Inhibition of Mitochondrial Fission Reverses the Immunoescape of Solid Tumors via IRE1α-XBP-1s-TPP2 axis

Hypoexpression of human major histocompatibility complex (MHC) class I is widely known to be an important strategy of immune evasion in most cases of malignancies that lead to poor prognosis. We demonstrated that mitochondrial dynamics can be exploited as an ecient target in regulating MHC-I expression and cancer immunogenicity. Clinically, MHC-I expression and fragmentation of mitochondria are both closely associated to patient survival but are negatively correlated to one another. Mechanistically, it was observed that endoplasmic reticulum (ER) stress, an integrated signal transduction pathway activated in most rapidly proliferating tumor cells, played a crucial role in connecting mitochondrial fragmentation and cancer cell immunogenicity particularly via the IRE1α-XBP-1 s axis. XBP-1 s, which is activated by imbalanced mitochondrial ssion and prolonged oxidative stress, served as a potent transcription factor, promoted the expression of aminopeptidase TPP2 and destructed the applicable antigenic peptide to impede MHC-I complex maturation and the activation of adaptive immune system upon cancer antigen. Our ndings highlight the importance of mitochondrial dynamics in determining solid tumor immunogenicity and suggest that mediating mitochondrial fragmentation might provide a novel approach in antitumor immunotherapy. present work revealed that mitochondrial ssion plays a key role in regulating cancer cell MHC-I membrane expression and immunogenicity by targeting UPR, particularly the IRE1α axis in vitro and in vivo. The current study proposed a novel model whereby mitochondrial ssion participates in cancer immune escape, providing promising targets for future clinical therapeutics. were constructed by transfecting with OVA lentiviral (Cat. No. 113030, Addgene). 2 × 10 5 B16F10-OVA cells were resuspended in 150 µL PBS and then injected subcutaneously into the anks of the C57BL/6 mice. 3 days after tumor implantation, Mdivi-1 (Cat. No. HY-15886, MedChemExpress, 2.5 mg/kg) or the vehicle control (DMSO) was given by tail vein injection for 5 days in a 7-day cycle daily[15]. Tumor growth was monitored and with calipers every two days after tumor implantation. The survival endpoint was when tumor reached a diameter of 15 mm. Tumor volumes calculated with the formula: 0.5 × length×(width) 2 . To further investigate the effect of Mdivi-1 on tumor metastasis, 2 × 10 5 cells were resuspended in 200 µL PBS and inoculated via tail vein, Mdivi-1 was then given the same way as described above.


Introduction
Downregulation of MHC-I is often associated with malignant transformation of cells. This abnormality results in a defective synthesis and expression of MHC-I-tumor antigen-derived peptide complexes which provide malignant cells with a mechanism to evade the host's immune system [1]. These peptide complexes are important in the interactions between malignant cells and cognate cytotoxic T cells.
Cytotoxic T lymphocytes (CTLs) alongside other lymphocyte subsets are crucial for cancer eradication. CTLs recognize antigenic peptides of 8-10 amino acid (aa) that are derived from tumor-associated antigens (TAAs) and then presented by MHC-I [2]. Loss or downregulation of MHC-I, a condition in which CTLs cannot recognize cancer cells, however, is a de ciency which has not gained su cient attention on how to gure out its reversal in clinical senario so far. Immunotherapy is the fourth option in line for cancer therapy, trailing behind surgery, chemotherapy, and radiotherapy. T cell activation to eliminate tumor cells is the most promising anti-cancer strategy since the development of chemotherapy, as demonstrated by the shrinkage of melanoma and non-small cell lung cancer (NSCLC) in response to checkpoint blockers [3]. Unfortunately, existing immunotherapies are either only effective in limited types of tumors, or ultimately develop drug resistance after initial response, both of which prevent them from achieving clinical success. Although little is known about the mechanisms mediating primary and acquired resistance to immunotherapy, downregulation of MHC-I has been reported in resistance to immune checkpoint inhibitors (ICIs) for the cases of lung cancer and melanoma [4]. The above clinical ndings have stimulated interests in identifying the molecular mechanisms underlying MHC-I downregulation in malignant cells with the expectation that this information may set the prerequisite for development of strategies to restore MHC-I expression on malignant cells.
Mitochondrion is one of the vital organelles in eukaryotic cells. As a dynamic organelle, it constantly undergoes ssion and fusion, and it has the unique ability to regulate its morphology in response to various cellular stimuli [5]. In fact, mitochondrial hyperfragmentation has been observed in several types of cancer cells. It was reported that the inhibition of mitochondrial ssion leads to the decrease in cell proliferation and migration among models of lung, colon, gastric, breast and thyroid cancer [6]. This further con rmed the hypothesis that a fragmented mitochondrial phenotype is essential to many cancers, though limited reports had attempted to connect mitochondrial fragmention with immune escape of cancer cells.
Endoplasmic reticulum (ER) is another central organelle that is closely associated with mitochondria. It functions primarily to process newly synthesized secretory and transmembrane proteins. Rapidly proliferating cancer cells require increased ER activity to facilitate the folding, assembly and transport of membrane and secretory proteins, and are thereby subjected to ER stress [7]. Adaptation to protein folding stress is mediated by the activation of an integrated signal transduction pathway dubbed unfolded protein response (UPR) [8], which results in the activation of three ER transmembrane kinases, namely, protein kinase-like endoplasmic reticulum kinase (PERK), inositol requiring 1α (IRE1α) RNase and activating transcription factor 6 (ATF6), which are otherwise maintained in an inactive state by the association of the chaperone BiP/GRP78 with their luminal domains [9]. ER's exclusive ability in processing transmembrane protein makes it a major concern in immunology research.
Our present work revealed that mitochondrial ssion plays a key role in regulating cancer cell MHC-I membrane expression and immunogenicity by targeting UPR, particularly the IRE1α axis in vitro and in vivo. The current study proposed a novel model whereby mitochondrial ssion participates in cancer immune escape, providing promising targets for future clinical therapeutics.
1. Low pSer616 DRP-1 correlates with high MHC-I and indicates better prognosis in cancer patients The ssion process in mammals is mediated by DRP-1 (dynamin-related protein), FIS1 ( ssion 1 homologue protein), and MFF (mitochondrial ssion factor), among which the central player is the highly conserved DRP-1, which belongs to a family of large GTPases that self-assemble to regulate mitochondrial membrane structure [10]. DRP-1 activity is modulated by phosphorylation of regulatory kinases at serine 616. Upon activation, DRP-1 moves from the cytosol to the mitochondrial where it assembles in multimers that constrict and divide the mitochondria [11]. Conversely, phosphorylation of DRP-1 at serine 637 inhibits ssion [12]. Tumor samples obtained from 127 patients with head and neck squamous cell carcinoma (HNSCC), 62 patients with NSCLC and 59 patients with malignant melanoma at Sun Yat-sen Memorial Hospital, Sun Yat-sen University (Guangzhou, China) between Jan. 2006 and Dec. 2010 were used for MHC-I and phospo-Ser616 DRP-1 staining, Spearman correlation and Kaplan-Meier survival analysis. Both immuno uorescence (IF) and immunohistochemistry (IHC) staining revealed that the expression of MHC-I was much higher in tumor tissues from patients with longer survival (> 5 years, post-surgery). In contrast, pSer616 DRP-1 level was signi cantly higher in patients who didn't survive 5 years following surgery ( Fig. 1A-C). Correlation among clinicopathological status and the expression of MHC-I or pSer616 DRP-1, univariate and multivariate analysis of factors associated with overall survival of these cancer patients are shown in Supplementary Table 1-2. It was indicated that the expression level of MHC-I and pSer616 DRP-1 have close association with node metastasis but no signi cant correlation with sex, age and clinical stage (Supplementary Table 1). The levels were also indicators of prognosis in HNSCC, NSCLC and malignant melanoma patients (Supplementary Table 2). Additionally, a Spearman order correlation analysis showed that MHC-I expression was negatively correlated to pSer616 DRP-1 level (Fig. 1D); high expression of MHC-I and low levels of pSer616 DRP-1 are indicators of better overall survival (OS) in these cancer patients respectively (Fig. 1E). Other than collectively demonstrate the clinical signi cance of MHC-I and pSer616 DRP-1 which were reported previously, the above data also suggest the correlation between mitochondrial ssion and MHC-I expression, making mitochondrial ssion a plausible leverage in rescuing tumor immunogenicity in solid tumors.

Mdivi-1 mediates MHC-I expression in syngeneic tumor models
To investigate the role of DRP-1 in MHC-I antigen expression and its functional properties, we used Mitochondrial division inhibitor 1, Mdivi-1, a selective cell-permeable inhibitor of DRP-1. It inhibits the self-assembly of DRP-1 by blocking DRP-1's GTPase activity [13]. By using B16F10 melanoma cells expressing the model tumor antigen, chicken-derived ovalbumin (OVA) and C57BL/6 mice, we found that injection of Mdivi-1 via tail vein three days after tumor subcutaneous implantation signi cantly slowed down the proliferation (or promoted the apoptosis) of tumor compared to vehicle control (DMSO), as indicated by tumor volume measurement ( Fig. 2A) and harvested tumor weight ( Fig. 2B-C). Flow cytometric analysis of cancer cells isolated from the harvested tumors and stained with mAbs showed that H-2K b antigen and immunodominant (ID) OVA epitope SIINFEKL [14] expression was upregulated (Fig. 2D) on cancer cells isolated from tumors with reduced size. These phenotypic changes were associated with increased IFN-γ-producing T cells (Fig. 2E) and CTLs in ltration in tumors (Fig. 2F). The above described data were paralleled with those obtained from mice inoculated with tumor cells via the tail vein. Monitoring the latter model with an IVIS Lumina imaging system showed that Mdivi-1 treatment e ciently prevented metastasis spread (Fig. 2G); furthermore, the survival of tumor-bearing mice was prolonged (Fig. 2H). In recent years, Mdivi-1 has emerged as a promising therapeutic agent for stroke, myocardial infarction, neurodegenerative diseases and cancers [15]. For the rst time, our research revealed that Mdivi-1's activity against established tumor is at least partially due to the enhanced membrane expression of MHC-I on cancer cells, which subsequently contributed to recognition by the immune system and CTLs activation.
3. Mdivi-1 improves the adoptive T cell therapy (ACT) in PDX tumor models by upregulating MHC-I To further investigate the clinical relevance of our ndings, we explored whether Mdivi-1 could arouse the immune dormancy and improve ACT effects in cancer patient-derived xenograft (PDX) model implanted in immunocompromised NOD/SCID mice (Fig. 3A). Successful PDX engraftments were established in 11 out of 63 (17.4%) primary HNSCC samples and 6 out of 32 (18.9%) primary NSCLC samples.
Similarly, Mdivi-1 treatment effectively inhibited tumor growth, as evaluated through tumor-volume measurements (Fig. 3B) and the harvested PDX tumor weights (Fig. 3C). Double immunostaining in the PDX tissues also showed consistent results as the above, indicating that adoptive transfer with CTLs and Mdivi-1 treatment resulted in massive apoptosis of cancer cells (EpCAM + TUNEL + ) ( Fig. 3D and Supplementary Fig. 1A). To further investigate the causes of differences in tumor growth and apoptosis rate, we dissociated the PDX and found that Mdivi-1 treatment signi cantly upregulated the membrane expression of MHC-I with or without CTL transfer (Fig. 3E) as in syngeneic models, which also corresponded to and could at least partially explain the results of double immuno uorescence staining in PDX tissues depicted above. ACT was signi cantly upregulated after Mdivi-1 treatment as compared to vehicle control (DMSO) group.
This was proven by more in ltration of CD8 + T cells within cancer nests (Fig. 3F) and more in ltration of IFN-γ-producing T cells [16] (Fig. 3G). Moreover, we isolated the PDX-in ltrating CTLs and observed that perforin and granzyme B, markers associated with cytotoxic activity [17], were upregulated after Mdivi-1 treatment (Fig. 3H-I and Supplementary Fig. 1B). In a word, these ndings suggest that the inhibition of mitochondrial ssion in PDX that are transferred with tumor-speci c CTLs is able to at least partially overcome tumor immune evasion by upregulating MHC-I expression of cancer cells, thus improving the therapeutic effects of ACT on tumors.

Mdivi-1 improves the cytotoxic function of CTLs against autologous cancer cells in vitro
The above-described shrinkage of tumor and upregulation of immune response prompted us to determine Mdivi-1's effect on the cytotocix function of CTLs in vitro (Fig. 4A). CTLs were primed by tumor-lysate-pulsed DCs (dendritic cells) for 5 days [18] and co-cultured with autologous cancer cells at different E/T (effector/target) ratios. After 12 hours, death of cancer cell (indicated as EpCAM + ) was examined by PI and Annexin V uptake using ow cytometry ( Supplementary Fig. 1C). The production of Th1 cytokines [19] and residual E/T ratio were quanti ed using Enzyme-Linked ImmunoSorbent Assay (ELISA) after 24 hours (Fig. 4B) and ow cytometry after 3 days (Fig. 4C) based on CD8 and EpCAM expression, which indicated that the generation of DCs ( Supplementary Fig. 1D) and CTLs was feasible and the cytotoxic function of tumor-speci c CTLs was enhanced with increasing E/T ratio. Next, in order to determine the appropriate concentration and time of administration for Mdivi-1 in vitro, we performed sequential treatment [20] by using an MTS Assay Kit (Abcam, ab197010) and ow cytometry on HNSCC and NSCLC primary cancer cells and tumor-speci c CTLs to evaluate cell viability. It was decided that the treatment regimen of 25-50 µM for 72 hours can e ciently upregulate MHC-I membrane expression ( Fig. 4E) with comparatively small anti-proliferative effect (Fig. 4D). By pretreating Mdivi-1 according to the regimen described above, the cytotoxic function of HNSCC and NSCLC primary cancer cells that were co-cultured with tumor-speci c CTLs was dramatically enhanced. This was quanti ed by PI and Annexin V uptake by ow cytometry (Fig. 4F), Th1 cytokines production by speci c ELISAs (Fig. 4G) and residual E/T ratio by ow cytometry (Fig. 4H) compared with control group, respectively. In addition, preincubating cancer cells with neutralizing anti-MHC-I antibody [21] abrogated most of the cytotoxicity on target cancer cells. Similar results can be replicated by using melanoma cell line FO-1, a cultured cell line lost MHC-I expression completely due to a defect in B2M gene expression [22], further con rming the restriction and indispensability of MHC-I in cytotoxic function. Together, the data above suggested that inhibition of mitochondrial ssion greatly enhances antitumor immunity with comparatively smaller number of CTLs in vitro.

Inhibition of mitochondrial ssion restores the downregulation of cancer cells membrane expression of MHC-I
Since the restoration of cancer cells' MHC-I membrane expression and immunogenicity by inhibiting mitochondrial ssion had been tested through in vitro and in vivo experiments, we need to further explore the mechanism that contributes to this phenotype. Given that chemical inhibitor may have off-target effects, the results obtained by Mdivi-1 were further con rmed using a DRP-1 siRNA (small interfering RNA, GenePharma, Suzhou, China) [23]. Inhibiting mitochondrial ssion by Mdivi-1 or DRP-1 siRNA upregulated MHC-I membrane expression on tongue squamous cell carcinoma (TSCC) cell line SCC-9, CAL-27 and primary HNSCC cancer cells as described previously (Supplementary Fig. 2A and Supplementary Fig. 3A) The results of TSCC cell lines were further veri ed by immuno uorescence staining ( Supplementary  Fig. 2B). We then extended the analysis to include additional solid tumors including NSCLC, osteosarcoma, melanoma and mouse melanoma (Supplementary Fig. 2A and Supplementary Fig. 3D). From there, similar phenotype was observed, suggesting that the method of mitochondrial dynamics regulation is universally applicable. However, even though the membrane expression was signi cantly upregulated, protein level of MHC-I molecules (Supplementary Fig. 2C and Supplementary Fig. 3B) and mRNA level of HLA-A/B/C and B2M (Supplementary Fig. 2D and Supplementary Fig. 3C) stayed almost the same with or without mitochondrial ssion interference. Moreover, overexpression of DRP-1 (Generay Biotech, Shanghai, China) exhibited opposite results as that of membrane MHC-I ( Supplementary  Fig. 3E), whereas similar results were shown for transcription and translation (Supplementary Fig. 3F-G). This paradox implies that the regulation process probably happens after translation.

Inhibition of mitochondrial ssion alleviates oxidative stress and UPR in cancer cells
Mitochondria are the original source of oxygen free radicals, in particular the generation of reactive oxygen species (ROS). Excessive mitochondrial ssion may lead to change of structural organization and arrangement of electron transport chain (ETC) components within the mitochondrial membrane. This organizational arrangement of the ETC may result in perturbation of ETC activity, causing ROS overproduction [24]. ROS are small molecules that are highly active due to the presence of unpaired electrons They are important mediators of in ammation, and recent ndings have linked ER stress to the generation and accumulation of intracellular ROS, a state commonly referred to as oxidative stress. Increased protein-folding in the ER promotes ROS generation which impairs mitochondrial membrane through calcium accumulation in the mitochondria, leading to excessive ROS generation [25]. It is also reported that ER stress may further induce mitochondrial ssion [26]. Through this forward cycle, ROS exacerbates ER stress and further impairs mitochondrial function [10]. It has been widely accepted that inhibition of mitochondrial ssion effectively reduces mitochondrial ROS production and thus terminates the abovementioned cycle, followed by attenuated ER stress [27].
It was observed that SCC-9 and CAL-27's fragmented mitochondria have been distinctly transformed into lamentous phenotype by using Mdivi-1 and siDRP-1 (Fig. 5A). 10-N-nonyl acridine orange, a uorescent dye that binds to non-oxidized cardiolipin but not to oxidized cardiolipin (Fig. 5B) [28], was evaluated as well as JC-1 ( Supplementary Fig. 4A) [29], another key event re ecting oxidative stress, were signi cantly upregulated with the transformation of mitochondrial morphology. Flow cytometry analysis of intracellular ROS further validates the theory of terminating the 'stress cycle' through mitochondrial ssion inhibition ( Fig. 5C and Supplementary Fig. 4B) as described above, which could further inhibit ER Stress, as indicated by several important markers in immunoblotting (Fig. 5D). In addition, overexpression of DRP-1 e ciently promoted intracellular ROS as expected ( Supplementary Fig. 4C). With that, we summarize that inhibiting mitochondrial ssion is an effective method to attenuate oxidative stress and ER Stress within cancer cells.
7. IRE1α-XBP-1 s is the most signi cant regulator axis of MHC-I among UPR Recently, accumulating evidence revealed the connection between ER stress and MHC-I: The overexpression of an ER stress-inducing misfolded protein or the constitutive expression of nATF6 or XBP-1 s were found to be associated with decreased levels of MHC-I in 293T cells [30]. Similarly, ER stress induced by palmitate or glucose deprivation reduces class I antigen presentation in mouse thymoma cells. Moreover, inhibition of ER stress response in Hereditary Haemochromatosis (HH) cells leads to the restoration of MHC-I [31]. We used dominant negative (DN) plasmids (Addgene 20745, 36954) and siRNA (GenePharma) to interfere with the three signaling pathways and found that IRE1α, the most conserved arm of the UPR [3], is the most signi cant regulator of MHC-I ( Supplementary Fig. 5A). Within minutes of unfolded proteins accumulation, BiP dissociates from PERK, IRE1α and ATF6 and preferentially binds to the unfolded proteins, resulting in the activation of PERK and IRE1α via luminal domain homodimerization and autophosphorylation of ATF6 via proteolytic cleavage [9]. Active IRE1α then excises a 26-base intron in XBP-1 (X-box binding protein 1) mRNA. Re-ligation of spliced XBP-1 shifts the open reading frame, and its translation produces the homeostatic transcription factor XBP-1 s. Spliced XBP-1 transcription factors then translocate into the nucleus where it binds to UPR elements (UPRE) and activates many of the UPR target genes. In this situation, the unspliced XBP-1 which was indicated as XBP-1u, degraded rapidly [32]. Knockdown of XBP-1 s by siRNA (GenePharma) restored MHC-I to the level similar to that of transfected DN IRE1α ( Supplementary Fig. 5D). Co-transfection of XBP-1 s and DN IRE1α rescue MHC-I level to normal status, which further con rm this regulatory mechanism ( 8. Bioinformatic analysis based on microarray indicated TPP2 as the potential target of XBP-1 s There are several assumptions about how ER stress affects the expression of MHC-I on cell surface, such as (i) to relieve the folding pressure, premature protein is removed from ER including MHC-I [31], (ii) ER stress induced blockage of protein synthesis via PERKmediated eIF2α phosphorylation contributed to diminished peptide loading [34], (iii) other antigen processing machinery ((APM, such as TAP (transporter associated with antigen presentation) translocates peptides from the cytosol to the ER lumen where loading onto MHC-I molecules takes place and tapasin, plays a decisive role in the formation of low off-rate MHC-I/peptide complexes)) members have been perturbed [30,35]. However, the molecular mechanism of this phenomenon has never been identi ed. The expression microarrays indicated that 624 mRNAs were upregulated and 652 mRNAs were found to be signi cantly downregulated (Fig. 6A) after the knockdown of XBP-1 s. Based on previous observations, downstream genes regulated by XBP-1 s are supposed to be involved in the post-translation procedure, namely translocation, metabolism, protein modi cation, et al. We next performed the GeneSet Enrichment Analysis (GSEA) with the FDR signi cance cutoff of 0.05, we found several associated genes which participated in immune, metabolism, post-translation and cancer pathways (Fig. 6C). The identi ed genes were speci cally exhibited on the basis of fold change and p value in Fig. 6B and the associated gene sets were displayed in Fig. 6D-E, with NES indicating normalized enrichment score. We divided these genes into four groups according to their recognized functions in cancer biology (Fig. 6G). Among these genes, tripeptidyl peptidase 2 (TPP2), attracted our attention, as it is a well-recognized mammalian aminopeptidase that removes tripeptides from the N-terminus of longer peptides at neutral pH. It is now well accepted that TPP2 play an essential role in some MHC-I antigen presentation and CD8 + T cells maturation [36]. Other genes such as CUL3, CTSF, MDM2, MRC2, CTSO, AKT2, ATG5, ATG7, UBE2R2 and ERAP1 that possess similar function of degrading or metabolizing speci c types of proteins are shown in Fig. 6F. To further identify the functionality of XBP-1, we referred to the human ENCODE database and found that XBP-1 was frequently enriched in the common region at TPP2 promoter in three different cancer cell lines (Fig. 6H). We next performed de novo motif discovery on XBP-1 peaks and observed that the 5'-AGCAGCACGTGATT-3' (reverse complement: 5'-TCGTCGTGCACTAA-3') motif was highly enriched (Fig. 6I). ChIP and luciferase reporter assays identi ed the transcriptional functionality of XBP-1 as well ( Fig. 6J-L). As expected, results obtained from the microarray were further veri ed by quantitative real-time PCR (qRT-PCR; Fig. 6M and Supplementary Fig. 6A) and immunoblotting (Fig. 6N). Moreover, examination of The Cancer Genome Atlas (TCGA) database further showed that overexpression of TPP2 is correlated with worse prognosis in a wide range of solid tumors ( Fig. 6O and Supplementary Fig. 6B).
9. TPP2 prevents MHC-1 maturation by the destruction of antigenic peptides Antigen presentation is a sophisticated process that involves a number of crucial events: (i) MHC-I folding and maturation, (ii) peptide generation and shuttle into the ER, (iii) assembly of the APM, (iv) tra cking of peptide loaded MHC-I through the Golgi apparatus to cell surface and (v) ultimately its recognition by CD8 + T lymphocytes. During this delicate process, any corrupted intracellular events could lead to the generation of high off-rate MHC-I/peptide complexes that dissociate prematurely during their journey towards cell surface [31]. Co-transfection of siTPP2 and XBP-1 s rescued MHC-I to its unaffected level in TSCCs and B16F10, further veri ed that TPP2 is the downstream gene regulated by XBP-1 s and is involved in MHC-I maturation (Fig. 7A). To further investigate its speci c role in antigen processing and presentation, we used B16F10 cell line that was stably transfected with either shTpp2 or shCtrl (small hairpin RNA, Genechem, Shanghai, China). When transiently transfected with OVA (Generay) and SIINFEKL plasmids (Cat. No. 102944, Addgene), it was observed that knockdown of Tpp2 e ciently promoted the SIINFEKL-H-2K b complex expression (Fig. 7B). The abovementioned stably transfected cells were then co-cultured with the isolated OVA-speci c CTLs derived from OT-1 transgenic mice ( Fig. 7B) [37]. We tested the production of mouse IFN-γ ( Fig. 7D) and the speci c lysis of cancer cells ( Fig. 7E) after co-culturing for a predetermined time and E/T ratio, It was revealed that Tpp2 knockdown signi cantly promoted recognition of the epitope by CTLs. The results above were all evidences that point to the predominant role of Tpp2 in destroying peptides, i.e. cleave them to sizes smaller than required to bind to MHC-I (i.e.: < 8-10 amino acids) [14]. Given the fact that TPP2 regulates the immunogenicity of cancer cells by degrading antigenic peptides, it can be inferred and experimentally veri ed that the inhibition of mitochondrial ssion would not affect the expression of other immunological-related or unrelated transmembrane proteins (Fig. 7F). The precision in this method avoids any off-target occurrences and side effects. Suppressing the degradation of antigenic peptides is a possible mechanism to which mitochondrial dynamics regulate cancer immunogenicity. The detailed mechanism of present study is exhibited in Fig. 7G.

Discussion
Immune escape of tumor is the central mechanism that leads to di culty in early diagnosis, in nite proliferation and metastasis without restriction and ultimately death of patients. Loss or downregulation of membrane MHC-I has always been considered to be an important cause of tumor escaping from immunological surveillance [2] and is also frequently found in solid tumors, including malignant melanoma, breast cancer, stomach cancer, colon cancer, and bladder cancer [1]. In this study, we characterized MHC-I membrane expression as a prognosis indicator in HNSCC, NSCLC and melanoma patients, which has con rmed the ndings of several previous research [1][2]4].
However, there has been little research on the restoration of MHC-I expression to reverse the immunoescape phenotype and promote CTLs recognition and eradication of cancer cells.
Mitochondria are the indispensable regulatory centers shared by all eukaryotic cells and have always been the focus of basic scienti c research. In recent years, the unique mitochondrial dynamics has gradually received more and more attention in addition to studies on mitochondrial metabolism and apoptosis. The mitochondrial ssion and fusion are both closely related to important physiological processes. For example, mitochondrial ssion produces smaller, fragmented mitochondria, which are important for mitochondrial movement to regions of high energy demand or to allow for equal mitochondrial distribution to daughter cells following mitosis [38]. Mitochondrial ssion is also implicated in the release of cytochrome C into the cytosol to trigger apoptosis, or for the destruction of damaged cellular organelles [39]. More importantly, impaired fusion and enhanced ssion have been frequently observed in solid tumors [40]. Several studies have demonstrated that mitochondrial ssion is required to maintain the metastasis potential and proliferating rate of breast, thyroid, and glioblastoma cancer cells [41]. It was also observed that tumor growth may be blocked by inhibition of mitochondrial ssion through knockdown of DRP-1 [6]. In the present study, we have also determined that pSer616 DRP-1 is a indicator of poor prognosis in HNSCC, NSCLC and melanoma. Our research also revealed for the rst time that mitochondrial ssion can be utilized as a promising target to regulate immunogenicity of cancer cells which can eventually in uence the biological behavior and clinical outcome of tumors. This is an important discover compared to the existing research regarding mitochondrial dynamics. Interestingly, constrast to mitochondrial ssion, fusion of mitochondria into linear or tubular networks limits deleterious mutations in mtDNA (mitochondrial DNA), induces supercomplexes of the ETC thus maximizes OXPHOS (oxidative phosphorylation) activity, enhances ER interactions important for Ca 2+ ux and protect mitochondria from autophagic degradation [42]. In addition, mitochondria elongate as a survival mechanism in response to nutrient starvation and stress, linking fusion to cell longevity and persistence [43]. This indicates that inhibiting mitochondrial ssion namely promoting fusion, can dramatically mitigate the damage that traditional methods such as chemotherapy and radiotherapy poses on non-transformed cells, which was a major concern of side effect and an obstacle hampering many therapies from clinical application.
ER stress caused by abnormal accumulation of unfolded proteins in ER is a hallmark feature of secretory cells and many diseases, including diabetes, neuro-degeneration, and cancers [8]. The UPR can promote survival under conditions of transient and mild ER stress.
For example, ER stress promotes angiogenesis through stimulating VEGF expression and secretion. It also induces cancer cell dormancy through G1 arrest in response to decreased cyclin D1 downstream of PERK activation. The robust upregulation of GRP78 and other ER chaperones by the UPR can enhance the ER protein folding capacity and reestablish ER homeostasis, which protects the cancer cells from apoptosis and allows for recurrence once favorable growth conditions return [44]. However, if ER stress is prolonged and severe, coupled with failure of compensatory mechanisms, apoptosis will be initiated through activation of transcription factor CHOP by downstream signaling from PERK via ATF4. Although the molecular mechanisms underlying this switch remain poorly understood, each apical UPR sensor holds a dualistic role in propagating adaptive as well as toxic signals [32]. Thus, from a 'simplistic' point of view one would like to promote severe ER stress in cancer cells by therapeutics that either block the pro-survival pathways and/or promote the pro-apoptotic signals emanating from the UPR, such as using proteasome inhibitors like HIV protease inhibitors [48] to further increase the protein burden in the already challenged ER, or HSP90 inhibitors which can activate all three UPR branches [46]. The above methods are all plotting to further aggravate ER stress, although they do cause damage to cancer cells, it is inevitable that non-transformed cells may severely suffer as well. In recent years, there have been several reports regarding small molecule inhibitors targeting UPR, such as the IRE1α-speci c inhibitor 4µ8c [3], ER Stress inhibitor TUDCA [31], inhibitor of PERK's kinase domain named GlaxoSmithKline (GSK) [47], et al. However, their application is limited due to off-target effect and pancreatic toxicity [52]. Moreover, the molecular insights on apoptosis/survival decisions during ER stress are still too limited and the risk exists that the anticancer drug in question might actually end up blocking ER stressmediated apoptosis, thereby promoting rather than preventing tumor progression. Not surprising that there are con icting data in literature considering the impact of inhibiting PERK or IRE1α in cancer therapy [48]. The treatment strategy proposed by us of targeting mitochondrial ssion rather than ER stress subtly bypass the above drawbacks we may encounter. Regulating mitochondrial ssion is an easier, safer and more controllable way as compared to ER stress.
Rescuing loss or downregulation of MHC-I is a promising method to initiate a cytotoxic immune response against the transformed cells. A strong link between UPR and the MHC-I antigen presentation pathway defects was established by previous research [30] and our study. However, the broad impact of UPR on distinct cellular mechanisms, combined with complexity of the antigen presentation pathway, does not make the establishment of a mechanistic link between ER stress response activation and MHC-I impairment an easy task [31]. By using expression microarray, bioinformatic analysis, ChIP-qPCR and luciferase reporter assay, TPP2 has emerged as a plausible link between MHC-I expression and UPR, especially XBP-1 s. Peptides that bind to MHC-I are produced from intracellular proteins as a byproduct of protein catabolism. The major protease responsible for the initial cleavage of cellular proteins into oligopeptides is the proteasome. The majority of peptides produced by proteasomes are very rapidly hydrolyzed into amino acids by the concerted action of aminopeptidases and endopeptidases in the cytosol. However, a small fraction of peptides escape destruction and are transported by TAP into ER where ones of the right size and sequence bind to newly assembled MHC-I. However, proteasomes more frequently generate peptides (~ 10-20%) that are too long to bind to MHC-I molecules but can serve as potential antigenic precursors. These long precursors can be converted to MHC-I binding peptides by aminopeptidases, or may be completely degraded to amino acids by aminopeptidases and endopeptidases [49]. When closely examined, most aminopeptidases preferentially degrade relatively short peptides. They have little or no activity on peptides that are longer than about 16 amino acids when tested in vitro, while TPP2 works as an exception for its ability to degrade longer peptides [50]. TPP2, the aminopeptidase is reported to be involved in the generation of many epitopes recognized by CD8 + T lymphocytes, however, it also destroys epitope-containing peptides. Due to the fact that very few peptides su ce for immune surveillance by CD8 + T cells, the productive/destructive balance is not always positive for the generation of MHC-I ligands [51]. Although the exact effect of TPP2 on class I antigen processing and presentation is still controversial, our data suggested a moderate yet a predominantly destructive role of TPP2. Meanwhile, we have also put forward several post-translation related genes and pathways that require further exploration.

Methods
Patients and tissue samples 127 HNSCC samples, 62 NSCLC samples and 59 melanoma samples collected at Sun Yat-sen Memorial Hospital, Sun Yat-sen University (Guangzhou, China) between January 2006 and December 2010 were used for MHC-I, pSer616 DRP-1 staining, Spearman correlation and Kaplan-Meier survival plotting. The date of death was obtained from patient records or through follow-up telephone calls. Survival time was calculated from the date of surgery to the date of death or to the last follow-up. Each patient has been followed up for at least 60 months. Additionally, tumor samples and peripheral blood samples obtained from 109 patients with HNSCC and 60 patients with NSCLC at Sun Yat-sen Memorial Hospital, Sun Yat-sen University between April 2015 and October 2018 were used for primary cancer cell and T cell isolation and analysis. 15-20 mL peripheral blood was obtained from each patient. The clinical features of these patients are provided in Supplementary Table 3

Primary cells isolation from tumors and peripheral blood of patients with cancer
Tumors were cut into small fragments (approximately 1 mm 3 ) and incubated for 30 minutes with collagenase type I and III (Worthington Biochemical), in RPMI 1640 medium containing 2% FBS (5 ml/g tumor tissue) at 37 ℃. The tumor pieces were transferred to a tissue digestion C-tube (Miltenyi Biotec) and further dissociated enzymatically and mechanically on a gentleMACS Dissociator (Miltenyi Biotec) to obtain a single-cell suspension. Primary cancer cells were puri ed with EpCAM + microbeads (Cat. No. 130-061-101, Miltenyi Biotec) [18].
CD3 + T cells were isolated from peripheral blood with EasySep™ Human T Cell Isolation Kit (Cat. No. 17951, Stemcell) according to the manufacturer's instructions. The isolated cancer cells were discarded after ten passages.

Preparation of cancer cells lysate
The isolated autologous cancer cells were incubated with 0.01% EDTA-solution for 5 minutes, carefully detached with a cell scraper, washed twice in PBS and resuspended at a density of 5 × 10 6 /ml in serum-free medium. The cell suspensions were frozen at -80℃ and disrupted by four freeze-thaw cycles. For the removal of crude debris, the lysate was centrifuged for 10 minutes at 300 g. The supernatant was collected and passed through a 0.2 µm lter. The protein concentration of the lysate was determined by commercial assay (Cat. No. 5000002, BioRad) [21].

Generation of DCs and tumor-speci c CTLs
DCs were generated as previously described with slight modi cations [18,21]. In brief, peripheral blood mononuclear cells (PBMCs) were isolated from the peripheral blood of patients with HNSCC or NSCLC by SepMate™ tube for density gradient centrifugation (Cat. No. 86450, Stemcell) and were subsequently allowed to adhere in culture asks for 1 hour. The initial adherent cell fraction was harvested and cultured in DMEM containing 50 ng/ml GM-CSF, 20 ng/ml IL-4 (PeproTech) and 10% heat-inactivated FBS for 6 days. The cultures were replaced with fresh medium and cytokines every three days, and cell differentiation was monitored through light microscopy. DCs were matured through incubation with 100 ng/ml LPS (Sigma) and 500 U/ml IFN-γ (PeproTech) for 48 hours and then pulsed for 24 hours with cancer cell lysates (200 µg protein/ 1 × 10 6 cells/ ml) as prepared above. DCs maturation were con rmed to be 90% pure by ow cytometric analysis for speci c functional markers CD80, CD83 and CD86 (Supplementary Fig. 1D). To generate tumor-speci c CTLs, we incubated the isolated CD3 + T cells with antigen-speci c DCs (5:1) in RPMI 1640 medium supplemented with 25 U/ml IL-2 (Peprotech) and 10% heat-inactivated FBS for 5 days. The generated CTLs' ability to kill autologous cancer cells was determined by PI and Annexin V uptake of cancer cells co-cultured ( Supplementary Fig. 1C), Th1 cytokine production (Fig. 4B) and residual E/T ratio (Fig. 4C).

Patient-derived xenograft (PDX) implantation
Primary specimens were collected from patients with HNSCC and NSCLC who underwent tumor resection at Sun Yat-sen Memorial Hospital, Sun Yat-Sen University. Four-week-old female NOD/SCID mice purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd. (Beijing, China) and maintained under pathogen-free conditions were used for PDX transplantation, with three mice per experimental group. The PDX procedure was performed as previously described [20]. Brie y, a small incision was made on the ank of anaesthetized NOD/SCID mice. The primary HNSCC and NSCLC samples were then minced into fragments of 2-3 mm 3 and implanted into the subcutaneous tissue. The incision was then closed with sutures. The time from patient collection to mouse implantation ranged from 1-3 hours. Tumor formation was monitored with calipers per week following implantation.
Adoptive cell therapy (ACT) CD3 + T cells were isolated from peripheral blood of the same patients with HNSCC or NSCLC. Mature DCs were generated and veri ed as described above, then incubated with autologous CD3 + T cells (DC/T cell ratio 1:5) for 5 days. Then, 2.5 × 10 6 T cells and 0.5 × 10 6 DCs were intravenously transfused into each PDX-bearing mouse via tail vein after palpable tumor formation. Mdivi-1 (2.5 mg/kg) or the vehicle control (DMSO) was given on the same day as CTL transfer by tail vein injection for 5 days daily in a 7-day cycle. Tumor growth was monitored and recorded with calipers weekly after ACT. The survival endpoint was when tumors reached a diameter of 15 mm. The dissociation of harvested PDXs was the same as primary cell isolation as described above. The tumor in ltrated T cells (TILs) were puri ed with CD8 microbeads (

Enzyme-linked ImmunoSpot (ELISpot)
To analyze the tumor in ltrated T cells, single cell suspensions were generated as previously described from harvested tumors or PDXs and were rested overnight to get rid of living tumor cells via plastic adherence. Viable cells were separated via density gradient centrifugation and added to the ELISpot plate [16]. The number of IFN-γ-producing T cells was determined with IFN-γ ELISpot kit (Cat. No. 2210001 and 211001, Dakewe) according to the manufacturer's protocol.
Cytotoxic and co-culture assays Primary specimens were collected as described above. 46 HNSCC and 28 NSCLC samples with autologous peripheral blood were used for co-culture analysis. Tumor-speci c CTLs, generated by incubation with tumor antigen pulsed DCs as described above, were co-cultured with the autologous cancer cells at indicated E/T ratio. Cancer cells were pretreated with Mdivi-1 at 50 µM for 72 hours or left untreated before co-culture. After 12 hours, EpCAM + cells were harvested and death was assessed by ow cytometry with apoptosis detection kit (Cat. No. 88-8007-74, eBioscience) that stains for Annexin V and PI, according to the manufacturer's instructions. The percentages of apoptotic cells include the percentages of early (Annexin V + P − ) and late apoptotic cells (Annexin V + PI + ). Speci c apoptosis was calculated as: percentage of induced apoptosis-percentage of spontaneous apoptosis)/(100%-percentage of spontaneous apoptosis) × 100% as previously described [52]. We mainly focused on late apoptosis. Supernatant was collected at 24 hours after co-culture to measure IFN-γ, IL-2 and TNF-α release using speci c ELISAs (Cat. No. ELH-IFNg-1, ELH-IL2-1 and ELH-TNFa-1, Raybiotech). After 72 hours of culture at 37 ℃, adherent cancer cells and T cells were collected altogether and residual cancer cells and T cells were assessed by ow cytometric analysis based on EpCAM and CD8 expression respectively. To determine MHC-I restriction of cancer cells lysis, the target cells were preincubated with MHC-I blocking antibody w6/32 (Cat. No. sc-32235, Santa Cruz, 10 µg/mL) for 2 hours at 37 ℃ before coculture [21]. The result was further veri ed by using immunode cient FO-1 melanoma cell line [53].

Mitochondrial staining
To visualize the changes in mitochondrial morphology, we planted cells onto coverslips and treated as described. Then, the cells were stained for 30 minutes with 0.02 µM MitoTracker Red CMXRos (Cat. No. M7512, ThermoFisher) at 37 ℃ protected from light. The images were acquired with laser scanning confocal microscopy (LSM 800 with Airyscan, Zeiss).

Measurements of oxidative stress
To evaluate the oxidative stress within cancer cells, we tested several indicators as follow: oxidation of cardiolipin, intracellular ROS and mitochondrial membrane potential (ΔΨm) [28]. We assessed cardiolipin oxidation by using uorescent dye NAO (Cat. No. A1372, ThermoFisher), which binds to non-oxidized cardiolipin, but not oxidized cardiolipin. We planted cells onto coverslips and treated as described, then incubated cells with 100 nM NAO for 15 minutes at 37 ℃ protected from light. The images were acquired with laser scanning confocal microscopy (LSM 800 with Airyscan, Zeiss). Intracellular ROS production was measured by ow cytometry using the oxidation sensitive dye DCFH-DA (Cat. No. S0033, Beyotime Biotechnology). The cells were seeded in 6-well plates and treated as described. After washing with preheated PBS for three times, the cells were incubated with DCFH-DA at 37 ℃ for 25 minutes and detached, suspended to be tested. The green uorescence was measured using FITC channel. Similarly, mitochondrial membrane potential (ΔΨm) was measured by ow cytometry using MitoScreen (JC-1) Kit (Cat. No. 551302, BD Pharmingen) according to the manufacturer's protocol.
The anti-CD3/CD28 activated CTLs described above were co-cultured with pretreated B16F10 at indicated E/T ratio. After 12 hours, GFP + cells were harvested and death was assessed by ow cytometry with apoptosis detection kit (Cat. No. 88-8007-74, eBioscience) that stains for annexin V and PI, according to the manufacturer's instructions. The calculation of speci c death is described above. Supernatant was collected at 24 hours of culture to measure IFN-γ using speci c ELISA (Cat. No. ELM-IFNg-1, Raybiotech).

Bioinformatic analysis
We obtained XBP-1 ChIP-seqs of T47D, HS578T and MDA-MB-231 cells from ENCODE, then processed them by ENCODE processing pipeline. To predict the potential XBP-1 binding sites at TPP2 promoter region, we used motif-counter (https://bio.tools/motifcounter) to scan TPP2 promoter region from both strands with XBP-1 motif obtained from JASPAR database.

Chromatin immunoprecipitation assays (ChIP)
ChIP assays were performed as previously described [54]. Brie y, cells (5 × 10 6 ) were washed with PBS and incubated for 10 min with 1% formaldehyde at room temperature. Crosslinking was halted with 0.1 M glycine for 5 min. The cells were washed twice with PBS and lysed for 1 h at 4 °C in a lysis buffer, then sonicated into chromatin fragments with an average length of 500-800 bp, as assessed via agarose gel electrophoresis. The samples were precleared with Protein-A agarose (Roche) for 1 h at 4 °C on a rocking platform. Then, 5 µg of speci c antibodies was added and the samples rocked overnight at 4 °C. Immunoprecipitated DNA was puri ed using the QIAquick PCR puri cation kit (Qiagen) according to the manufacturer's protocol. The nal ChIP DNA was then used as a template in qPCR with the primers in Table S 5. ChIP-grade anti-XBP-1 antibody (Santa cruz, sc-8015) and anti-RNA polymerase II antibody (Abcam, ab5131) were used in this study.
Luciferase reporter assay A luciferase assay was carried out as previously described with modi cations [55]. Brie y, pGL4-TPP2-wide type(-wt) or mutant (mut) was obtained by cloning a 2000 bp DNA fragment (from TPP2 transcriptional starting site) into the pGL4.20-Basic vector upstream of the luciferase reporter gene. The pGL4.20 derivated reporter vectors were transfected into cells and the stable cell lines were obtained through puromycin selection for two weeks. The pRL-TK plasmid delivering Renilla Luciferase was co-transfected as a control. The luciferase activities were measured using a Dual Luciferase Reporter Assay Kit (Promega), and the target effect was presented as the luciferase activity of the reporter vector with the target sequence relative to that without the target sequence. Immuno uorescence and immunohistochemistry staining Antigen retrieval was performed by incubation of the slides in a pressure cooker for 5 minutes in 0.01 M citrate buffer, pH 6.0 and subsequent treatment with 3% hydrogen peroxide for 5 minutes. Slides were incubated overnight at 4 ℃ with antibodies as follow: HC-10 + HC-A2 (targeting MHC-I, a generous gift from Dr. Koichi Sakakura, Massachusetts General Hospital, Harvard Medical School), pSer616 DRP-1 (Cat. No. bs-12702R-HRP, Bioss, 1:300) and EpCAM (Cat. No. ab213500, Abcam, 1:16000). Immunohistochemical staining was performed according to the manufacturer's instructions. In order to simplify the analysis, expression of MHC-I and pSer616 DRP-1 were scored based on both intensity (0-3) and extent of staining (1)(2)(3). A multiplicative staining score was calculated by multiplying the intensity and extent scores to yield scores on a 9-point scale from 1 to 9. When dealing with multiple scores per patient, the individual scores were averaged to obtain a nal score. Cases were divided based on staining scores into 3 groups: weak (0-2), moderate (3)(4)(5) and strong (6-9) [19]. For immuno uorescence, specimens were incubated with MHC-I (Cat. No. sc-32235, Santa Cruz, 1:200), pSer616 DRP-1 (Cat. No. 4494, Cell Signaling Technology, 1:3200), EpCAM (Cat. No. ab213500, Abcam, 1:100) and CD8 (Cat. No. ab93278, Abcam, 1:500). The accuracy of automated measurements was con rmed through independent evaluation by two pathologists. Cells stained with the indicated antibodies were counted at 400 × magni cation at least ten elds per section. The stained slides were imaged using an upright uorescence microscope (Axio Imager A2, ZEISS).

Statistics
All data are expressed as the mean ± standard error of mean (s.e.m.). All statistical analyses were performed in SPSS Windows version 13.0. Spearman correlation analysis were used to assess the relationship between MHC-I and pSer616 DRP-1 expression. Kaplan-Meier survival curves were plotted and log-rank tests were performed. All experiments were performed at least in triplicates and the exact numbers of independent experiments with similar results are indicated in the gure legends. All statistical analyses of experiments were performed with two-tailed Student's T tests unless otherwise stated. p < 0.05 was considered statistically signi cant.     Mdivi-1 improves the cytotoxic function of CTLs targeting autologous cancer cells in vitro. A. Scheme of tumor-speci c CTLs generation and in vitro co-culture. B. 24 hours later, Th1 cytokines production was quanti ed by speci c ELISAs (mean ± s.e.m; n = 5; **, p<0.001 compared with non-coculture group). C. Residual E/T ratio was evaluated by ow cytometry after 3 days based on CD8 and EpCAM expression (mean ± s.e.m; n = 3; *, p<0.01, **, p<0.001 compared with E/T ration=1:1 group). D. Cell viability of primary cancer cells and PI uptake of tumor-speci c CTLs after sequential treatment by MTS and ow cytometry. E. Evaluation of MHC-I membrane expression in HNSCC and NSCLC primary cancer cells by ow cytometry. F indicates the fold change of MFI normalized to mock (mean ± s.e.m; n = 7; #, p<0.05, **, p<0.001 compared with mock). F. 12 hours after co-culture, EpCAM+ cancer cells were harvested and cell death was examined with ow cytometry based on the uptake of PI and Annexin V. E/T ratio was 5:1 (mean ± s.e.m; n = 4; **, p<0.001 compared with mock). G. Th1 cytokines production was quanti ed by speci c ELISAs after 24 hours. E/T ratio was 5:1 (mean ± s.e.m; n = 5; **, p<0.001 compared with mock). H. Evaluation of residual E/T ratio by ow cytometry based on CD8 and EpCAM expression respectively following 3 days. E/T ratio was 5:1 (mean ± s.e.m; n = 3; *, p<0.01, **, p<0.001 compared with mock). E indicates effector cells, namely T cells. T indicates targeted cells, namely cancer cells. For experiments that were conducted at different time points, the isotype tubes were adjusted to ensure that their uorescence intensity was at similar levels. NC indicates negative control.