Endoplasmic Reticulum Stress Contributes to Mitochondrial Exhaustion of CD8⁺ T Cells

In response to antigen, T cells undergo rapid expansion, engaging in up to 15 to 20 cell divisions, and proliferation is characterized by differentiation to an effector phenotype from a naïve state (1). To support the bioenergetic demands of growth and proliferation, naïve T cells that primarily engage oxidative phosphorylation (OXPHOS) shift to generate adenosine triphosphate (ATP) via increased OXPHOS and aerobic glycolysis as they become functional effector cells (2, 3). Highly differentiated effector T cells rely primarily on glycolysis and are characterized by a loss of mitochondrial integrity (4) that likely accounts for their inability to effectively continue OXPHOS. The cell-intrinsic mechanisms that consume metabolic energy and impart mitochondrial exhaustion in T cells are unknown.

Programmed cell death protein 1 (PD-1) is expressed on highly differentiated effector T cells that have become exhausted due to chronic antigen exposure (5). In cancers, programmed death-ligand 1 (PD-L1) is expressed by tumor cells to further impair antitumor effector function of PD-1⁺ T cells (6). Checkpoint blockade therapy is a means to reinvigorate effector function of exhausted T cells by inhibition of the PD-1–PD-L1 interaction (7). Unfortunately, anti–PD-1 therapy remains ineffective for the majority of cancer patients due to the initial lack of response or loss of durable responses (8–10). In early-stage exhaustion, PD-1⁺ T cells show diminished OXPHOS and glycolysis, and long-term chronically exhausted PD-1^high T cells exhibit a dependence on glycolysis due to dysfunctional mitochondria (11). T-cell–intrinsic factors that drive PD-1⁺ CD8⁺ tumor-infiltrating lymphocyte (TIL) metabolic exhaustion in response to tumor antigen are unknown, and simple methods to characterize the metabolic state of PD-1⁺ TILs in patient tumors are ill-defined.

T-cell expansion initiated by antigen recognition requires T effector cells to greatly increase new protein synthesis and initiate posttranslational modifications (12). Imbalances of unfolded and misfolded proteins are detected by endoplasmic reticulum (ER) stress sensors inositol-requiring enzyme-1 (IRE1α), protein kinase R-like endoplasmic reticulum kinase (PERK), and activating transcription factor 6 (ATF6; ref. 13). The acute arm of the stress response aims to alleviate the burden of unfolded or misfolded proteins in the ER to restore proteostasis. In the face of insurmountable stress, the unfolded protein response (UPR) initiates cell death. PERK mediates the terminal UPR through regulation of transcription factors activating transcription factor 4 (ATF4) and C/EBPα homologous protein (CHOP), which induce downstream target ER oxidoreductase 1 (ERO1α; refs. 14–17). As a catalyst of disulfide bond formation in the ER, ERO1α facilitates oxidation-reduction reactions (Redox). Excess ERO1α produced in response to chronic protein folding induces accumulation of reactive oxygen species (ROS) and impairs survival (15, 16). The PERK-specific cell stress response has not been measured in T cells or applied to the rubric of antitumor immunity.

Here, we report that the PERK axis contributes to ERO1α activation in T effector cells. We find that this axis consumes metabolic energy in T cells and drives a protein profile synonymous with oxidative stress. We show that accumulation of mitochondrial ROS (mtROS) is a hallmark of ER-induced mitochondrial exhaustion, connecting ER biology to metabolic function in T cells. Memory T cells show reduced expression of mtROS and ERO1α compared with T effectors. PERK KO, and PERK, or ERO1α inhibitor–treated T cells exhibited superior tumor control compared with T effectors. Tumor antigen–specific PD-1⁺ CD8⁺ TILs acquire mtROS in tumors. In sarcoma patients and a sarcoma mouse model, we find that high proportions of PD-1⁺ CD8⁺ TILs express mtROS. In vivo treatment with a PERK inhibitor reduced mtROS in PD-1⁺ CD8⁺ TILs, increased tumor clearance, and extended animal survival in combination with anti–PD-1 therapy.

T-cell–specific deletion of PERK on a C57BL/6J background was accomplished by cross of PERK^loxP (Eif2ak3^tm1.2Drc/J) mice with OT-1 (C57BL/6-Tg(TcraTcrb)1100Mjb/J) or Lck-Cre (B6.Cg-Tg(Lck-icre)3779Nik/J) mice to obtain OT-1xPERK^f/f and LckCrexPERK^f/w mice, respectively. These lines were crossed to obtain OT1⁺Lckcre⁺PERK^f/f mice and OT1⁺Lckcre⁻PERK^f/f littermate controls. Lck-Cre, PERK^loxP, OT-1, C57BL/6J, Ly5.2 (B6.SJL-Ptprc^a Pepc^b/BoyJ), and P-mel (B6.CgThy1^a/CyTg(TcraTcrb)8Rest/J) mice were obtained from The Jackson Laboratory. All animal experiments were approved by the Medical University of South Carolina (MUSC) Institutional Animal Care and Use Committee, and the Division of Laboratory Animal Resources at MUSC maintained all mice.

This work was determined by MUSC Institutional Review Board to be exempt under protocols Pro00055941 and Pro00055960. Normal donor patients undergoing routine non–cancer-associated surgery (Pro00055941) or patients undergoing surgical removal of high-grade deep pleomorphic undifferentiated sarcomas (Pro00055960) granted consent under MUSC Biorepository surgical consent forms. Tissue samples were deidentified. Studies were conducted in accordance with the Declaration of Helsinki, International Ethical Guidelines for Biomedical Research Involving Human Subjects (CIOMS), Belmont Report, or U.S. Common Rule. Blood (8 mL) was collected in EDTA coated tubes and peripheral blood mononuclear cells (PBMCs) were isolated via Histopaque-1077 centrifugation (Sigma). Sarcoma tissue was collected on ice and immediately cut into 2 mm³ pieces and dissociated to a single-cell suspension using Human Tumor Dissociation Kit and gentleMACS dissociator (Miltenyi Biotec) according to the manufacturer's protocol.

RNA was isolated with the RNeasy Mini Kit (QIAGEN) and single-strand cDNA was made with High Capacity RNA-to-cDNA Kit (Applied Biosystems, Thermo Fisher Scientific). TaqMan gene-expression assays (Applied Biosystems, Thermo Fisher) were used to perform real-time PCR using the StepOnePlus Real-Time PCR system (Applied Biosystems, Thermo Fisher). Gene expression for PERK (Eif2ak3), ATF4 (Atf4), CHOP (Ddit3), and ERO1α (Ero1) were normalized to Gapdh. For immunoblots, cell lysates were prepared in RIPA buffer (Sigma-Aldrich) and probed with PERK (Cell Signaling Technology), ERO1α (Santa Cruz Biotechnology) antibodies, or β-actin (Cell Signaling Technology) loading control.

Fluorochrome-conjugated monoclonal antibodies and respective isotype controls listed in Supplementary Table S1 were purchased from Thermo Fisher (eBioscience). Extracellular stains were performed in PBS supplemented with 2% FBS. IFNγ was probed after 4 to 6 hours of cell restimulation with Cell Stimulation Cocktail (eBioscience) and GolgiPlug Protein Transport Inhibitor (eBioscience). Foxp3/Transcription Factor Intracellular Staining Buffer Set was used according to the manufacturer's protocol (eBioscience). mtROS were measured with MitoSOX Red Mitochondrial Superoxide Indicator (3 μmol/L) loaded at 37°C for 30 minutes in PBS. Extracellular stains were added post dye incubation. For Annexin staining, MitoSOX dye loaded cells were washed and stained for using Annexin V–FITC Apoptosis Detection Kit (eBioscience) according to the manufacturer's protocol. Samples were run directly on a BD Accuri C6 flow cytometer. For FACS sorting of mtROS⁺ and mtROS⁻ cell fractions, samples were stained with MitoSOX Red Mitochondrial Superoxide Indicator and CD8 and sorted on a FACSAria IIu (BD Biosciences). Further analysis was performed post-sort for confirmation of population purity. For RNA ex vivo gene analysis, lymphocytes were isolated via Histopaque gradient (Sigma) and CD8⁺ T-cell Isolation Kit (Miltenyi) was used to obtain >90% purity from spleens and tumors.

MCA-205-OVA (Gift of William Redmond) and MC38 (Kerafast Inc.) cell lines were maintained in DMEM 10% FBS. B16F1-OVA and B16F10 (Gifts of Mark Rubinstein) were maintained in RPMI complete T-cell media. Cell lines were last determined to be Mycoplasma free in March 2017 (MCA-205-OVA), January 2018 (MC38), and 2012 (B16F1-OVA, B16F10). Cell lines were not authenticated in the past year. All growth media were supplemented with Plasmocin mycoplasma prophylactic (Invivogen). All cell lines were passaged three times prior to in vivo tumor inoculation, and a fresh vial of cells was prepared for individual experiments.

Spleens from OT-1 mice were dissociated, and RBC lysis was performed using ACK Lysing (Thermo Fisher Scientific) to achieve a single-cell suspension. T cells were activated and expanded for indicated time points from total splenocytes incubated with 1 μg/mL OVA 257–264 (Invivogen) or hgp100 25–33 (GenScript) peptide in complete T-cell media (200U rhIL2, NCI). Cells were washed, and media were changed after 3 days of initial activation and expansion. For inhibitor treatments, PERK (100 nmol/L, GSK2606414, Tocris) or ERO1α inhibitors (10 μmol/L, EN460, Cayman Chemical) were incubated with splenocytes for 10 minutes prior to introduction of peptide and reintroduced at the day 3 cell split. For central memory T-cell development, IL2 was replaced in T-cell media with rhIL15 (50 ng/mL, Shenandoah) at the day 3 cell split.

Oxygen consumption rate (OCR) was measured in nonbuffered RS media supplemented with HEPES under basal conditions and in response to 1 μmol/L oligomycin, 1.5 μmol/L FCCP, and 2 μmol/L rotenone + 1 μmol/L Antimycin A using the XFe96 Extracellular Flux Analyzer (Seahorse Bioscience). Cell-Tak (Corning) was used for T-cell adherence.

Proteomics were performed by Bioproximity, LLC. Samples were prepared for digestion using the suspension-trapping (S-trap, Protifi) method. Digested peptides were collected by centrifugation. Peptides were eluted with 80% acetonitrile, 5% ammonium hydroxide and lyophilized in a SpeedVac (Thermo Savant) to remove volatile components. Digestion mixtures were analyzed by UHPLC-MS/MS. LC was performed on an Easy-nLC 1000 UHPLC system (Thermo) interfaced to a quadrupole-Orbitrap mass spectrometer (Q-Exactive HF-X, Thermo Fisher) via nano-electrospray ionization using a source with an integrated column heater (Thermo Easy Spray source).

Tandem mass spectra were searched using X!Tandem and Open Mass Spectrometry Search Algorithm (OMSSA), requiring expectation value scores of 0.01 or better to be considered a match. Protein intensity values were calculated using OpenMS to measure the area under the curve of identified peptides. Searches were performed on Amazon Web Services–based cluster compute instances using the Proteome Cluster interface, which builds species- and genus-specific protein sequence libraries monthly from current UniProtKB distributions. Gene annotations were obtained from Ensembl Release 93 database, Mouse genes (v93GRCm38.p6) data set through the BioMart website. Official Gene Ontology categories to annotate mouse proteins based on corresponding gene symbols were used.

For transfer of 7-day expanded OT-1⁺ or PERK KO T cells, C57BL/6 mice were implanted subcutaneously (s.c.) with 2.5 × 10⁵ B16F1-OVA tumor cells, and 5 × 10⁵ T cells were infused via tail vein to 5 Gy irradiated mice after 7 days of tumor growth. For transfer of 7-day expanded Pmel or Pmel inhibitor–treated T cells, C57BL/6 mice were implanted s.c. with 2.5 × 10⁵ B16F10 tumor cells, and 2 × 10⁶ T cells were infused via tail vein to 5 Gy irradiated mice after 7 days of tumor growth. Tumor growth was measured every other day for 3 weeks. For tracking acquisition of mtROS/PD-1⁺ status in tumor antigen–specific CD8⁺ TILs, 2.5 × 10⁵ MCA-205-OVA were implanted s.c. to Ly5.2 mice. Naïve OT-1⁺ T cells were obtained via the Mouse CD8⁺ T-Cell Isolation Kit (Miltenyi Biotec), and 1 × 10⁶ CD8⁺ T cells were transferred via tail vein on day 7 of tumor growth. After 7 days of in vivo expansion, tumor-draining lymph nodes (TDLN) and tumors were harvested. Tumors were processed to single-cell suspension using the Mouse Tumor Dissociation Kit (Miltenyi Biotec) according to the manufacturer's protocol. For combination therapy experiments, 2.5 × 10⁵ MCA-205-OVA were injected s.c. to C57BL/6 mice, and tumors were established for 7 days. PERK inhibitor (50 mg/kg; GSK2606414, GlaxoSmithKline) or vehicle was administered twice daily via oral gavage as a suspension of 0.5% hydroxypropylmethyl cellulose + 0.1% Tween-80 in water. Two hundred micrograms aAnti–PD-1 (RMP1-14) or RatIgG2a isotype control (2A3; Bio X Cell) was administered every 4 days after 12 days of initial tumor growth. For CD8α cell depletion, 100 μg Anti-CD8α (53-6.7, Bio X Cell) was administered three times per week after 5 days of tumor growth.

Previously, we found that ER activation contributed to mitochondrial stimulation in CD4⁺ T cells through inositol 1,4,5-triphosphate receptor (IP(3)R)–mediated Ca²⁺ signaling. Inhibition of IP(3)R shifted T-cell fate and promoted cell persistence in tumors (18). Continuous disruptions in Ca²⁺ homeostasis impair proper protein folding and a high burden of misfolded proteins induces chronic stress on the ER. In the face of irresolvable stress, PERK integrates signals from Ca²⁺ binding protein BiP/grp78 to initiate the terminal UPR through activation of transcription factors ATF4 and CHOP. A key downstream target of ATF4/CHOP is ERO1α. As a facilitator of Redox reactions, excessive ERO1α activity can induce oxidative stress and impair cell survival as cells lose capability to scavenge ROS through antioxidant systems (17, 19, 20). We asked how the ER stress sensor PERK and the terminal UPR were regulated in differentiating T effector cells. We found that gene expression of PERK (Eif2ak3), ATF4 (Atf4), CHOP (Ddit3), and ERO1α (Eor1l) were significantly increased over the course of T-cell activation and differentiation (Fig. 1A–D). PERK protein increased in T cells and peaked after 5 days of expansion. ERO1α protein increased abundance as T cells differentiated (Fig. 1E). PERK gene expression peaked 7 days after activation, but protein expression was reduced after the 5-day time point. Our data point out a discrepancy between gene and protein regulation of PERK in T cells and indicate that further study of how PERK is post-transcriptionally regulated is needed.

To measure the role of PERK in induction of the terminal UPR in T cells, we created T-cell receptor (TCR) transgenic T-cell–specific conditional knockout mice (OT1⁺Lckcre⁺PERK^f/f, PERK KO). We measured genes associated with the chronic PERK response; Atf4, Ddit3, and Ero1l in naïve and 7-day expanded WT and PERK KO T cells. As expected, Eif2ak3, Atf4, Ddit3, and Ero1l were significantly reduced in PERK KO T cells (Fig. 1F–I). In line with these data, PERK and ERO1α proteins were decreased in PERK KO T cells (Fig. 1J). A longer exposure for PERK was performed to convey absence of PERK protein. These data reveal a role for chronic ER stress as a component of differentiating T effector cells.

Increased spare respiratory capacity (SRC) is a property associated with superior antitumor function of T cells (21). The molecular processes that consume T-cell metabolic energy are not well understood. Redox reactions for protein folding and disulfide bond formation in the ER are energy-consumptive processes that integrate ER activation with mitochondrial bioenergetics (17, 19, 22). We measured OCR in WT and PERK KO T cells and found that PERK KO T cells maintained superior SRC ompared with WT cells. To assess a potential contribution of ERO1α to this phenomenon, we treated T cells with a target-specific PERK (PERK I; ref. 23) or ERO1α inhibitor (ERO1α I; ref. 24) across the course of differentiation and expansion. We found that PERK I or ERO1α I–treated T cells preserved metabolic energy compared with WT controls (Fig. 2A–C). Loss of SRC is associated with a reduction in T-cell effector function (25). We measured capacity to secrete IFNγ in the abovementioned T-cell groups, and found that PERK KO, PERK I, or ERO1α I T cells had increased IFNγ secretion compared with WT controls (Fig. 2D). Inhibitor treatments did not significantly reduce cell numbers and experimental T-cell groups exhibited increased viability compared with respective WT controls (Supplementary Fig. S1A and S1B).

Activation of chronic PERK UPR mediated by ATF4/CHOP-ERO1α signaling has previously been implicated to drive protein synthesis, consume cellular energy, and contribute to death through oxidative cell stress (16, 17, 19). To gain insight into how PERK and downstream target ERO1α shape T effector cells, we assayed the proteomes of T effector, PERK KO, and ERO1α I–treated T cells. Shotgun liquid chromatography–tandem mass spectrometry (LC-MS/MS) was used to assess the top 100 proteins with greatest intensity expression in T effectors compared with PERK KO or ERO1α I–treated T cells (Supplementary Table S2). User-defined categories were used to qualify proteins into biologically functional groups. Proteomics revealed protein groups associated with energy and metabolism, redox reactions, and oxidative stress/DNA damage with increased expression in T effectors compared with PERK KO and ERO1α I–treated T-cell groups (Fig. 2E and F). Data are represented as fold increased expression values in T effectors compared with PERK KO or ERO1α I–treated T-cell groups. Of user-defined categories, the greatest overlap in PERK KO and ERO1α-treated T cells was in the “ER transport/cell stress” group (n = 6 proteins). Three of seven of redox-associated proteins overlapped as highly expressed in T effectors compared with both PERK KO and ERO1α I–treated T cells, highlighting lactate dehydrogenases, thioredoxins, and glutathione-S-transferases as protein families of potential relevance. Further analysis to validate these protein pathways to elucidate the role of Redox and oxidative stress in T cells is warranted.

Persistent ERO1α activation is bioenergetically costly to cells, and OXPHOS may be stimulated in efforts to replenish depleted ATP stores (19). We wished to establish a biomarker of energy loss and aging in T cells that may be associated with chronic stress on the ER consistent with pathogenic ERO1α activation. We reasoned that persistent mitochondrial activation through continued OXPHOS would result in accumulation of mtROS in T cells. The live-cell dye MitoSOX Red detects the superoxide anion, a precursor to the majority of cellular ROS, in mitochondrial membranes and is detectable by flow cytometry (26). Under normal physiologic conditions, detoxification of superoxide occurs through superoxide dismutase-mediated conversion to hydrogen peroxide (27). However, excess ROS is associated with disease and aging (28). We measured mtROS in OT-1⁺ T cells over the course of T effector cell differentiation. As naïve T cells lost CD62L expression, a measure of stemness, mtROS accumulated in T-cell mitochondria (Fig. 3A). Human CD8⁺ T cells underwent the same pattern over the course of 3 weeks of in vitro expansion (Fig. 3B).

To determine whether mtROS accumulation was a hallmark of energy loss in T cells, we FACS sorted day 3 and 5 mtROS⁻ and mtROS⁺ CD8⁺ T-cell subsets. The lowest 25% mtROS⁻ and highest 25% mtROS⁺ cells were obtained (Fig. 3C), and SRC among groups was measured. Accumulation of mtROS indicated energy loss in T cells (Fig. 3D). We next measured the capacity for IFNγ secretion among day 5 and 7 expanded mtROS⁻ and mtROS⁺ T-cell groups. We found that IFNγ secretion underwent a similar pattern to SRC among sorted T-cell populations (Fig. 3E). In day 7 cells, we asked whether mtROS⁺ may indicate early-stage apoptosis. We found that ∼8% of CD8⁺ mtROS^high T cells harvested on day 7 were in early-stage apoptosis (Fig. 3F). This was detectable only in day 7 samples, not in cells harvested at earlier time points. Our data show that mtROS can be used as a hallmark of metabolic energy loss and aging in T cells.

Through ERO1α activation, oxidative protein folding in the stressed ER has been linked to generation of cellular ROS and inhibition of cell survival (19). However, the specific effects of the PERK axis on mitochondrial activation have not been assessed. Given the maintenance of SRC and diminished oxidative stress–associated proteomic signature in PERK KO and ERO1α I T cells (Fig. 2), we reasoned that these T-cell groups may show reduced mitochondrial activation and low mtROS accumulation. We found that mtROS was significantly diminished in PERK I, ERO1 I, and PERK KO T cells compared with T effector controls (Fig. 4A). Indeed, in the abovementioned T-cell groups, maintained expression of CD62L was a mark of reduced cell aging (Fig. 4B).

Reacquisition (29) or continued expression of CD62L has been associated with memory T-cell formation (25, 30). We developed memory T cells in vitro with IL15 cytokine conditioning and found that memory T cells expressed reduced mtROS compared with T effector controls (Fig. 4C). These data are consistent with maintenance SRC of memory cells (25). We asked if memory T cells experience reduced activation of the chronic PERK axis. We found that Atf4, Ddit3, and ERO1l genes were reduced in memory T cells compared with T effectors (Fig. 4D). As expected, ERO1α protein expression was increased in T effectors, and ERO1α protein was not highly expressed in memory T cells. Gene expression of Eif2ak3 was not significantly different between T effector and T memory cells, but protein expression of PERK was reduced (Fig. 4D and E). These data, similar to data obtained in Fig. 1, may suggest posttranscriptional regulation of PERK in T cells. Together, the data indicate that ER programs may affect T-cell lineage fate.

Memory T cells promote durable antitumor responses (30). We used adoptive T-cell therapy mouse models to measure how PERK and ERO1α affect T-cell–specific tumor control. Mice bearing B16F1-OVA-expressing tumors were left untreated or treated with OT-1 (T eff) or PERK KO T cells, and tumor control was measured every other day for 3 weeks. OT-1-PERK KO T cells exhibited significantly greater tumor control compared with T effectors (Fig. 5A). Survival was significantly extended in tumor-bearing mice treated with PERK KO T cells (Fig. 5B). Although PERK can be acutely protective, the PERK axis can also impair T-cell–mediated antitumor immunity. To further test how the PERK axis contributes to T-cell–specific tumor control, P-mel T cells conditioned with PERK I or ERO1α I were transferred to mice bearing pathogenic B16F10 melanomas, and tumor growth was monitored for 3 weeks. Both inhibitor treatments of T cells yielded significant increases in tumor control compared with T effectors (Fig. 5C and D).

We next asked how the terminal PERK UPR is expressed among endogenous CD8⁺ TILs. CD8⁺ cells were sorted from spleens and tumors of mice bearing immunogenic MCA-OVA sarcomas, and gene expression was measured between splenocytes and TILs. The chronic PERK signaling axis was significantly upregulated in CD8⁺ TILs compared with splenic-matched controls (Fig. 6A). We next asked whether mtROS accumulated in CD8⁺TILs in mice bearing immunogenic tumors. In multiple tumor models, CD8⁺ TILs accrued mtROS compared with T cells in TDLNs. To expand this finding, we asked whether mtROS was a property of exhausted T cells in tumors. We measured coexpression of mtROS and PD-1 in CD8⁺ T cells from TDLNs and tumors of mice bearing MCA-205-OVA sarcomas or MC-38 colon carcinomas. In both mouse models mtROS⁺ cells were localized to the PD-1⁺ compartment and mtROS expression was significantly increased in TILs compared with TDLNs (Fig. 6B and C). In agreement with gene-expression data, splenocytes from MCA-205-OVA tumor-bearing mice also expressed reduced mtROS/PD-1⁺ CD8⁺ T cells compared with TILs (Supplementary Fig. S2A).

PD-1⁺ TILs comprise a pool of T cells enriched for tumor antigen specificity (31) and tumor-specific T cells home to and proliferate in tumors within 7 days of adoptive transfer (32). We reasoned that metabolically exhausted TILs would be in the tumor antigen–specific T-cell compartment. To address this hypothesis, we transferred naïve OT-1⁺ T cells to mice bearing 7 day-established MCA-205-OVA sarcomas and measured mtROS/PD-1⁺ coexpression among endogenous (CD45.1) and transferred (CD45.2) T cells recovered from tumors after 7 days of expansion (Fig. 6D). Approximately 70% of transferred CD45.2 CD8⁺ TILs coexpressed mtROS/PD-1 compared with ∼20% to 30% in the endogenous polyclonal pool of TILs, and this expression could be visualized in TDLNs (Fig. 6E). Our data agree with reports that PD-1⁺ T cells experience mitochondrial depolarization (33) and shed light on the molecular events within T cells that contribute to mitochondrial exhaustion.

We wished to determine whether mtROS⁺ CD8⁺ TILs could be identified in patients. Bone and soft-tissue sarcomas are immunogenic human tumor types, but have shown poor responsiveness to anti–PD-1 and combination checkpoint blockade therapies (34). We obtained tumor tissue from patients undergoing routine surgical removal of high-grade deep pleomorphic undifferentiated (HGD PU) sarcomas. These patients had not received radiation and had not undergone recent chemotherapy. Dissociation and analysis of CD8⁺ TILs from fresh HGD PU sarcomas showed a CD8⁺ T-cell infiltrate in all tumors and FACS analysis of PBMCs versus autologous TILs showed a significant increase in mtROS/PD-1⁺ CD8⁺ T cells in tumors (Fig. 7A). Our data show that PERK inhibition overcomes mitochondrial exhaustion in T cells in vitro (Fig. 4). MCA-205-OVA tumor–bearing mice were treated with PERK I for 7 days during the time period when T cells become functionally exhausted. Day 14 harvest of TILs showed that PERK I–treated PD-1⁺ CD8⁺ TILs expressed significantly reduced mtROS compared with vehicle-treated controls (Fig. 7B). In line with these data, PERK I–treated mice exhibited increased absolute numbers of CD8⁺ T cells in tumors (Fig. 7C). mtROS⁺ TILs comprise the pool of highly activated TILs measured by CD44 (Supplementary Fig. S2B). A reduction in mtROS⁺ TILs was concordant with reduced CD44⁺ expression among PERK I–treated TILs in both MCA-205-OVA and MC-38 tumor models (Supplementary Fig. S2C). These data suggest that PERK I effects on CD8⁺ TILs were not due to increased activation associated with reduced formation of immunosuppressive TIL populations or reduced tumor growth. In vivo PERK I treatment augmented anti–PD-1 therapy as 5 of 7 combination therapy-treated mice achieved complete response compared with 2 of 7 mice treated with anti–PD-1 therapy alone. Depletions of CD8⁺ cells over the course of therapy demonstrated that CD8⁺ cells were necessary for combination therapy effect (Fig. 7D and E). Combination therapy–treated mice exhibited 100% survival compared with ∼28% survival in the anti–PD-1 therapy condition (Fig. 7F). Together, our data indicate that targeting ER stress in combination with traditional immunotherapy may improve responses in patients.

Works by several groups have brought the ER to the forefront of programming pro- and antitumor immunity (35). ROS in the tumor microenvironment drive MDSC cell death through an ER stress–mediated mechanism. The shortened lifespan of MDSCs promotes their continued expansion in bone marrow (36). In multiple human tumor types, MDSCs in peripheral blood directly correlate with low overall survival rates in patients (37). Similarly, the tumor microenvironment appears to impair tumor-associated dendritic cell (tDC) antigen presentation, and antigen presentation is enhanced by impairing the IRE1α stress pathway in tDCs (38). Lastly, the role of ER stress, and in particular PERK, in augmenting tumor cell development is well defined (39). Together, these data indicate that targeting ER stress pathways in cancer patients may be a formidable strategy to promote antitumor immune function and impair immune suppression and tumor cell growth.

Few reports have focused on the role of ER stress in control of T-cell antitumor immunity, and little attention has been paid to the role of the ER in defining T-cell biology. Our previous work found that inhibition of ER activation improved T-cell mitochondrial function and aided antitumor immunity (18). Our proteomic screen yielded increased expression of multiple proteins associated with ER–mitochondrial cross-talk enriched in T effectors compared with PERK KO T cells. A second report found that inhibition of IRE1α stress signaling led to memory T-cell development in response to acute infection (40). These data agree with our data presented here that uncover a role for ER stress signaling, a response that affects T-cell differentiation and effector cell maturation.

Modulation of ER stress elements alone was able to affect mitochondrial bioenergetics and enhance T-cell–specific tumor control. Mechanistically, acute PERK-specific ER stress is protective to cells through momentary attenuation of protein synthesis mediated by phosphorylation of eukaryotic translation initiation factor 2A (EIF2α; ref. 13). The chronic PERK axis drives protein synthesis, energy consumption, oxidative cell stress, and death through activation of transcription factors ATF4 and CHOP (16, 17). Though PERK gene and protein expression were reduced in our PERK KO T cells, we do not discount the possibility of remnant PERK expression that may have conferred the protective effect of the acute PERK response combined with reduced lethal effects from the terminal UPR. Based on our findings, T cells present a paradigm in which to study the effect of biologically relevant chronic ER stress on cell development and death. Our data show that activation of ERO1α in rapidly dividing T effectors/TILs governs a cell-intrinsic mechanism that drives energy loss and oxidative stress. How the PERK-mediated transcription factors control expression of ERO1α in T cells remains unknown. This question needs to be addressed due to potential implications for cancer immunotherapy (41, 42).

Based on the durability of antitumor immunity associated with memory T cells in vivo (4, 21, 25, 43), we note that modulation of the PERK pathway could promote in vivo T-cell durability and contribute to long-term efficacy of anti–PD-1 therapy. Multiple potential benefits of targeting chronic ER stress have been identified in modulation of pro- and antitumor immunity (35). Development of novel ER stress axis inhibitors that may circumvent previous toxicities associated with inhibitors is on the horizon (44–48). Future research that defines the role of ER stress in T-cell biology is needed. Work that measures how tumor microenvironments drive cell stress signaling that affects T-cell antitumor function is essential.

L.R. Leddy reports receiving a commercial research grant from KCI. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K.E. Hurst, K.A. Lawrence, L.R. Leddy, J.E. Thaxton

Development of methodology: K.E. Hurst, K.A. Lawrence, Z.J. Walton, L.R. Leddy, J.E. Thaxton

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.E. Hurst, M.T. Essman, L.R. Leddy, J.E. Thaxton

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.E. Hurst, J.E. Thaxton

Writing, review, and/or revision of the manuscript: K.E. Hurst, M.T. Essman, J.E. Thaxton

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.E. Hurst, K.A. Lawrence, M.T. Essman, J.E. Thaxton

Study supervision: Z.J. Walton, J.E. Thaxton

Funding sources for this work were from NCI K12 CA157688, ACS IRG-97-219-14, and ACS IRG-16-185-17 to J.E. Thaxton. K12 CA157688 was granted via HHS, NIH, NCI to L. Michael and ACS IRG-97-219-14 and ACS IRG-16-185-17 to J. E. Thaxton.

We are thankful to Gyda Beeson and Dr. Craig C. Beeson for support with metabolic assays, and to Dr. Jeffrey Axten for insight and thoughtful commentary on the manuscript. We thank Dr. B. Arman Aksoy for qualification of protein groups. We are grateful to Drs. Zihai Li, Chrystal Paulos, Michael Lilly, and Gustavo Leone for mentorship within the Hollings Cancer Center.

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.