed by Western blot, demonstrating an enrichment of the mitochondrial protein TOM20 and ATP-B (the mitochondrial inner membrane ATP synthase subunit) in the mitochondrial fraction, whereas cytosolic GAPDH was almost absent (Supplementary Fig. S6A and S6B). Vaccinated or mock-treated mice were then challenged s.c. with B16 tumor cells 12 days after vaccination. We observed that vaccination with B16-derived mitochondria, but not with C57BL/6 MAF-derived mitochondria, produced a response capable of controlling tumor growth (Fig. 5A; Supplementary Fig. S7A). This effect was tumor specific, as mice vaccinated with mitochondria isolated from B16 cells and then challenged with the syngeneic, but unrelated, Lewis lung carcinoma, failed to control Lewis lung carcinoma growth (Fig. 5B; Supplementary Fig. S7B). Depletion of CD8⁺ T cells resulted in enhanced tumor growth, with no significant difference between mice vaccinated with B16 mitochondria and the control group, demonstrating a role for CD8⁺ T cells in controlling tumor growth this model (Fig. 5C; Supplementary Fig. S7C).

To further demonstrate the ability of CD8⁺ T cells from mitochondria vaccinated mice to recognize B16 cells in vitro, mice were vaccinated with purified B16 mitochondria or vehicle control and then challenged with B16 cells. Splenocytes from these mice were then cocultured with B16 cells overnight. Splenocytes from naïve mice served as the no-challenge control. These experiments demonstrated that the frequency of B16-specific CD8⁺ T cells isolated from mice vaccinated with B16 mitochondria was significantly greater than B16-specific CD8⁺ T cells isolated from the control groups (Fig. 5D).

In conclusion, these results indicate that B16-derived mitochondria are a source of tumor antigens capable of inducing protective CD8⁺ T-cell responses in vivo.

Finally, we extended our studies to humans by investigating the presence of CD8⁺ T cells specific for mutated mitochondrial-localized proteins in human cancers. The frequencies of nonsynonymous somatic mutations in mitochondrial proteins that are expressed by genomic DNA from 9,508 samples across 31 cancer types from TCGA (version 02-04-2018) were investigated. We observed the presence of mutations of mitochondrial-localized proteins across different tumor types, with the highest average frequency present in endometrioid cancers (Fig. 6A). We therefore focused our studies on endometrial cancer patients and studied the immune response in 1 patient with a hypermutated phenotype caused by the loss of function of the proofreading DNA polymerase epsilon (POLE; ref. 25).

By comparing the sequences obtained from tumor and germline DNA, tumor-specific, nonsynonymous single-nucleotide variations were identified. RNA-sequencing was also performed to check the expression of potential neoepitopes (Supplementary file S1). A pipeline was used to predict mutant peptide binding affinity to the patient's HLA allele HLA-A*02:01. Epitopes were then prioritized based on their mitochondrial or nonmitochondrial localization, and 60 peptides were synthesized for each location group. The patient's PBMCs were stimulated and expanded in the presence of each peptide, and HLA-A2 tetramers loaded with mitochondrial or not mitochondrial-derived peptides were used to assess the presence of neoantigen-specific T cells in the expanded PBMC. We identified CD8⁺ T cells recognizing 4 neoantigens derived from 2 mitochondrial-localized mutated proteins and 3 neoantigens derived from proteins localized in the cytosol (Fig. 6B; Supplementary Fig. S8), which were capable of specifically killing peptide-pulsed autologous EBV-LCLs (Fig. 6C) and producing IFNγ (Fig. 6D).

Clinical results have demonstrated a significant correlation between the number of predicted HLA-binding peptides derived from mutated proteins in tumor cells and the efficacy of immune-checkpoint blocking antibody treatment in cancer patients (1). Thus, our ability to improve the identification and ranking of immunogenic neoantigens is needed to optimize the development of cancer vaccines and the effectiveness of checkpoint blockade therapies. Although several strategies are currently pursued to improve algorithms predicting immunogenicity of neoantigens (40, 41), further improvements are still required.

The results of our studies highlight the utility of considering the location of antigenic proteins in mitochondria as an additional criterion to prioritize neoantigen predictions, as we showed an increased immunogenicity of mitochondrial-localized OVA and NY-ESO-1 proteins. We demonstrated that their enhanced immunogenicity is due to their increased stability and to the activation of the cGAS/STING pathway. In contrast, cytosolic localized OVA and NY-ESO-1 proteins, which are rapidly degraded by the proteasome and efficiently directly presented by tumor cells in vitro, fail to induce strong antigen-specific CD8⁺ T-cell responses in vivo. These results are supported by previously published papers demonstrating a correlation between protein stability and their cross-priming abilities (13, 32, 38). The more efficient cross-priming of mitochondrial-localized proteins is likely due to the efficient uptake of mitochondria by cross-priming CD103⁺ DCs, compared with cytosolic OVA, and by a combination of the higher steady-state amount and enhanced DC maturation, possibly due to mitochondrial DNA.

The rate at which antigenic proteins are degraded in cross-presenting DC is also a determining factor in controlling the immunogenicity of cross-presented antigens (37, 42), which can be modulated by several factors, including the rate at which antigenic proteins exit from the endosomes/lysosomes to the cytosol in cross-presenting DCs (43), the expression of lysosomal proteases (44, 45), and the pH within lysosomes (44, 46, 47), which is shown to be higher in the lysosomes of DCs and improves cross-presentation of antigenic proteins (48). We found that the two forms of stable OVA (i.e., mod cytoOVA and mtOVA), which were localized in different compartments of B16 cells, displayed different abilities to prime the OVA_257–264–specific T-cell response. Although further experiments are warranted to dissect the mechanisms controlling these results, it is tempting to speculate that the observed differential priming abilities of B16 mod cytoOVA and B16 mtOVA cells may reflect differences in their processing events in cross-presenting DCs.

Different recognition pathways have been shown to provide the relevant signals for CD8⁺ T-cell priming, including extracellular uric acid generated during cell death, which may stimulate an inflammatory response, mediated by NLRP3 inflammasome activation (49). Although we did not observe any difference between WT mice and NLRP3-deficient mice, we showed that lack of the STING/cGAS pathway significantly reduced priming of CD8⁺ T cells specific for mitochondrial-localized proteins (34, 35). In contrast, DNGR-1, which was previously shown to be involved in the cross-presentation of cell-associated antigens (37), appeared not to be involved in the cross-presentation of mitochondrial OVA. Previous papers have provided insights into the mechanisms by which mitochondrial-derived proteins may intersect the antigen processing and presentation pathway (15, 16). However, these papers fail to demonstrate whether mitochondrial antigenic proteins can be efficiently taken up by DCs in vivo and cross-presented to antigen-specific CD8⁺ T cells.

We have extended results obtained with model antigens (i.e., OVA and NY-ESO-1), by using endogenous B16 mitochondria as a source of specific mitochondrial antigens (39) and demonstrated their ability to induce B16-specific CD8⁺ T-cell responses capable of controlling B16 growth in vivo. These results extend our previous findings obtained with mitochondrial-localized OVA and NY-ESO-1 protein by highlighting the strong immunogenicity of endogenous mitochondrial-localized antigenic proteins. In our analysis of CD8⁺ T cells from an endometrial cancer patient, we identified an equal number of T-cell clones specific for either mitochondrial or nonmitochondrial neoantigens. However, we note that the expansion potential of the CD8⁺ T cells that are specific for mitochondrial-localized epitopes is higher. This is consistent with the concept that tumor mitochondrial-localized neoantigens are able to prime a better CD8⁺ T-cell response during tumor development.

Although cross-presentation events of epitopes derived from mitochondrial-localized proteins are involved in inducing strong priming of tumor-specific T cells, such response would not be sufficient to control tumor growth in vivo, unless the tumor cells are able to directly present epitopes derived from mitochondrial-localized proteins. Our results demonstrated that mtOVA B16 cells can directly present the OVA SIINFEKL epitope in vivo, albeit less efficiently than B16 cytoOVA cells, and that the combination of the higher frequency of OVA-specific T cells and their ability to recognize B16 mtOVA cells in vivo accounted for the greater control of tumor growth, compared with cytoOVA B16 cells. It has been previously shown that direct presentation of epitopes derived from mitochondrial proteins relies on the generation and trafficking of mitochondrial-derived vesicles induced by heat shock, but not by IFNγ stimulation (15). We demonstrated that in order to detect the H-2K^b/SIINFEKL complex on the surface of B16 mtOVA and cytoOVA cells in vitro, B16 cells needed to be treated with IFNγ and TNFα. H-2K^b/SIINFEKL complexes were detectable ex vivo by flow cytometry on the surface of B16 mtOVA cells, suggesting that the inflammatory tumor microenvironment induces upregulation of SIINFEKL/H-2 K^b complexes. Indeed, we observed that upon depletion of CD8⁺ T cells, B16 mtOVA cells lack surface expression of H-2K^b/SIINFEKL complexes, suggesting that cytokine expression by infiltrated tumor-specific CD8⁺ T cells may be required to induce direct presentation of the SIINFEKL OVA epitope to detectable amounts.

In conclusion, our results demonstrate that the location of antigenic proteins in mitochondria significantly enhanced their ability to elicit a high frequency of antigen-specific CD8⁺ T-cell responses in vivo. Such enhanced immunogenicity is controlled by cross-priming dependent events, which are facilitated by the steady-state amount of mitochondrial-localized proteins and by the activation of the STING–cGAS pathway. Our data showed a greater immunogenicity of mitochondrial-localized, mutated proteins during tumor development; boosting this preexisting immune response through personalized vaccination would be a novel strategy to enhance the efficacy of cancer immunotherapy treatments.

No potential conflicts of interest were disclosed.

Conception and design: G. Prota, U. Gileadi, A.V. Lechuga-Vieco, J.A. Enríquez, V. Cerundolo

Development of methodology: G. Prota, U. Gileadi, A.V. Lechuga-Vieco, J.-L. Chen, T.N. Schumacher

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Prota, U. Gileadi, M. Rei, S. Galiani, M. Bedard, V.W.C. Lau, M. Artibani, S. Gordon, J. Rehwinkel, A.A. Ahmed, V. Cerundolo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Prota, U. Gileadi, M. Rei, A.V. Lechuga-Vieco, S. Galiani, L.F. Fanchi, Z. Hu, J.A. Enríquez, A.A. Ahmed

Writing, review, and/or revision of the manuscript: G. Prota, U. Gileadi, M. Rei, A.V. Lechuga-Vieco, S. Gordon, J. Rehwinkel, J.A. Enríquez, A.A. Ahmed, V. Cerundolo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Prota, J.-L. Chen

Study supervision: G. Prota, A.A. Ahmed, V. Cerundolo

This work was funded by the U.K. Medical Research Council, the Oxford Biomedical Research Centre, and Cancer Research UK (program grant C399/A2291). J.A. Enríquez was funded through MINECO (SAF2015-65633-R and SAF2015-71521-REDC). The Centro Nacional de Investigaciones Cardiovasculares Carlos III (CNIC) is supported by MINECO and the Pro-CNIC Foundation and is a SO-MINECO recipient (award SEV-2015-0505). A.A. Ahmed is supported by Ovarian Cancer Action, and Oxford Biomedical Research Centre, National Institute of Health Research. J. Rehwinkel acknowledges funding from the Wellcome Trust (grant number 100954). A.V. Lechuga-Vieco was supported by a Postdoctoral Fellowship from the Fundacion Alfonso Martin Escudero (Spain). The authors thank Oliver Schulz, Neil Rogers, and Caetano Reis e Sousa for providing DNGR-1 knockout and BATF3 knockout mice. The authors thank Kevin Maloy for providing MyD88 knockout mice. The authors thank Mariolina Salio and Giorgio Napolitani (University of Oxford) for critical revision of this article.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.