Abstract

   The role of tumor microenvironment (TME)–associated inadequate protein
   modification and trafficking due to insufficiency in Golgi function,
   leading to Golgi stress, in the regulation of T cell function is
   largely unknown. Here, we show that disruption of Golgi architecture
   under TME stress, identified by the decreased expression of GM130, was
   reverted upon treatment with hydrogen sulfide (H[2]S) donor GYY4137 or
   overexpressing cystathionine β-synthase (CBS), an enzyme involved in
   the biosynthesis of endogenous H[2]S, which also promoted stemness,
   antioxidant capacity, and increased protein translation, mediated in
   part by endoplasmic reticulum–Golgi shuttling of Peroxiredoxin-4. In in
   vivo models of melanoma and lymphoma, antitumor T cells conditioned ex
   vivo with exogenous H[2]S or overexpressing CBS demonstrated superior
   tumor control upon adoptive transfer. Further, T cells with high Golgi
   content exhibited unique metabolic and glycation signatures with
   enhanced antitumor capacity. These data suggest that strategies to
   mitigate Golgi network stress or using Golgi^hi tumor-reactive T cells
   can improve tumor control upon adoptive transfer.
     __________________________________________________________________

   The Golgi apparatus can be targeted in antitumor T cells to enhance
   their immunotherapeutic capacity.

INTRODUCTION

   Adoptive transfer of tumor-reactive T cells has shown promising results
   in metastatic melanoma and advanced B cell malignancies ([70]1, [71]2).
   However, a quantitative or qualitative decrease of the transferred
   antitumor T cells in the tumor-bearing host typically results in tumor
   recurrence, leaving substantial room for improvement ([72]3–[73]6).
   Strategies to improve antitumor T cell function by altering
   mitochondrial bioenergetics ([74]7, [75]8) or its metabolites ([76]9),
   mitigating endoplasmic reticulum (ER) stress ([77]10), or inducing
   protective autophagy ([78]11) are being widely tested. Multiple studies
   have identified oxidative and ER stress within the tumor
   microenvironment (TME) as major contributors of immune cell dysfunction
   and immune evasion ([79]12–[80]16). Similarly, the mammalian Golgi
   apparatus serves important roles in the transport, processing, and
   targeting of proteins and, when under stress, mounts a stress response
   where its unique structure can be fine-tuned to adapt different Golgi
   functions to specific cellular needs ([81]17, [82]18). While the
   synchronized activity of these cellular organelles is being
   increasingly recognized for maintaining quality control and ensuring
   cell survival and function ([83]19), specifics of Golgi dynamics in the
   TME and the role of the Golgi stress response in shaping T cell
   function have thus far been understudied.

   Similar to our recent study where carbon monoxide–mediated transient
   activation of the ER stress PERK pathway led to increased mitochondrial
   biogenesis and reprogramming of antitumor T cells to effectively treat
   established tumors upon adoptive T cell transfer ([84]11), transient
   activation of Golgi stress mediated by monensin was shown to stimulate
   the reverse trans-sulfuration pathway via cystathionine γ-lyase (CSE)
   (the biosynthetic enzyme for cysteine and an important regulator of
   redox homeostasis) to mitigate the toxicity associated with cysteine
   deprivation in Huntington’s disease ([85]20). Given that CSE-,
   cystathionine β-synthase (CBS)–, and 3-mercaptopyruvate
   sulfurtransferase (3-MST)–mediated secretion of hydrogen sulfide
   (H[2]S), an endogenous signaling gaseous transmitter that also
   mitigates Golgi stress ([86]17), has been shown to regulate the immune
   response in mammals ([87]21), we hypothesized that H[2]S may establish
   a reduced stress state in antitumor T cells and that increasing H[2]S
   could potentiate the antitumor T cell response by mitigating ER and
   Golgi stress.

   Here, we show that T cell intrinsic H[2]S signaling supports overall
   protein translation and improves T cell effector function by reducing
   ER and Golgi stress. The levels of H[2]S or CBS also inversely
   correlated with exhaustion, and replenishing H[2]S exogenously during T
   cell activation or expansion led to an increase in the central memory
   (Tcm) phenotype by engaging the NAD^+-Sirt1-Foxo1 axis. Proteomics
   analysis highlighted that increased abundance of free thiols in
   Peroxiredoxin-4 (Prdx4) was also in part responsible for the
   H[2]S-mediated Tcm phenotype. ER-localized Prdx4 was found to
   translocate to the Golgi under conditions of oxidative stress. Further,
   delineating T cells based on Golgi content highlighted that the T cell
   subsets with high Golgi content (Golgi^hi subset) exhibit long-term
   tumor control upon adoptive transfer. Further, human CD19 chimeric
   antigen receptor (CAR)–T cells overexpressing CBS or sorted for
   Golgi^hi significantly increased the survival of human lymphoma
   xenografted mice compared to mice treated with Golgi^lo CAR-T cells.
   Collectively, these results not only provide insight into the important
   role of endogenous H[2]S in regulating the T cell immune response but
   also highlight the Golgi network as a previously unidentified
   therapeutic target for enhancing the efficacy of immunotherapy for
   cancer.

RESULTS

H[2]S promotes generation of central memory antitumor T cells

   Given the limited knowledge on the role of H[2]S in primary T cells, we
   determined the kinetics of endogenous H[2]S production in antitumor T
   cells during T cell receptor (TCR)–mediated activation. Melanoma
   epitope gp100-reactive CD8^+ T cells (from Pmel-transgenic mouse
   spleen) were stained with Cell Trace Violet (CTV) proliferation dye and
   activated in vitro with gp100 cognate antigen. After 3 days of
   activation, the T cells were stained with WSP-1 dye to quantify
   intracellular levels of H[2]S production in different generations. T
   cells within the first generation of proliferation (G1) displayed
   significantly increased levels of intracellular H[2]S compared to the
   naïve T cells ([88]Fig. 1A). However, with progressive T cell
   proliferation (G2 to G4), intracellular H[2]S returned to basal levels.
   Corroborating these findings, mRNA samples collected from activated T
   cells before activation (time 0), and at 24, 48, and 72 hours
   demonstrated significant up-regulation of Cbs (one of the primary
   enzymes responsible for H[2]S production) 24 hours after activation,
   followed by a return to basal levels of expression by 72 hours
   ([89]Fig. 1B).

Fig. 1. H[2]S promotes generation of central memory (Tcm) antitumor T cells.

   [90]Fig. 1.
   [91]Open in a new tab

   (A) Splenocytes from Pmel mice were stained with CTV dye and stimulated
   with gp100 antigen (1 μg/ml) + IL-2 (100 IU/ml) followed by staining
   with WSP-1 dye to quantify H[2]S production using FACS (n = 4
   independent samples). (B) RT-PCR analysis performed on Pmel CD8^+ T
   cells at various time points following activation to quantify mRNA
   levels of CBS (n = 4). (C and D) Pmel T cells treated with 0.5 mM
   GYY4137 for 3 days. (C) FACS analysis for expression of CD62L and CD44
   (n = 4). (D) FACS analysis for expression of Tcf1/7 and Sca-1 (n = 4).
   (E and F) WT and Cbs-KO splenocytes activated with anti-CD3 and
   anti-CD28 and expanded with addition of H[2]S to the Cbs-KO cells. (E)
   Analysis of intracellular H[2]S by FACS using WSP-1 dye and (F) FACS
   analysis of relative percentage of CM to EM cells (n = 3). (G and H)
   Activated Pmel T cells transduced with lentiviral vector to overexpress
   CBS enzyme (Cbs^td) and on day 7 analyzed for (G) intracellular H[2]S
   production using WSP-1 dye and (H) frequency of CM (CD62L^+ CD44^+) and
   Sca-1 expression (n = 4). (I to L) B16-F10 tumors implanted in C57BL/6
   mice and resected on day 14 to culture in vitro. Supernatant from the
   cultured tumor cells was extracted and transferred to cultures
   containing Pmel T cells and gp100 antigen with or without 0.5 mM
   GYY4137. T cells cultured under the conditions of the exhaustion assay
   were analyzed for expression of (J) CBS by FACS and (K) intracellular
   H[2]S using WSP-1 dye and quantified by FACS (n = 4). (L) T cells from
   tumor supernatant exhaustion assay were also analyzed for the
   expression of CD62L and CD44 via FACS (n = 4). All data shown represent
   the mean ± SEM and were analyzed by two-sided Student’s t test or
   one-way analysis of variance (ANOVA). ns, P > 0.05; *P ≤ 0.05; **P ≤
   0.01; ***P ≤ 0.001; ****P ≤ 0.0001.

   We next determined whether restoring H[2]S levels in activated T cells
   would alter their phenotype. To achieve steady-state levels of H[2]S in
   the T cell culture media, we elected to use the water-soluble H[2]S
   donor GYY4137 that slowly releases sustained levels of H[2]S up to 7
   days in culture ([92]22). Given the cytotoxic nature of H[2]S at high
   concentrations, we selected 0.5 mM as the optimal dose of GYY4137 (fig.
   S1A) in all experiments where Pmel T cells were activated with cognate
   antigen gp100 for 7 days in presence of rIL2 (50 IU/ml) and gated on
   the CD8^+ fraction for analysis (fig. S1B). We observed that T cells
   generated in the presence of H[2]S donor had a significant increase in
   the Tcm phenotype (as determined by CD62L^+CD44^+ coexpression)
   compared to those expanded with interleukin-2 (IL-2) alone ([93]Fig.
   1C), and this increase in Tcm phenotype was consistently maintained
   over 6 days (fig. S1C). Further analysis revealed that H[2]S treatment
   significantly enhanced the expression of Tcf7 and Sca1, markers related
   to the T cell stemness phenotype ([94]Fig. 1D). A similar
   dose-dependent increase in the Tcm phenotype was observed when the
   fast-release H[2]S donor NaHS was used, added to the culture media
   daily (fig. S1D), supporting that H[2]S plays a role in generation of
   Tcm phenotype. A quantitative polymerase chain reaction (qPCR) analysis
   of the activated T cells that were FACS (fluorescence-activated cell
   sorting) sorted based on effector (CD62L^+CD44^+) and central memory
   (CD62L^+CD44^+) fraction showed increased expression of all three
   H[2]S-producing enzymes (CBS, CSE, and 3-MST) in the Tcm fraction, with
   CBS being most significantly up-regulated (fig. S1E). Treatment of T
   cells with IL-15 and IL-6, cytokines known to induce the Tcm phenotype
   in T cells ([95]23) and play a key role in memory formation ([96]24),
   respectively, resulted in significantly increased expression of both
   CBS and CSE (fig. S1F).

   TCR-activated T cells from mice lacking CBS expression (Cbs-KO, fig.
   S1G) not only showed reduced H[2]S accumulation ([97]Fig. 1E) but also
   exhibited reduced Tcm fraction when programmed in the presence of IL-15
   (fig. S1H) and inferior ability to persistent following adoptive
   transfer compared to wild-type T cells (fig. S1I). However, adding
   H[2]S to the Cbs-KO T cells led to a restoration of the Tcm fraction
   ([98]Fig. 1F). These data confirm that H[2]S levels play an important
   role in maintaining the Tcm phenotype. Overexpression of CBS (fig. S1J)
   in activated CD8^+ T cells also resulted in a significantly enhanced
   H[2]S accumulation ([99]Fig. 1G) and a concomitant increase in the Tcm
   population and Sca1 expression ([100]Fig. 1H).

   The solid TME is characterized as being highly immunosuppressive,
   leading to T cell exhaustion. Thus, we next used an in vitro model of
   TME-induced T cell exhaustion using supernatants collected from B16-F10
   murine melanoma cells along with chronic antigen stimulation ([101]25).
   Pmel CD8^+ T cells were activated and cultured under optimal culture
   conditions, with tumor supernatant plus repeated stimulation with
   gp100, or with tumor supernatant plus repeated stimulation with gp100
   plus H[2]S donor ([102]Fig. 1I). This model was used to determine the
   impact of T cell exhaustion on CBS expression and H[2]S production by
   gating Pmel T cells on TIM3^+PD1^+ (terminally exhausted) and
   TIM3^−PD1^− cells. We found that terminally exhausted T cells expressed
   significantly less CBS ([103]Fig. 1J) with less H[2]S production
   ([104]Fig. 1K). Addition of H[2]S to the culture not only enabled
   activated T cells to preserve the Tcm phenotype in the TME exhaustion
   assay ([105]Fig. 1L) but also decreased expression of T cell exhaustion
   markers PD1, TIM3, Lag3, and CD38 (fig. S1K), suggesting a role for
   H[2]S in combating TME-induced T cell exhaustion. These data suggest
   that endogenous H[2]S levels play an important role in regulating T
   cell effector versus memory and exhaustion phenotypes.

H[2]S supports T cell effector function and protein translation

   Next, we determined the effect of H[2]S treatment on the transcriptomic
   profile of antitumor T cells. Pmel CD8^+ T cells were activated and
   expanded in the presence or absence of the H[2]S donor before RNA
   sequencing (RNA-seq). Principal components analysis (PCA) demonstrated
   a distinct transcriptomic profile for T cells treated with the H[2]S
   donor as compared to T cells activated under standard culture
   conditions ([106]Fig. 2A). Of note, several of the most significantly
   up-regulated genes were genes involved in chemokine receptor activity
   and leukocyte migration, including CCR2, CCR5, CXCR2, GRP15, and CD177
   ([107]Fig. 2, B and C). The major pathways significantly up-regulated
   in H[2]S-treated T cells included pathways related to immune receptor
   activity, T cell signaling, and cytokine activity ([108]Fig. 2, C and
   D, and fig. S2A).

Fig. 2. H[2]S supports T cell effector function and protein translation.

   [109]Fig. 2.
   [110]Open in a new tab

   (A to D) Melanoma epitope gp100 TCR reactive CD8^+ T cells were
   activated for 3 days with gp100 and IL-2 with or without 0.5 mM
   GYY4137. On day 3, the cells were collected for RNA-seq analysis (n = 3
   independent samples). (A) Principal components analysis (PCA) of
   control versus H2S-treated samples. (B) Heatmap displaying
   significantly up-regulated and down-regulated gene groups. (C) Volcano
   plot demonstrating significantly up-regulated and down-regulated genes
   using false discovery rate (FDR)–adjusted P value < 0.05. (D) Pathway
   enrichment analysis of significantly up-regulated genes. (E) Schematic
   of experimental design for cytokine and translation assays. (F) Pmel
   CD8^+ T cells were activated for 3 days with gp100 and IL-2 with or
   without 0.5 mM GYY4137. On day 3, the cells were washed and
   restimulated with gp100 for 4 hours, followed by measurement of
   cytokine production via FACS. (G) Active protein translation was
   assessed using the Click-iT Plus OPP Alexa Fluor 647 Protein Synthesis
   Assay Kit (Thermo Fisher Scientific) (n = 3 independent samples) in
   Pmel T cells after restimulation (n = 3 independent samples). (H)
   Levels of EIF2a, phosphorylated 4EBP1, and phosphorylated rS6 evaluated
   via FACS in Pmel T cells after restimulation (n = 4 independent
   samples). (I) Pmel CD8^+ T cells were activated for 2 days with gp100
   and IL-2 followed by transduction with either control lentivirus or
   CBS-expressing lentivirus. On day 5, the cells were washed and
   re-stimulated with gp100 for 4 hours. Quantification of protein
   translation assessed using Click-iT Plus OPP Alexa Fluor 647 Protein
   Synthesis Assay Kit and quantification intracellular cytokines assessed
   using FACS (n = 3 independent samples). MFI represents mean
   fluorescence intensity. All data shown represent the mean ± SEM and
   were analyzed by two-sided Student’s t test or one-way ANOVA, unless
   otherwise specified. ns, P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001;
   ****P ≤ 0.0001.

   We next sought to determine the impact of H[2]S on T cell effector
   function. Pmel CD8^+ T cells were activated and expanded with standard
   culture conditions or with the addition of the H[2]S donor and
   subsequently restimulated with gp100 cognate antigen ([111]Fig. 2E).
   Pmel T cells treated with the H[2]S donor displayed enhanced secretion
   of effector cytokines tumor necrosis factor–α (TNFα) and interferon-γ
   (IFN-γ), as well as increased levels of cytolytic molecule granzyme B
   and CD107 expression (a marker of degranulation) ([112]Fig. 2F)
   ([113]26). Given the importance of protein translation in maintaining a
   robust antitumor response and the repression of T cell translation that
   occurs in the TME ([114]27, [115]28), we determined whether H[2]S
   influences protein translation. Notably, T cells treated with the H[2]S
   donor demonstrated significantly enhanced up-regulation of overall
   protein translation as measured by O-propargyl-puromycin (OPP)
   incorporation into the nascent polypeptide chain during translation
   upon restimulation with cognate antigen ([116]Fig. 2G) ([117]29).
   H[2]S-treated T cells also displayed increased total levels of EIF2α
   and phosphorylated levels of S6 ribosomal protein, essential factors
   regulating the rate-limiting step of protein synthesis, along with
   increased phosphorylation of 4EBP1 (one of the key negative regulators
   of protein synthesis which is inhibited by phosphorylation) ([118]Fig.
   2H). Similarly, Pmel CD8^+ T cells transduced with CBS overexpressing
   vector displayed increased cytokine production and overall protein
   translation when reexposed to cognate tumor antigen ([119]Fig. 2I),
   demonstrating that enhancing H[2]S signaling increases effector
   function in CD8^+ T cells.

   It has previously been shown that in CD4^+ T cells, H[2]S promotes
   Tet-mediated Foxp3 demethylation to drive regulatory T cell (Treg
   differentiation) ([120]30). However, recent studies have shown that
   reduced Tet2 expression results in the improved function of CD19
   CAR-engineered T cells ([121]31). We observed that in CD8^+ T cells,
   H[2]S treatment results in a decrease of Tet activity (fig. S2B). These
   findings suggest distinct roles of H[2]S signaling in CD8^+ T cell
   programming.

H[2]S enhances antitumor efficacy of T cells in vivo

   Given the Tcm phenotype, reduced exhaustion, and enhanced effector
   function of T cells treated with H[2]S in vitro, we next assessed the
   utility of H[2]S-based strategies in tumor control. First, we used Pmel
   T cells expanded with or without the H[2]S donor that were subsequently
   transferred to B16-F10 murine melanoma-bearing immunocompetent C57BL/6
   mice (fig. S3A, schematic). The H[2]S-treated Pmel T cells displayed a
   superior ability to control tumor growth and prolong overall survival
   compared to control Pmel T cells ([122]Fig. 3A), which also correlated
   with increased persistence (right panel). Enhanced CD62L^+CD44^+ Tcm
   phenotype and Sca1 expression was also noted in the group that received
   H[2]S-pretreated T cells (fig. S3B). Next, to determine whether H[2]S
   treatment would reprogram tumor-infiltrating lymphocytes (TILs) and
   render them more effective, we obtained TILs from B16-F10 tumor
   engrafted in Pmel mice and then expanded them with or without the H[2]S
   donor (fig. S3C, schematic). Following adoptive transfer into
   tumor-bearing mice, we observed that mice receiving H[2]S-treated TILs
   exhibited significant reduction in tumor growth and extended overall
   survival ([123]Fig. 3B), which also correlated with its enhanced
   persistence.

Fig. 3. H[2]S enhances the antitumor efficacy of T cells in vivo.

   [124]Fig. 3.
   [125]Open in a new tab

   (A to D) Indicated cells were expanded ex vivo and transferred into
   mice bearing subcutaneously inoculated tumors. Tumor size was measured
   three times per week until end point size of 400 mm^2 is reached. All
   tumor control experiments were repeated twice. (A) B16-F10 melanoma
   tumor-bearing C57BL/6 mice treated with melanoma epitope gp100 TCR
   reactive CD8^+ T cells cultured with IL-2 alone or IL-2 + 0.5 mM
   GYY4137 (n = 10 mice per group). Frequency of Pmel T cells in
   peripheral blood at day 21 after transfer assessed by FACS. (B) B16F10
   tumor-bearing C57BL/6 mice treated with TILs isolated from B16-F10
   tumors grown subcutaneously on C57BL/6 mice and expanded with IL-2
   alone or IL-2 + 0.5 mM GYY4137 (n = 10 mice per group). Frequency of
   transferred T cells in peripheral blood at day 10 after transfer
   assessed by FACS. (C) Raji tumor-bearing NSG mice treated with human
   CD19 CAR-T cells cultured with IL-2 alone or IL-2 + 0.5 mM GYY4137 (n =
   10 mice per group). CAR-T cells were injected 3 days following Raji
   cell inoculation. Frequency of CD34^+ transferred CAR-T cells in
   peripheral blood at day 21 after transfer assessed by FACS. (D) Human
   PBMCs were transduced with either CD34-CD19-CAR-T construct or
   Cbs-CD34-CD19-CAR-T construct. Transduced CAR-T cells were sorted on
   CD34^+ cells and adoptively transferred into CD19^+ Raji tumor-bearing
   NSG mice 6 days after Raji cell inoculation (n = 10 mice per group).
   Frequency of CD34^+ transferred CAR-T cells in peripheral blood at day
   21 after transfer assessed by FACS. For all survival outcomes,
   Kaplan-Meier curves were used to display the results. Median survival
   time and corresponding 95% confidence interval were calculated for each
   experimental condition. A log-rank test was used to compare the
   outcomes across experimental conditions. ns, P > 0.05; *P ≤ 0.05; **P ≤
   0.01; ***P ≤ 0.001.

   Given the demonstrated efficacy of using H[2]S to enhance adoptive cell
   therapy (ACT) protocols in murine models of melanoma, we determined
   whether this strategy would be equally efficacious in controlling human
   tumors. Thus, human B cell lymphoma Raji cells engrafted in NSG mice
   were treated with human CD19 CAR-transduced T cells that were generated
   in the presence or absence of H[2]S (fig. S3D, schematic). As expected,
   mice receiving CD19 CAR-T cells expanded with H[2]S had a significant
   reduction in tumor growth and an increase in overall survival
   ([126]Fig. 3C), as well as increased persistence (right panel). To
   further support the translational applicability of these findings, we
   designed a CAR-T construct to incorporate Cbs. These
   CD19-Cbs–engineered CAR-T cells (Cbs^td CAR-T cells) also exhibited
   better tumor control as compared to the CD19 CAR-engineered T cells
   ([127]Fig. 3D) with increased persistence (right panel). This suggests
   that antitumor T cells with increased H[2]S signaling exhibit better
   persistence in vivo and CAR-T cells engineered to express CBS can
   potentiate tumor control.

H[2]S alters metabolic profile and enhances the mitochondrial function of T
cells

   Next, we sought to characterize the metabolic status of antitumor T
   cells treated with exogenous H[2]S. Comprehensive metabolomics analysis
   revealed that Pmel CD8^+ T cells treated with the H[2]S donor possessed
   a distinct profile of metabolites relating to enrichment of several
   metabolic pathways ([128]Fig. 4A and fig. S4A). Of note, several of the
   top pathways affected by H[2]S treatment involved metabolic pathways
   that are known to be critical for antitumor immunity, including serine,
   vitamin B6, and nicotinamide metabolism ([129]Fig. 4A) ([130]9,
   [131]32). Specifically, the pathway of nicotinate and nicotinamide
   metabolism was enriched, with increased levels of NAD^+ and decreased
   levels of nicotinamide in H[2]S-treated T cells ([132]Fig. 4, B and C).
   This corresponded to increased expression of NAD(P)H quinone
   oxidoreductase 1 (NQO1) identified in the RNA-seq analysis, an
   antioxidative enzyme that modulates the differentiation of T[H]17 cells
   by regulating reactive oxygen species (ROS) levels ([133]33). These
   data imply that H[2]S renders increased antioxidant capacity to T
   cells, which may contribute to increase persistence and Tcm phenotype
   (as reported earlier) ([134]2, [135]34).

Fig. 4. H[2]S alters metabolic profile and enhances the mitochondrial
function of T cells.

   [136]Fig. 4.
   [137]Open in a new tab

   (A to C) Pmel T cells activated for 3 days with gp100 and IL-2 with or
   without 0.5 mM GYY4137 and collected for comprehensive metabolomics
   analysis. (A) Pathway enrichment analysis based on significantly
   different metabolites. (B) Volcano plot showing significantly
   up-regulated and down-regulated metabolites based on –log10(P) > 1. (C)
   NAD^+ and nicotinamide quantification from metabolomics dataset. (D)
   SIRT1 deacetylase activity in activated Pmel T cells (means of
   normalized OD values). (E and F) Pmel T cells were activated with gp100
   for 3 days in the presence of vehicle, GYY4137 (0.5 mM), or GYY4137 +
   sirt1 inhibitor EX527 (10 μM). (E) Nuclear and cytoplasmic protein
   fractions were isolated and Western blot analysis was performed to
   determine the relative expression of nuclear and cytoplasmic Foxo1 (n =
   3). (F) Expression of CD62L and CD44 analyzed by FACS (n = 3). (G)
   Splenocytes from wild-type and Cbs-knockout mice +/- GYY4137 activated
   with anti-CD3 and anti-CD28 antibodies followed by intranuclear
   staining for Foxo1. (H) Pmel T cells activated and expanded for
   metabolic flux analysis. Quantification of oxygen consumption rate
   (OCR), spare respiratory capacity (SRC), maximal respiration, and basal
   respiration (n = 10 experimental replicates representative of three
   separate experiments). (I) Pmel T cells activated and expanded to day 5
   for characterization of mitochondria using FACS, quantified by MFI
   values for mitoFM Deep Red, TMRM, and mitoSOX dyes (n = 4). (J)
   Representative confocal microscopy images of mitochondria from control
   or H[2]S-treated T cells. (K) Human CD19 CAR-T cells expanded for 3
   days with IL-2 with or without 0.5 mM GYY4137 and collected for
   metabolic flux analysis. Shown is the quantification of mean OCR, SRC,
   maximal respiration, and basal respiration. Data shown represent the
   mean ± SEM and were analyzed by two-sided Student’s t test or one-way
   ANOVA. ns, P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤
   0.0001.

   Of note, an increase in levels of NAD^+ has been linked to the enhanced
   antitumor efficacy of CD8^+ T cells through its key role as a substrate
   for Sirt1 ([138]9). We found that the H[2]S-treated T cells were also
   characterized by increased Sirt1 activity ([139]Fig. 4D). Previous
   studies have shown that the transcription factor Foxo1 regulates the
   Tcm phenotype ([140]35) and that NAD^+-dependent Sirt1 is responsible
   for its deacetylation and activity ([141]9). Recent studies have also
   shown that Foxo1 overexpression promotes a stem-like phenotype in CAR-T
   cells derived from either healthy human donors or patients, which
   correlates with improved mitochondrial fitness, persistence, and
   therapeutic efficacy in vivo ([142]36, [143]37). Thus, we probed
   whether H[2]S treatment increased nuclear translocation of Foxo1 via
   enhanced Sirt1 activity in T cells. We observed that H[2]S treatment
   led to increased nuclear localization of Foxo1 and that Sirt1 inhibitor
   Ex527 treatment neutralized H[2]S-mediated nuclear localization of
   Foxo1 ([144]Fig. 4E) and Tcm phenotype ([145]Fig. 4F). Expression of
   Foxo1 was reduced in Cbs-KO T cells and was restored upon H[2]S
   treatment ([146]Fig. 4G). These data indicate that the H[2]S-mediated
   increase in Tcm phenotype may be dependent on Sirt1-mediated
   enhancement of nuclear localization of Foxo1.

   Next, given the distinct metabolite profile observed in H[2]S-treated T
   cells, we analyzed the mitochondrial fitness of the T cells generated
   in the presence of H[2]S. We found that Pmel T cells activated in the
   presence of the H[2]S donor did not alter basal respiration but
   resulted in a significant increase in both maximal respiration and
   spare respiratory capacity (SRC) ([147]Fig. 4H). Further, the Pmel T
   cells treated with the H[2]S donor had increased overall mitochondrial
   mass as measured by mitoFM staining, decreased mitochondrial membrane
   potential as measured by tetramethylrhodamine methyl ester (TMRM)
   staining, and decreased mitochondrial ROS as measured by mitoSOX
   staining ([148]Fig. 4I). Notable differences in mitochondrial
   organization and morphology were also observed with H[2]S treatment,
   including more extended and dispersed mitochondria in the control group
   compared to more compact and continuous mitochondria in the
   H[2]S-treated cells ([149]Fig. 4J). Similarly, the human CD19 CAR-T
   cells treated with the H[2]S donor displayed a significant increase in
   maximal respiration and SRC ([150]Fig. 4K). Given the previously
   demonstrated importance of low mitochondrial membrane potential and low
   mitochondrial ROS in metabolically fit antitumor T cells with enhanced
   antioxidant capacity and stemness ([151]8, [152]34), these findings
   suggest an overall increase in mitochondrial fitness in the presence of
   H[2]S.

H[2]S reduces oxidative stress and Golgi-ER network stress in antitumor T
cells

   Reducing T cell intrinsic ROS generation has been shown to alleviate T
   cell exhaustion and improve the efficacy of T cell immunotherapy. Given
   that H[2]S has been identified to up-regulate cellular antioxidant
   defense mechanisms ([153]38–[154]40), we next sought to determine its
   ability to combat oxidative stress in antitumor T cells. Using hydrogen
   peroxide (H[2]O[2]) to induce oxidative stress in vitro, Pmel CD8^+ T
   cells treated with the H[2]S donor exhibited a significant decrease in
   apoptotic cell death ([155]Fig. 5A). H[2]S treatment also resulted in
   increased expression of glutamate cysteine ligase (GCL) holoenzyme
   catalytic subunit (GCLC) and modifier subunit (GCLM) ([156]Fig. 5B),
   which catalyzes the rate-limiting step in the formation of the cellular
   antioxidant glutathione (GSH) to maintain cellular GSH homeostasis. In
   line with this observation, we noted that H[2]S-treated Pmel T cells
   and CD19 CAR-T cells overexpressing CBS exhibited an increase in
   overall cell-surface thiol (-SH) expression and intracellular
   glutathione (iGSH) ([157]Fig. 5C and fig. S5A), correlating to their
   increased resistance to oxidative stress–induced cell death. Given the
   observed increase in total surface thiols, we next used a proteomic
   approach to identify proteins with cysteine thiols sensitive to H[2]S
   treatment (table S1). Our proteomics data revealed several key proteins
   involved in regulating ER stress that display enhanced abundance or an
   increase in the extent of free thiols in response to H[2]S treatment
   (table S2), including the key ER localized proteins Prdx4, ER
   oxidoreductin 1 (Ero1), mesencephalic astrocyte-derived neurotrophic
   factor 1 (MANF1), and the 60S ribosomal protein L7a (Rpl7a) ([158]Fig.
   5D).

Fig. 5. H[2]S reduces oxidative stress and Golgi-ER network stress in
antitumor T cells.

   [159]Fig. 5.
   [160]Open in a new tab

   (A to F) Pmel T cells activated for 3 days with or without 0.5 mM
   GYY4137. (A) Cells treated for 12 hours with H[2]O[2] (50 mM). FACS
   analysis quantifying 7-AAD^+annexin V^+ and 7-AAD^−annexinV^−
   populations (n = 3). (B) mRNA levels of GCLM and GCLC. (C) Measurements
   of surface thiols by Alexa Fluor 488 C[5] maleimide (n = 3) and
   intracellular glutathione by ThiolTracker Violet (n = 4). (D)
   Proteomics analysis quantifying thiol modifications on cysteine
   residues with statistically significant results based on one-sample t
   test with FDR < 0.5 with a median of three normalized ratios using
   log2(fold change treated versus control) (n = 3). (E) Fold increase in
   MFI for phosphorylated-PERK, phosphorylated-IRE1a, and ATF4 after
   12-hour incubation with 50 mM H[2]O[2] (n = 3). (F) FACS quantification
   of cells expressing high phosphorylated-PERK levels after exhaustion
   assay (n = 2). (G) Percentage of cells expressing high
   phosphorylated-PERK in Pmel cells overexpressing CBS compared to Pmel
   cells transduced with control lentivirus (n = 3). (H) Fold change in
   GM130 MFI in Pmel cells using exhaustion assay (n = 3). (I) Cell
   viability and mitochondrial ROS following monensin treatment (1 mM).
   (J) MFI values for GM130 and PD1 in TILs isolated from B16-F10 tumors
   and expanded +/- GYY4137 (n = 3). (K) GM130 MFI values in Pmel T cells
   overexpressing CBS compared to control cells after transfer into
   tumor-bearing mice and TIL isolation after 7 days (n = 4). (L) Pmel T
   cells cultured with B16-F10 supernatant and gp100, followed by FACS
   sorting for PD1^+TIM3^+ and PD1^−TIM3^− cells. Sorted cells stained
   with anti-Giantin antibody and analyzed for Golgi area via ImageStream.
   (M) FACS analysis of GM130 expression in PD1^+TIM3^+ and PD1^−TIM3^−
   populations. (N) Pmel cells treated with 1 mM monensin and analyzed for
   intracellular ROS (DCFDA). (O) FACS analysis of GM130 expression in
   Pmel cells treated with 50 mM H[2]O[2] (6 hours). Data represent mean ±
   SEM. Analyzed by two-sided Student’s t test or one-way ANOVA. ns, P >
   0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001.

   While Prdx4 supports redox homeostasis by metabolizing H[2]O[2] in the
   ER, its loss leads to oxidative stress and toxicity ([161]41). Ero1 is
   an oxidoreductase enzyme that catalyzes the formation and isomerization
   of protein disulfide bonds in the ER, generating H[2]O[2] in the
   process ([162]42). Given the ER-Golgi intricate association and that
   Golgi stress response has been shown to reprogram cysteine metabolism
   ([163]20), we hypothesized that H[2]S levels regulate oxidative stress
   within the ER and Golgi network. We observed that markers of ER stress,
   including phospho-PERK, phospho-IRE1a, and ATF4, were increased when
   control Pmel T cells were exposed to oxidative stress; however, this
   increase was mitigated in H[2]S-treated T cells ([164]Fig. 5E). The
   Cbs-KO T cells also exhibited increased ER stress, measured by
   increased pPERK, which was reduced in the presence of H[2]S (fig. S5B).
   We also noted a significant increase in ER stress in T cells exposed to
   the in vitro TME exhaustion model, which was mitigated with H[2]S
   treatment ([165]Fig. 5F). Further, overexpression of CBS resulted in a
   decrease in percentage of T cells experiencing high levels of ER stress
   (phospho-PERK^hi) ([166]Fig. 5G).

   Golgi stress has recently been identified as an important mediator of
   redox imbalance in human cells ([167]43), and H[2]S signaling has also
   been shown to be protective against Golgi stress ([168]17). To track
   Golgi stress, we chose to determine the expression of GM130, a Golgi
   tethered protein that has been shown to control Golgi morphology in
   response to changes in cellular conditions ([169]44). Previous studies
   have shown that Purkinje neurons in mice that lack the GM130 exhibit
   Golgi fragmentation and decreased secretory trafficking, leading to
   ataxia and cell death ([170]45). Thus, we hypothesize that strategies
   that maintain Golgi homeostasis and avoid Golgi disruption in T cells
   would also result in increased persistence and preserve effector
   functions in the TME. Induction of T cell exhaustion using the TME
   exhaustion model resulted in Golgi stress, as measured by a significant
   decrease in GM130 expression ([171]Fig. 5H). This decrease in GM130
   expression was mitigated in the presence of the H[2]S donor ([172]Fig.
   5H). To further determine the impact of Golgi stress on T cells, we
   used a known Golgi stressor, monensin, to induce Golgi stress in Pmel T
   cells in the presence or absence of H[2]S. Monensin treatment induced
   cell death and cellular accumulation of mitochondrial ROS, which was
   partially mitigated with H[2]S treatment ([173]Fig. 5I). TILs isolated
   from B16-F10 tumors and expanded under normal control conditions
   (without H[2]S) or with H[2]S revealed significantly higher GM130
   expression and lower PD1 expression in the H[2]S-treated group
   ([174]Fig. 5J). These results demonstrate that Golgi stress is induced
   in the TME and that H[2]S can be used to reprogram TILs to reduce Golgi
   stress, correlating to a decrease in T cell exhaustion. Similarly, Pmel
   T cells transduced with CBS and adoptively transferred into
   tumor-bearing mice maintained significantly higher GM130 expression
   when isolated from the tumors ([175]Fig. 5K). Monensin did not induce
   ER stress, nor did thapsigargin induced Golgi stress (fig. S5C),
   highlighting the exclusivity of organelle stress and need for
   mitigating them is important to obtain a “stress-free” T cell.

   Golgi dispersion under conditions of stress has been identified as a
   key feature of Golgi dysfunction ([176]44). Thus, we induced Golgi
   stress with monensin and characterized Golgi dispersion with
   ImageStream analysis, showing that Golgi dispersion significantly
   increased when T cells were treated with monensin (fig. S5D). Next, we
   induced T cell exhaustion using the previously described in vitro TME
   model and characterized Golgi dispersion in terminally exhausted T
   cells (PD1^+TIM3^+) compared to healthy T cells (PD1^−TIM3^−). Like
   monensin treatment, we observed a significant increase in Golgi
   dispersion in PD1^+TIM3^+ T cells compared to PD1^−TIM3^− ([177]Fig.
   5L). This correlated with our observation that PD1^+TIM3^+ terminally
   exhausted T cells had significantly lower expression of GM130 compared
   to PD1^−TIM3^− T cells ([178]Fig. 5M). Given our observation that H[2]S
   treatment protects antitumor T cells from oxidative stress, we next
   sought to determine whether oxidative stress plays a role in mediating
   Golgi stress. When Pmel T cells were treated with monensin to induce
   Golgi stress, a significant increase in ROS (2',7'-dichlorofluorescein
   diacetate, DCFDA) was observed, which was mitigated with H[2]S
   treatment ([179]Fig. 5N). The changes in Golgi stress (quantified by
   GM130 levels) mediated by oxidative stress (using H[2]O[2]) were
   mitigated by H[2]S ([180]Fig. 5O). Further, human T cells were also
   evaluated for Golgi dispersion by confocal microscopy, along with
   mitochondria staining to establish whether there is any spatial
   relationship between the Golgi and mitochondria. Compared to vehicle
   control, H[2]S-treated T cells had more compact Golgi organization. We
   also observed a close spatial relationship between the Golgi and
   mitochondria with areas of overlap between the two organelles (fig.
   S5E).

Prdx4 regulates H[2]S-mediated inhibition of Golgi stress

   Our proteomics screen ([181]Fig. 5D) suggested that an increase in free
   thiols at cysteine 54 in Prdx4 is promoted by H[2]S treatment. Prdx4 is
   a key peroxiredoxin enzyme that is involved in regulating redox balance
   and oxidative stress within the ER. Given the delicate redox balance
   necessary for the native disulfide bonds, we postulated that a
   localized mechanism for the detection and elimination of ROS produced
   during the oxidative folding process may require Prdx4. Thus, we
   hypothesized that ER and Golgi network stress mitigated by H[2]S is
   mediated through the Prdx4 localization and activity in these
   organelles. First, we sought to establish whether Prdx4 also localizes
   within the Golgi apparatus in T cells using a proximity ligation assay
   (PLA), a tool that allows in situ detection of endogenous proteins with
   high specificity and sensitivity, with antibodies directed against
   Giantin (a conserved Golgi membrane protein) and Prdx4. We observed
   that Prdx4 localizes within the Golgi apparatus upon T cell activation
   ([182]Fig. 6A). This localization was further increased upon induction
   of acute oxidative stress, suggesting a previously unidentified role of
   Prdx4 in responding to oxidative stress within the Golgi ([183]Fig.
   6A). Prolonged oxidative stress ultimately resulted in loss of Prdx4
   localization within the Golgi, resembling a colocalization profile
   similar to cells treated with monensin to induce Golgi stress
   ([184]Fig. 6B). Notably, this loss of colocalization was mitigated with
   H[2]S treatment ([185]Fig. 6B).

Fig. 6. Prdx4 regulates H[2]S-mediated inhibition of Golgi stress.

   [186]Fig. 6.
   [187]Open in a new tab

   (A) Pmel T cells were activated for 3 days before treatment with
   vehicle or 50 mM H[2]O[2] for 6 hours. Prdx4-Golgi colocalization was
   assessed by PLA using mouse anti-Giantin and rabbit anti-Prdx4
   antibodies. (B) Activated Pmel T cells treated with vehicle or 50 mM
   H[2]O[2] for 6 or 24 hours with monensin (1 mM). PLA was performed
   using anti-Giantin and anti-Prdx4 antibodies. (C) Experimental design
   using Prdx4-targeted siRNA to silence expression of Prdx4 in Pmel T
   cells. (D) GM130 expression assessed under control versus monensin (1
   μM) conditions in T cells treated with control-siRNA or Prdx4-siRNA.
   (E) GM130 expression assessed under control versus 50 mM H[2]O[2]
   conditions in T cells treated with control-siRNA or Prdx4-siRNA with or
   without 0.5 mM GYY4137. (F) Protein translation assessed in Pmel T
   cells activated with or without GYY4137 and treated with control-siRNA
   or Prdx4-siRNA. (G) GM130 expression in Pmel T cells treated with
   vehicle, GYY4137, or catalase (10 μg/ml). (H and I) Site-directed
   mutagenesis performed to mutate cysteine-54 of the Prdx4 gene.
   Prdx4-targeted shRNA was used to knock down expression of Prdx4 in
   Jurkat T cells, followed by transfection with plasmid containing
   wild-type or mutated Prdx4. (I) GM130 expression assessed after
   treatment with 50 mM H[2]O[2] with or without GYY4137. (J) B16-F10
   tumors were injected subcutaneously into flanks of Pmel mice and grown
   to ~100 mm^2. TILs and T cells from tumor-draining lymph nodes were
   isolated via Ficoll spin followed by magnetic bead positive selection.
   FACS analysis of Prdx4 expression in TILs gated on PD1^loLag3^lo versus
   PD1^hiLag3^hi. (K) Lentiviral supernatant was generated using
   GFP-tagged control plasmid or GFP-tagged Prdx4-overexpression plasmid.
   Pmel T cells were transduced with control or Prdx4 lentivirus. Pmel T
   cells overexpressing Prdx4 and control T cells were treated with
   H[2]O[2] (50 mM) followed by staining with anti-GM130. Data shown
   represent mean ± SEM and analyzed by two-sided Student’s t test or
   one-way ANOVA. ns, P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P
   ≤ 0.0001.

   Next, using siRNA targeting Prdx4, we knocked down expression of Prdx4
   in CD8^+ Pmel T cells ([188]Fig. 6C and fig. S6, A and B). We then
   subjected the cells to conditions of oxidative and Golgi stress in the
   presence or absence of H[2]S treatment. Of note, knockdown of Prdx4
   resulted in a significant decrease in GM130 expression ([189]Fig. 6D).
   This decrease was further exacerbated with monensin treatment,
   demonstrating an increase in susceptibility to Golgi stress in the
   absence of Prdx4 ([190]Fig. 6D). Then, to determine whether the
   protective effect of H[2]S on Golgi stress is dependent on the presence
   of Prdx4, we induced Golgi stress in the presence or absence of H[2]S
   with control (siCtl) and Prdx4 knockdown (siPrdx4) T cells. Of note,
   while H[2]S prevented the decrease in GM130 expression in control T
   cells, this protective effect was absent in Prdx4 knockdown T cells
   ([191]Fig. 6E). A similar dependence on the presence of Prdx4 was
   observed for the effect of H[2]S on increasing protein translation,
   which was not observed upon H[2]S treatment in the Prdx4 knockdown T
   cells ([192]Fig. 6F). In accordance, the Prdx4-silenced T cells showed
   reduced production of effector cytokines (IFN-γ and TNFα) and cytolytic
   molecules (granzyme B and perforin) (fig. S6C).

   Given the key role of Prdx4 in scavenging superoxide species and our
   previous findings indicating a role for oxidative stress (specifically
   H[2]O[2]) in disrupting the Golgi, we hypothesized that it is an
   organelle-specific H[2]O[2] scavenging function of Prdx4 that is
   critical for its ability to protect against Golgi stress. To test this,
   we induced Golgi stress in control or Prdx4 knockdown T cells in the
   presence of H[2]S or the direct H[2]O[2] scavenger catalase. As
   observed previously, the protective effect of H[2]S in mitigating Golgi
   stress was absent when Prdx4 was knocked down, but protection against
   Golgi stress could be rescued with the addition of catalase ([193]Fig.
   6G). Given the increase in free thiol at cysteine 54 in Prdx4 with
   H[2]S, we transfected Prdx4 knockout Jurkat cells with plasmids
   containing either wild-type Prdx4 or a version of Prdx4 mutated at the
   homologous human residue, C51A ([194]Fig. 6H). The protective effect of
   H[2]S on reducing H[2]O[2]-induced Golgi stress was only observed in
   Jurkat cells expressing wild-type Prdx4 and not in the cells expressing
   the cysteine-mutated form of the protein ([195]Fig. 6I). A similar
   Prdx4-dependent effect of H[2]S on reducing ER stress was observed
   Jurkat cells expressing either the normal or cysteine-mutated form of
   Prdx4 (fig. S6D).

   To further establish the physiological relevance of T cell Prdx4
   expression in the TME in vivo, we isolated TILs from B16-F10 tumors and
   assessed levels of Pdrx4 in PD1^hiLag3^hi terminally exhausted TILs
   versus PD1^loLag3^lo TILs. Notably, we observed a significant decrease
   in Prdx4 expression in PD1^hiLag3^hi terminally exhausted TILs
   ([196]Fig. 6J). In addition, we observed a significant decrease in
   Prdx4 expression in antigen-experienced TILs compared to
   antigen-experienced T cells in the tumor draining LNs, suggesting a
   suppressive effect of the TME on sustained Prdx4 expression (fig. S6E).
   To further verify the role of Prdx4 in sustaining T cell function and
   mitigating Golgi stress, lentivirus supernatant was generated to
   overexpress Prdx4 in activated Pmel T cells (fig. S6F). Pmel T cells
   overexpressing Prdx4 were more resistant to loss of GM130 compared to
   control Pmel T cells ([197]Fig. 6K). Upon restimulation, Pmel T cells
   overexpressing Prdx4 produced more cytolytic cytokines compared to
   control Pmel T cells (fig. S6G). Collectively, these findings
   demonstrate a protective effect of H[2]S in reducing Golgi stress that
   is at least partially dependent on the H[2]O[2]-scavaging capacity of
   Prdx4.

Golgi^hi cells exhibit enhanced functionality and stem-like features

   On the basis of our findings that T cells exposed to the TME exhibit
   increased Golgi stress, which correlated with a decrease in GM130
   levels and a decrease in antitumor function, we sought to determine
   Golgi content itself would correlate with T cell antitumor function.
   Thus, activated CD8^+ T cells were stained using a fluorescent dye to
   label the Golgi and were subsequently sorted into cells possessing high
   Golgi content (Golgi^hi) and low Golgi content (Golgi^lo) ([198]Fig.
   7A). We observed significantly increased levels of GM130, intracellular
   H[2]S, and protein translation in the Golgi^hi T cells upon
   restimulation with gp100 tumor antigen compared to the Golgi^lo T cells
   ([199]Fig. 7B). The Golgi^hi subset also exhibited an increase in the
   Tcm phenotype, with Golgi^hi cells having a higher proportion of
   CD62L^+CD44^+ cells ([200]Fig. 7C) and higher expression of Sca-1 and
   CD27 ([201]Fig. 7D).

Fig. 7. Golgi^hi cells have enhanced functionality and stem-like features.

   [202]Fig. 7.
   [203]Open in a new tab

   (A to K) Pmel T cells were activated for 3 days with or without
   GYY4137, then stained with BODIPY-TR Ceramide dye to label Golgi,
   followed by FACS-sorting into Golgi^hi (upper 30%) and Golgi^lo (lower
   30%) based on BODIPY-TR Ceramide dye fluorescence. (B) FACS
   quantification of GM130 and H[2]S levels in Golgi^hi versus Golgi^lo
   cells. Sorted cells were restimulated (gp100 1 μg/ml) and analyzed for
   protein translation using the Click-iT Plus Protein Synthesis Kit (n =
   4). (C) FACS quantification of CD62L^+CD44^+ populations in sorted
   cells (n = 4). (D) FACS quantification of Sca-1 and CD27 in sorted
   cells (n = 4). (E) FACS analysis of TIM3 expression in sorted cells
   after in vitro exhaustion assay (n = 5). (F) Sorted cells were
   restimulated (gp100 1 μg/ml) and the supernatant was collected for
   multiplex cytokine ELISA. Heatmap illustrating log10(fluorescence
   intensity) of cytokines for Golgi^lo versus Golgi^hi groups (n = 3).
   [(G) and (H)] Overall survival for (G) B16-F10 tumor-bearing C57BL/6
   mice treated with FACS-sorted Pmel T cells (n = 10) and (H) Raji
   tumor-bearing NSG mice treated with FACS-sorted CD19-CAR-T cells (n =
   10). (I) Sorted Pmel T cells were injected into B16F10-bearing mice.
   TILs isolated by Ficoll separation and CD8^+ selection stimulated with
   gp100 followed by FACS analysis of TNFα and IFN-γ (n = 4). (J) qRT-PCR
   analysis of mRNA levels for glycosylation enzymes in Golgi^hi versus
   Golgi^lo human CD8^+ T cells. (K) Golgi^hi versus Golgi^lo human CD8^+
   T cells analyzed using antibody-based platform with MALDI-IMS to assess
   N-linked glycosylation. (L and M) Levels of branched N-glycans with
   individual branched N-glycans that were significantly different between
   groups. (N) ELISA quantification of IFN-γ production after CD19-CAR-T
   cells were treated with vehicle or Phostine-PST3.1a (10 μM) and
   cocultured with Raji cells (n = 3). (O) Pmel T cells treated with
   vehicle or Phostine-PST3.1a before restimulation for FACS cytokine
   analysis (n = 4). Data shown represent mean ± SEM, analyzed by
   two-sided Student’s t test or one-way ANOVA. ns, P > 0.05; *P ≤ 0.05;
   **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001.

   In the TME exhaustion assay, Golgi^lo cells had a significantly higher
   expression of TIM3 as compared to Golgi^hi cells, indicating a
   resistance to T cell exhaustion in the Golgi^hi subset ([204]Fig. 7E).
   Functionally, Golgi^hi cells secreted more pro-inflammatory cytokines
   upon restimulation with cognate antigen, including IL-2, IL-4, RANTES,
   and TNFα ([205]Fig. 7F). We observed that treating Golgi^lo T cells
   with H[2]S resulted in partial restoration of their effector functions
   (fig. S7A). Golgi^hi cells were also characterized by a significant
   increase in mitochondrial mass compared to Golgi^lo (fig. S7B). Similar
   to H[2]S-treated T cells, Golgi^hi T cells also displayed a significant
   increase in SRC, demonstrating a correlation between Golgi^hi cells and
   enhanced mitochondrial fitness (fig. S7C). Transmission electron
   microscopy analysis of Golgi^hi versus Golgi^lo cells revealed large,
   healthy-appearing mitochondria in Golgi^hi cells compared to small,
   dense mitochondria in the Golgi^lo cells (fig. S7D). Metabolomics
   analysis revealed significant differences in the metabolite profiles of
   Golgi^hi versus Golgi^lo cells (fig. S7E). Of note, several key
   metabolic pathways were found to be up-regulated in the Golgi^hi cells
   compared to Golgi^lo cells, including GSH metabolism,
   nicotinate/nicotinamide metabolism, and mitochondrial electron
   transport chain (fig. S7, F and G). Similar to H[2]S-treated T cells,
   Golgi^hi cells had a significantly higher NAD^+/NADH ratio (fig. S7H).

   We next sought to determine the antitumor capacity of Golgi^hi versus
   Golgi^lo cells, hypothesizing that Golgi^hi cells would exert superior
   tumor control when adoptively transferred into tumor-bearing hosts. We
   found that Golgi^hi Pmel T cells displayed significantly better tumor
   control (fig. S7I) and improved survival ([206]Fig. 7G) in
   B16-F10–bearing C57BL/6 mice. In addition, when human CD19 CAR-T cells
   were sorted into Golgi^hi and Golgi^lo populations and adoptively
   transferred to the NSG mice engrafted with Raji tumor cells, the
   Golgi^hi CAR-T cells displayed superior control of lymphoma (fig. S7J)
   and prolonged host survival ([207]Fig. 7H). Adoptively transferred
   Golgi^hi cells were detected at higher circulating frequencies after
   transfer (fig. S7K), exhibited reduced exhaustion and preserved Tcm
   phenotype (LAG3^loCD62^hiCD27^hi) (fig. S7L), and maintained superior
   effector function when isolated from the TME ([208]Fig. 7I). When
   cocultured with Raji cells, Golgi^hi CAR-T cells also induced
   significantly more tumor cell death compared to the Golgi^lo CAR-T
   cells (fig. S7M) and produced more perforin and granzyme B (fig. S7N).
   These data confirm that sorting T cells based on the status of the
   Golgi apparatus identifies T cells with superior ability to control
   tumor cells.

   To further identify the pathways that define the Golgi^hi versus
   Golgi^lo subsets, RNA-seq analysis was performed on human CD19 CAR-T
   cells, which revealed a district transcriptomic profile between the two
   subsets (fig. S7O). Fascinatingly, the top two up-regulated pathways in
   the Golgi^hi subset involved microtubule and tubulin binding (fig.
   S7P). Of note, previous studies have identified a critical role for
   Golgi proteins in coordinating with the microtubule organizing center
   to facilitate transportation of key signaling molecules to the
   immunological synapse ([209]46). In addition, RNA-seq analysis revealed
   significant differences in levels of key enzymes involved in regulating
   N-glycan branching in the Golgi, including an increase in MGAT1 (β1,6
   N-acetylglucosaminyltransferase I, a negative regulator of N-glycan
   branching) and a decrease in MGAT5A/B (β1,6
   N-acetylglucosaminyltransferase Va/Vb, the rate-limiting enzyme in
   N-glycan branching), which were validated by reverse transcription
   polymerase chain reaction (RT-PCR) ([210]Fig. 7J). Given the critical
   role of the Golgi in posttranslationally modifying proteins via
   glycosylation, we profiled the N-linked glycosylation on Golgi^hi
   versus Golgi^lo T cells using a method published by Dressman et al.
   ([211]47). While an overall increase in total glycosylation in Golgi^hi
   cells was observed, it was not statistically significant ([212]Fig.
   7K). However, a notable decrease in overall branched N-glycans was
   observed ([213]Fig. 7, L and M), which correlated with the reduced
   expression of Mgat5a/b. Levels of branched N-glycans have been shown to
   be directly correlated with the T cell activation threshold in an
   Mgat5-dependent manner, while Mgat1 has been shown to paradoxically
   inhibit the activity of Mgat5, leading to decreased N-glycan branching
   ([214]48–[215]50). To further validate the importance of decreased
   Mgat5 activity and thus low N-glycan branching in the Golgi^hi subset,
   we treated activated CD8^+ T cells with the Mgat5 inhibitor Phostine
   PST3.1a, a selective inhibitor of Mgat5 enzymatic activity, which has
   been shown to have antitumor activity in in vivo models of glioblastoma
   ([216]51). As predicted, inhibition of Mgat5 activity in both human and
   murine T cells resulted in a more potent effector response upon
   encountering tumor antigen, as evidenced by an increase in IFN-γ
   production by CD19 CAR-T cells when cocultured with Raji tumor cells
   ([217]Fig. 7N) and an increase in cytolytic cytokine production by Pmel
   T cells upon restimulation with gp100 peptide ([218]Fig. 7O). Thus, a
   role for the Golgi in modulating protein glycosylation and rendering a
   robust antitumor phenotype to T cells could be a key factor in
   determining immunotherapeutic outcomes.

DISCUSSION

   The cumulative role of cellular organelles in shaping the life and
   function of a cell has been long acknowledged ([219]52). While each
   organelle plays a specific role in the growth and development of T
   cells, numerous studies have thus far focused on targeting
   mitochondria-, ER-, or lysosome-related pathways to improve the
   antitumor T cell immune response. Strategies mitigating stress in these
   organelles have shown to improve T cell fitness and enhance tumor
   control. Increasing evidence suggests that the Golgi apparatus also
   plays a crucial function in sensing and integrating external and
   internal cues to promote cellular homeostasis. The Golgi apparatus is
   essential for maintaining normal cell physiology because it supports
   cell survival, promoting cell proliferation, and facilitating cell-cell
   communication and migration. These roles are partly influenced by
   established Golgi functions, such as posttranslational modifications,
   lipid production, intracellular trafficking, and protein secretion
   ([220]19). Because intracellular organelles are tightly regulated under
   various stress conditions, we hypothesized that Golgi apparatus
   disruption under oxidative stress could alter lipid and protein
   modification, packaging, and transport, resulting in suboptimal
   antitumor T cell function.

   Disruption of Golgi architecture and functions, termed Golgi stress,
   has been previously shown to alter redox balance and affect cell
   survival ([221]43). Golgi stress inducers, including monensin and
   brefeldin A, have been widely shown to impair Golgi structure and
   function. These Golgi stressors have been shown to up-regulate CSE and
   endogenous H[2]S generation, whereas inhibition of the CSE/H[2]S system
   results in increased susceptibility to Golgi stress ([222]17). Thus, we
   hypothesized that treating T cells with exogenous H[2]S would overcome
   Golgi stress and restore Golgi apparatus function to enhance antitumor
   T cell response. The role of H[2]S in biological processes has
   increasingly become the focus of research in recent years. A particular
   focus has been on the cytoprotective and antioxidant properties that
   H[2]S appears to have in cells that are exposed to high levels of
   oxidative stress ([223]53–[224]62). We found that with traditional T
   cell activation methods, T cells markedly up-regulate H[2]S production;
   however, as T cells continue to proliferate, this initial increase in
   H[2]S returns to baseline levels. These results initially suggested to
   us that H[2]S signaling may play an important role in supporting T cell
   activation and function and that sustained H[2]S signaling could
   produce robust antitumor T cells. Antitumor T cells expanded with
   exogenous H[2]S or overexpressing CBS to increase endogenous H[2]S
   production in T cells leads to high levels of cytolytic cytokines and
   sustained levels of protein translation upon TCR stimulation.

   Our data demonstrate that H[2]S is an important immunomodulatory
   signaling molecule that can be used to alter multiple factors in T
   cells to enhance their antitumor capacity and that this approach can be
   used to program TILs and genetically modify T cells with potent
   antitumor phenotype. Treating T cells with H[2]S donors or increasing
   endogenous production of H[2]S supports previously established
   signatures of robust antitumor T cells, such as enhanced stemness,
   increased mitochondrial function, and reduced susceptibility to
   oxidative stress and ER stress upon chronic antigen stimulation. Our
   data also highlight that the increased Tcm phenotype observed in
   H[2]S-treated T cells can be attributed to enhanced NAD^+ levels,
   through the NAD^+-Sirt1-Foxo1 axis ([225]9, [226]63). Foxo1 has also
   been recently shown to be important for determining CAR-T cell memory
   phenotype and function ([227]36, [228]37). Thus, it is likely that
   H[2]S acts at multiple levels to render a stress-free Tcm phenotype
   that results in improved persistence in vivo upon adoptive transfer to
   bring improved tumor control.

   In other cell types, H[2]S has been shown to reduce ER stress,
   particularly in the context of oxidative stress ([229]64, [230]65). We
   observed a similar effect of H[2]S in T cells, both in reducing overall
   oxidative stress and in preventing ER stress upon chronic antigen
   stimulation. Our data show that dysfunction within the Golgi apparatus
   is another critical factor that needs to be considered when generating
   tumor-reactive T cells for adoptive therapies and that H[2]S treatment
   during the expansion process can be used for effective programming.
   However, it must be noted that given the interdependence of organelle
   function in shaping the cellular response, a limitation of this study
   remains in determining the sole role of Golgi stress in altering the
   immune response when mitochondria or ER stress is also affected.
   Nonetheless, these findings offer insight into the role of H[2]S
   signaling in regulating both ER and Golgi network stress in T cells and
   offer previously unidentified therapeutic strategies for improving
   antitumor T cell response.

   While the ER and its associated ribosomes are responsible for
   synthesizing and folding proteins, the Golgi apparatus is a closely
   associated organelle responsible for further modification and sorting
   of synthesized proteins received from the ER. We were thus particularly
   interested in the status of the Golgi apparatus in T cells, given the
   importance of Golgi processing of the secreted factors that are
   required for T cell effector function. Similar to ER stress, multiple
   studies have shown that cells can also experience Golgi stress,
   characterized as a fragmentation of the Golgi apparatus and an
   inability to process proteins properly ([231]66–[232]68). Studies have
   recently shown that H[2]S is an important protective regulator of Golgi
   stress ([233]17). We found that Golgi stress was a characteristic of
   exhausted T cells and that treatment with H[2]S could reduce Golgi
   stress in antitumor T cells. Our data also demonstrate an important
   role for the thiol-specific peroxidase Prdx4 in regulating ER and Golgi
   stress in antitumor T cells. Prdx4 reactive cysteines are particularly
   susceptible to oxidation, rendering Prdx4 inactive when H[2]O[2] levels
   are high ([234]69). Strategies to target and selectively reduce the
   functional cysteine residues of Prdx4 and other molecules that regulate
   Golgi function will have the potential for high translational value to
   optimize immunotherapy. These findings identify Golgi stress as a
   previously unidentified therapeutic target in cancer immunotherapy and
   identify H[2]S-based therapy as a potential strategy for mitigating
   Golgi stress to enhance antitumor immunity.

   Further, it is of substantial interest to the scientific and medical
   communities to identify simple phenotypic attributes of potent
   antitumor T cells. For example, Sukumar et al. ([235]8) earlier
   demonstrated that T cells with low mitochondrial membrane potential as
   measured with TMRM dye had robust antitumor capacity and that T cells
   with low mitochondrial membrane potential could be sorted and used for
   ACT to produce durable tumor control. Similarly, we show that T cells
   with high and low Golgi mass have distinct functionality profiles and
   that sorting on Golgi^hi cells produces a subset of T cells with
   superior ability to control tumors. Whether it is increased expression
   of key cell signaling molecules in Golgi^hi fraction resulting from
   asymmetric cell division and distribution that contributes to long-term
   maintenance of T cell function and control ([236]70) or whether it is
   reduced activity of Mgat5-dependent N-glycan branching that lowers T
   cell threshold of activation and leads to increased functionality
   ([237]49) will need to be dissected in the future. Regardless, this
   approach to cell selection will likely be broadly applicable to
   multiple forms of ACT for treating cancer, including TIL and CAR-T
   therapy.

   Overall, it is of critical importance to continue to identify novel
   therapeutic strategies for enhancing the ability of the immune system
   to control and eliminate tumors. An improved understanding of
   regulation of the T cell immune response at the organelle level can
   help devise effective antitumor therapies focused on reducing organelle
   stress, limiting organelle damage, improving inter-organelle
   cross-talk, and restoring organelle homeostasis, and could be useful to
   improve immunotherapy options. Ultimately, we believe that H[2]S
   signaling plays a key role in immune regulation at multiple levels,
   including reduction of Golgi stress in antitumor T cells, which can be
   used to boost antitumor immunotherapeutic strategies.

MATERIALS AND METHODS

Materials availability

   This study did not generate new or unique reagents.

Data and code availability

   Single-cell RNA-seq data have been deposited at Gene Expression Omnibus
   (GEO) and are publicly available as of the date of publication. The
   mass spectrometry proteomics data have been deposited to the
   ProteomeXchange Consortium via the Proteomics Identification Database
   (PRIDE) partner repository. Accession numbers are listed in the key
   resources table. This paper does not report original code.

Mice

   C57BL/6, B6-Rag^−/−, Pmel, NSG, and Cbs^−/− mice were obtained from the
   Jackson Laboratory (Bar Harbor, ME). Animals were maintained in
   pathogen-free facilities, and experimental procedures were approved by
   the Institutional Animal Care and Use Committees of Medical University
   of South Carolina, Charleston (approval no. IACUC-2018-00628-1). For
   tumor experiments, an equal number of age- and gender-matched (both
   male and female) mice were randomly assigned for the experiments when
   they were between 8 and 10 weeks old. No influence of sex on the result
   of the studies was observed.

Cell lines

   B16-F10 and Jurkat cells were obtained from the American Type Culture
   Collection (ATCC), suggesting male origin. Raji cells (ATCC no. CCL-86)
   were a gift from M. Nishimura (Loyola University Chicago).

Generation of Prdx4 knockout and Prdx4 mutant cells

   Jurkat cells were transduced with human Prdx4 shRNA lentiviral
   particles expressing a puromycin resistance gene. Puromycin was added
   to the culture media to selectively expand the transduced Jurkat cells.
   Prdx4 knockdown was confirmed by RT-PCR and Western blot analysis. To
   generate the Prdx4 mutant plasmid, primers were designed based on the
   coding sequence of the canonical gene of interest (Prdx4). The coding
   sequence was converted into the amino acid codon sequence using Expasy
   to mutate the amino acid of interest. Roughly 15 to 20 amino acids
   upstream and downstream of the mutated codon were selected, and the New
   England Biolabs Tm Calculator was used to adjust the primer length,
   projected annealing temperature, and GC content. The Harvard Reverse
   Complement Tool was used to produce the reverse primer sequence.
   Primers were ordered from Integrated DNA Technologies, including
   5′-phosphorylation for plasmid ligation. The template plasmid
   (containing the wild-type gene of interest for mutation) and the
   primers were then used with the QuikChange XL Site-Directed Mutagenesis
   kit (Agilent Technologies, no. 200516) per the manufacturer’s
   instructions. The successful mutation was confirmed via sequencing of
   the plasmid (Genewiz, Azenta Life Sciences).

Overexpression of Prdx4

   To generate the lentiviral particle containing prdx4 plasmid, 293T
   Lenti-X cells were seeded in complete Dulbecco’s Modified Eagle’s
   Medium (DMEM) overnight in a 10-cm tissue culture plate. The next day,
   4 hours before transfection, the cells were treated with 25 μM
   chloroquine. Following chloroquine treatment, cells were transfected
   with 10 μg of either Prdx4 plasmid or mock plasmid and 7.5 μg of psPAX2
   packaging plasmid and 2.5 μg of pMD2.G envelope plasmid through
   Lipofectamine 3000 according to the manufacturer’s protocol. The next
   day, the medium was replaced with fresh complete DMEM and allowed to
   grow for 24 hours. The next day, media containing virus particles were
   collected and filtered through a 0.45-μm syringe filter. Supernatants
   containing viral particles of either mock or Prdx4 insert were diluted
   at 1:1 ratio with fresh complete Iscove’s Modified Dulbecco’s Medium
   (IMDM) before mouse T cell transduction. T cells were collected from
   spleen of healthy Pmel mouse, and 1 × 10^6 cells/ml were transduced
   with diluted viral supernatant by spinoculation method in the presence
   of protransdusin at 2000 rcf at 32°C for 2 hours. Twenty-four hours
   later, cells were collected, washed, and checked for green fluorescent
   protein (GFP) expression before use for further experimental analysis.

T cell differentiation

   Naïve total T cells were purified from the total splenocytes of 6- to
   9-week-old C57BL/6 mice, first by incubating the cells with
   biotinylated anti-CD19, anti-Gr1, anti-mouse ter-119, anti-CD11b,
   anti-CD11c, anti-NK1.1, anti-CD25, and anti-CD105 (Cell Signaling
   Technology), followed by negative selection with streptavidin magnetic
   particles (BD Biosciences). Purified T cells were then activated with
   soluble anti-CD3 (1 μg/ml) and anti-CD28 (1 μg/ml) in the presence of
   IL-2 (100 IU/ml). Total splenocytes from 6- to 9-week-old Pmel
   transgenic mice (bearing class I restricted CD8^+ T cells) were
   activated with gp100 melanoma antigen (1 μg/ml) in the presence of IL-2
   (100 IU/ml). Within experiments, mice were age and sex matched. T cells
   were cultured in IMDM media supplemented with 10% fetal bovine serum
   (FBS), 4 mM l-glutamine, penicillin (100 U/ml), streptomycin (100
   μg/ml), 55 μM β-mercaptoethanol under 7% CO[2], and atmospheric oxygen
   at 37°C in a humidified incubator. T cells were restimulated to
   evaluate intracellular cytokines by flow cytometry either with phorbol
   myristate acetate (PMA)/ionomycin for 4 hours or soluble anti-CD3 (1
   μg/ml) and anti-CD28 (1 μg/ml) or with gp100 melanoma antigen for 6
   hours in the presence of Golgi inhibitors. In some experiments, in
   vitro differentiated T cells were treated with either the vehicle
   control or H[2]S donor GYY4137 (0.5 mM).

Retroviral transduction

   For CD19-CAR-T generation, human peripheral blood mononuclear cells
   (PBMCs) were obtained from healthy donors by Ficoll gradient spin and
   activated for 3 days with soluble anti-CD3 antibody (Okt-3, 1 μg/ml).
   For Cbs overexpression, freshly isolated Pmel T cells were activated
   with gp100 peptide (1 μg/ml) for 3 days. CD19-CAR-T and CBS-CD19-CAR-T
   viral supernatant was generously gifted by M. Nishimura (Loyola
   University Chicago). Cbs viral supernant for Pmel transduction was
   generated using Cbs human tagged ORF clone (Origene no. RC201755L4).
   After 3 days of activation, T cells were plated at a concentration of 2
   × 10^6 cells/ml in complete media onto non–tissue culture–treated
   24-well plates (USA Scientific) coated with Retronectin. Viral
   supernatant (1 ml) was added on top of the T cells, and the plate was
   spun at 2000g for 2 hours and 32°C. After spinning, 1 ml of media was
   removed and replaced with fresh media containing IL-2 (200 IU/ml)
   before the cells were incubated overnight. The cells were collected,
   washed, and plated the following day for use in further experiments.

Adoptive T cell protocol

   B16-F10 (0.3 × 10^6) melanoma tumor cells were injected subcutaneously
   into the left flank of 8- to 10-week-old C57BL/6 or Rag^−/− mice. After
   tumor establishment, recipient mice were injected intraperitoneally
   (i.p.) with cyclophosphamide (4 mg per mouse) before adoptively
   transferring intravenously either Pmel, Pmel-Cbs^td, or TILs (1 ×
   10^6). After adoptive T cell transfer, recipient mice were given IL-2
   (50,000 U per mouse, i.p.) for three consecutive days. Raji cells (0.5
   × 10^6) were injected subcutaneously into the left flank of 8- to
   10-week-old NSG mice. After tumor establishment, CD19-CAR-T cells (5 ×
   10^6) were adoptively transferred intravenously. After adoptive T cell
   transfer, recipient mice were given IL-2 (50,000 U per mouse, i.p.) for
   three consecutive days. For all tumor control experiments, mice were
   randomly assigned to treatment groups and labeled using coded ear
   punch. Tumor measurements were then conducted in a blinded fashion
   until final analysis.

In vitro TME exhaustion assay

   B16-F10 (0.3 × 10^6) melanoma tumor cells were injected subcutaneously
   into the left flank of 8- to 10-week-old C57BL/6 mice. Once the tumors
   reached a size of approximately 150 mm^2[,] the mice were euthanized
   and the tumors were removed. The tumors were then processed into
   single-cell suspension using a mouse tumor dissociation kit (Miltenyi
   Biotec, no. 130-096-730). The tumor cell suspension was then plated in
   six-well plates with IMDM media supplemented with 10% FBS, penicillin
   (100 U/ml), streptomycin (100 μg/ml), and 55 μM β-mercaptoethanol under
   7% CO[2], and atmospheric oxygen at 37°C in a humidified incubator.
   After 5 days of culture without changing the media, the supernatant was
   removed from the wells and spun down to remove any tumor cells.
   Three-day activated Pmel T cells were then resuspended in the tumor
   supernatant along with gp100 antigen (0.1 μg/ml). To promote chronic
   antigen stimulation, gp100 (0.1 μg/ml) was added every day for 4 days.
   After 4 days of culture with tumor supernatant and chronic stimulation,
   the T cells were removed and analyzed for expression of exhaustion
   markers and functional assays.

Flow cytometry

   Staining for cell surface markers was performed by incubating cells
   with the antibody at 1:200 dilutions in FACS buffer [0.1% BSA in
   phosphate-buffered saline (PBS)] for 30 min at 4°C. For intracellular
   cytoplasmic proteins, surface markers were stained before
   fixation/permeabilization (BD Cytofix/Cytoperm Kit, BD Biosciences, San
   Jose, CA). For staining of transcription factors, cells were stained
   with surface markers and fixed/permeabilized with a FoxP3 staining
   buffer set (eBioscience, San Diego, CA). For Cbs, pIRE1α, pPERK, and
   ATF4 intracellular staining, surface markers were stained before
   fixation/permeabilization, followed by primary unconjugated antibody
   staining and subsequent incubation with fluorochrome-conjugated
   secondary antibody (Jackson ImmunoResearch Laboratories, West Grove,
   PA). In addition, MitoTracker Red (Cell Signaling Technology no. 9082),
   LIVE/DEAD Fixable Yellow Dead Cell Stain Kit (Invitrogen no.
   [238]L34959), DCFDA dye (Abcam no. ab113851), and WSP-1 dye (MCE no.
   HY-124409) were used to evaluate mitochondrial mass, cell viability,
   cellular ROS, and H[2]S production, respectively, following the
   manufacturer’s protocol. Samples were acquired on LSRFortessa and
   analyzed with FlowJo software (Tree Star, OR).

Immunoblotting

   For evaluation of the protein level, cell pellets were washed in PBS
   and lysed in radioimmunoprecipitation assay buffer (Thermo Fisher
   Scientific, Waltham, MA), including protease/phosphatase inhibitors,
   vortexed, and incubated for 20 min on ice. Cell lysates were then
   centrifuged at 12,000 rpm for 15 min at 4°C. The supernatants were
   collected, and proteins were quantified with a BCA protein assay kit
   (Thermo Fisher Scientific, Waltham, MA). For immunoblot analyses, 20 μg
   of protein lysates per sample was denatured in 4× Loading dye and
   boiled using a heating block at 95°C for 10 min before loading to SDS
   gradient gels 4 to 20% (Bio-Rad Criterion, 1-hour runs). Gels were
   semidry transferred onto polyvinylidene difluoride, and the membranes
   were blocked with 3% milk in 0.1% tris-buffered saline with tween 20
   (TBST). Next, the membrane was probed with the following primary
   antibodies: anti-Prdx4 (Proteintech, 10703-1-AP), anti-eIF2α (Cell
   Signaling Technology, no. 9722), or anti–β-actin (Signaling Technology,
   no. 4967L) overnight at 4°C followed by 1-hour incubation with
   horseradish peroxidase–conjugated secondary antibody (Cell Signaling
   Technology, Danvers, MA) and using a Clarity Western ECL Substrate
   (Bio-Rad, Hercules, CA).

Real-time quantitative PCR

   Total RNA was extracted from pellets of the indicated T cell subsets (2
   × 10^6 cells) using TRIzol reagent (Life Technologies, Grand Island,
   NY). cDNA was generated from 1 μg of total RNA using the iScript cDNA
   Synthesis Kit (Bio-Rad, Hercules, CA). SYBR Green incorporation
   quantitative real-time PCR was performed using an SYBR Green mix
   (Bio-Rad, Hercules, CA) in the CFX96 Detection System (Bio-Rad,
   Hercules, CA). The expression of different genes was quantified
   relative to Actb. For RT-PCR arrays, RT^2 Profiler PCR Arrays (Qiagen)
   were used according to the manufacturer’s instructions.

RNA-seq analysis

   Cells were immediately pelleted by centrifugation at 4°C and
   resuspended in 1 ml of TRIzol. RNA concentration was measured using a
   NanoDrop 8000. RNA quality was assessed using an Agilent 4200
   TapeStation and RINe values ranged from 9.7 to 10. Total RNA (250 ng)
   was used in the construction of libraries with the New England Biolabs
   NEBNext Poly(A) mRNA Magnetic Isolation Module (catalog no. 7490L) and
   Ultra II Directional RNA Library Prep Kit for Illumina (catalog no.
   7760L) according to the manufacturer’s instructions. Dual-indexed
   libraries were pooled to and sequenced at VANTAGE (Vanderbilt
   University Medical Center) on an Illumina NovaSeq 6000 (S4 flow cell)
   to a depth of approximately 25 million paired-end 150-bp reads per
   library. Reads were aligned to the mouse mm10 reference genome using
   STAR (v2.7.1a). Only uniquely mapped reads were retained for further
   analyses. Quality control metrics were assessed by the Picard tool
   ([239]http://broadinstitute.github.io/picard/). Gencode annotation for
   mm10 (version M25) was used as reference alignment annotation and
   downstream quantification. Gene level expression was calculated using
   featureCounts (v2.0.1). Counts were calculated based on protein-coding
   genes from the annotation file. Counts were normalized using counts per
   million reads (CPM). Genes with no reads in either Control or Treated
   samples were removed. To infer potential experimental confounders, we
   calculated surrogate variables using the sva package in R. Differential
   expression analysis was performed in R using DESeq2 (v1.34) with the
   following model: gene expression ~ Treatment + nSVs. We estimated
   log[2]fold changes and P values. P values were adjusted for multiple
   comparisons using a Benjamini-Hochberg correction [false discovery rate
   (FDR)]. Differentially expressed genes were considered for FDR < 0.05.
   Mouse Gene ID was translated into Human Gene ID using the biomaRt
   package in R. The functional annotation of differentially expressed
   genes was performed using clusterProfiler (v4.2). A Benjamini-Hochberg
   FDR (FDR < 0.05) was applied as a multiple comparison adjustment.

Transmission electron microscopy

   The cells were pelletized and fixed in 2% phosphate-buffered
   glutaraldehyde for 1 hour. Next, the pellets were rinsed in 0.1 M
   phosphate-buffered rinse and then postfixed in 2% aqueous osmium
   tetroxide for 1 hour. After rinsing in distilled water, the pellets
   were dehydrated through a series of graded ethyl alcohol; 50% ethanol
   (EtOH) for 15 min, 70% EtOH for 15 min, 95% EtOH for 15 min, and lastly
   twice with 100% EtOH for 15 min each. The dehydrant was removed using
   the intermediate fluid, propylene oxide, one change of 10 min each.
   Next, the pellets were infiltrated with a 1:1 solution of propylene
   oxide and Embed 812 (Electron Microscopy Sciences, Ft. Washington, PA)
   for 1 hour. The infiltration was continued using a 1:2 solution of
   propylene oxide and Embed 812 overnight. The pellets were embedded in
   Embed812 the following day and polymerized in a 60°C oven for 48 hours.
   Preliminary 1/2-μm sections were cut and stained with Toluidine Blue
   and examined using a light microscope. Then, with the cell types
   identified, the 70-nm thin sections were cut and stained with uranyl
   acetate and lead citrate and allowed to dry. The sections were viewed
   on the JEOL 1010, and images were taken with a Hamamatsu electron
   microscope camera.

Confocal microscopy

   Alexa Fluor 488 (anti-Giantin) and Alexa Fluor 647 (anti-TOMM20)
   fluorescence were imaged in a Zeiss LSM 880 NLO inverted laser scanning
   confocal microscope (Thornwood, NY) using a 63× 1.4 numerical aperture
   plan-apochromat oil immersion lens. Alexa Fluor 488 and Alexa Fluor 647
   were excited at 488 and 633 nm, respectively. Emitted light was
   detected with an Airyscan super-resolution detector at BP 495 to 550 nm
   for Giantin label and LP 654 nm for TOMM20. Z-stack Airyscan images
   were processed using the Huygens Professional deconvolution and image
   analysis software (Scientific Volume Imaging, The Netherlands). After
   images were deconvolved using Huygens’ Deconvolution Express (Standard
   Profile) that determines optimal parameters, we performed
   three-dimensional (3D) surface rendering of deconvolved images with
   watershed augmentation using Huygens’ Surface Renderer.

Metabolomics

   Different metabolites’ intracellular levels were determined by
   performing comprehensive hydrophilic metabolites analysis using an
   Liquid chromatography–mass spectrometry (LC–MS) platform (Metabolomics
   Core Facility, Northwestern University). Data were then analyzed using
   MetaboAnalyst software. Samples were loaded equivalently across the
   platform and normalized to Bradford values before statistical analysis.

Extracellular flux assays

   Oxygen consumption rate (OCR) and extracellular acidification rate
   (ECAR) were determined using the Seahorse Xfe96 analyzer (Agilent
   Technologies, Santa Clara, CA). Briefly, T cells (0.5 × 10^6 per well)
   were plated on a Cell-Tak coated Seahorse culture plate for 30 min.
   OCR, a measure of OXPHOS, was analyzed under basal condition, or in
   response to 1.0 μM oligomycin, 1.0 μM fluoro-carbonyl cyanide
   phenylhydrazone (FCCP), and 2 μM rotenone, plus 100 nM antimycin A.
   ECAR, a measure of glycolysis, was measured under basal conditions and
   in response to glucose (5.5 mM), oligomycin (1.0 μM), and
   2-deoxyglucose (100 mM). All reagents were purchased from Sigma-Aldrich
   (St. Louis, MO).

Protein translation assay

   Pmel T cells were restimulated with gp100 peptide (0.1 μg/ml), and the
   Click-iT Plus OPP Alexa Fluor 647 Protein Synthesis Assay Kit (Thermo
   Fisher Scientific no. [240]C10458) was used to measure protein
   translation using flow cytometry following the manufacturer’s protocol.

Multispectral imaging flow cytometry analysis

   T cells were stained with conjugated antibodies for surface markers as
   described in the figure legends followed by fixation and
   permeabilization with BD Cytofix/Cytoperm Kit (BD Biosciences, 554722).
   The Golgi was labeled using anti-Giantin antibody (Abcam, ab80864) as
   the primary antibody followed by secondary antibody staining with
   anti-rabbit immunoglobulin G (IgG) conjugated to Alexa Fluor 488
   (ab150077). The cells were then imaged using ImageStreamX mark II
   imaging flow cytometer and analyzed using IDEAS 6.2. Spectral overlap
   was compensated for using single-stain controls. Image analysis for
   Golgi area was performed using the protocol established by
   Eisenberg-Lerner et al. ([241]44). First, cells were gated on single
   cells using the area and aspect ratio features and then gated on
   focused cells using the Gradient RMS feature. Golgi area was calculated
   using the anti-Giantin fluorescence signaling, using the Threshold_50
   mask that includes the 50% highest-intensity pixels of the Golgi
   staining, with a mask defined as Area_Threshold_50 considered as the
   Golgi area to compare between the relevant conditions.

Proximity ligation assay

   PLAs were performed using NaveniFlex Cell MR RED (Cayman, no. 39505)
   according to the manufacturer’s instructions. Anti-Prdx4 (Proteintech,
   10703-1-AP) and anti-Giantin (Abcam, ab37266) were used as primary
   antibodies. T cells were fixed with 4% PFA for 15 min then
   permeabilized with 0.1% Triton X-100 for another 15 min at room
   temperature. PLA blocking agent was then used to block nonspecific
   binding, and the cells were then incubated overnight in the presence of
   antibodies of interest. After washing, secondary PLA probes conjugated
   to oligonucleotides were added to the cells, then a ligase was added to
   the samples to ligate the oligonucleotides that are in close proximity.
   DNA rolling-circle amplification was then performed using the PLA
   polymerase in the presence of fluorescence-bound oligonucleotide
   probes, which then yielded a fluorescent signal only where the two
   targets of interest are interacting in close proximity. Cells were
   imaged using the Olympus FV10i laser scanning confocal microscope.
   Signal quantification was performed using the Duolink in Situ Image
   Tool software.

Isolation of tumor-infiltrating T cells

   To obtain tumor-infiltrating T cells from subcutaneously established
   solid B16-F10 melanoma-bearing mice, tumors were excised, chopped
   finely using tweezers and scissors, and then digested with collagenase
   type IV (2 mg/ml; Stemcell Technologies, Vancouver, BC) for 45 min. The
   tumors were filtered through 70 cell strainers (BD Biosciences, San
   Jose, CA). The cell suspension was washed in culture medium twice by
   centrifugation at 1500 rpm for 10 min at 4°C. After the second wash,
   the cells were resuspended in 6 ml of PBS and layered carefully over 3
   ml of Ficoll-Paque (GE Healthcare) followed by centrifugation at 1500
   rpm for 30 min at room temperature. The enriched TILs obtained at the
   interface as a thin buffy layer were washed with PBS twice and lastly
   resuspended in FACS staining buffer for further staining procedures.

Generation of bone marrow–derived dendritic cells

   To obtain bone marrow, mice were euthanized by CO[2] inhalation.
   Femoral bones were removed, and all remaining tissue was dissected off
   the bone. The ends of each bone were cut off and the bone marrow was
   flushed from the center of the bone. Bone marrow cells were then
   cultured on non–tissue culture–treated six-well plates in complete
   media for 7 days with granulocyte-macrophage colony-stimulating factor
   (GM-CSF; 10 ng/ml) and IL-4 (10 ng/ml) to generate dendritic cells. The
   medium was changed every 2 days. Floating cells were removed, and the
   loosely adherent cells were considered to be dendritic cells. FACS
   analysis was then performed to confirm the successful generation of
   dendritic cells.

N-linked glycosylation profiling of T cells by mass spectrometry imaging

   Antibody-based analysis of N-linked glycosylation was performed as
   fully described by Dressman et al. ([242]47, [243]71) In brief,
   amine-reactive slides were coated with antibodies at 200 ng per 1.5 μl
   spot and incubated at room temperature for 1 hour in a preheated
   humidity chamber. Bound antibodies were washed with 0.1%
   octyl-β-d-glucopyranoside in PBS (PBS-OGS) for 1 min, followed by
   blocking in 100 mM ammonium bicarbonate solution (pH 8) for 30 min.
   Antibodies were then deglycosylated by adding 100 μl of PNGase F PRIME
   (10 μg/ml) diluted in high-performance liquid chromatography
   (HPLC)-grade water into each well and placed back into the humidity
   chamber and incubated at 37°C for 2 hours. Following deglycosylation,
   antibody arrays were washed with PBS-OGS (3 min × 3) with gentle
   shaking followed by PBS washes (3×) and a water wash (1 min). T cells
   were washed (3×) in FACS buffer and resuspended in FACS buffer. Cell
   suspension (100 μl) was added to each well. Cell capture was performed
   at 4°C, shaking at 250 rpm for 1 hour. The 24-well module was then
   removed and the slide was placed in a slide mailer containing 10%
   neutral buffered formalin for 20 min. After 20 min, the slide was
   removed and placed in PBS at room temperature for up to 1 week before
   further processing. Sialic acid stabilization and derivatization were
   performed via a slide-based sequential amidation-amidation reaction
   with dimethylamine and propargylamine, termed AAXL (amidation-alkyne
   Xtra linker). To release N-glycans from captured cells, PNGase F Prime
   (0.1 μg/μl in HPLC water) was sprayed onto the slide using an M5
   TM-Sprayer (HTX Technologies). Slides were incubated for 2 hours at
   37°C in a preheated humidity chamber. Matrix-assisted laser
   desorption/ionization (MALDI) matrix α-cyano-4-hydroxycinamic acid
   (CHCA, 7 mg/ml in 50% acetonitrile/0.1% trifluoracetic acid) was
   sprayed using the same M5 TM-sprayer. Two passes of ammonium phosphate
   monobasic (5 mM) were sprayed across the slide to reduce matrix
   clustering and improve the signal. N-glycan imaging was conducted using
   a timsTOF-flex MALDI-Quadrupole Time-of-Flight (Q-TOF) mass
   spectrometer (Bruker) operated in positive ion mode at a mass/charge
   ratio (m/z) range of 700 to 4000. Images were collected using a
   SmartBeam 3D laser that operated at 10,000 Hz using the M5 small smart
   beam setting at a laser spot size of 100 μm run at a raster of 150 μm.
   Six hundred laser shots per pixel were collected with an ion transfer
   time of 120 μs, a prepulse storage of 25 μs, a collision radio
   frequency of 4000 Vpp, a multipole radio frequency of 500 Vpp, and a
   collision cell energy of 25 eV.

Proteome-wide analysis of reactive cysteine thiols by LC-MS/MS–based
proteomics

   Cells were treated with or without H[2]S for 3 days of activation
   followed by 3 days of expansion. For analysis of differentially
   reactive cysteine residues, the sample preparation and analysis were as
   described by van der Reest et al. ([244]72) with minor modifications.
   Cells were lysed in freshly made 9 M urea, 50 mM tris (pH 8) buffer
   with Universal Nuclease (100 U/ml; Thermo Fisher Scientific, Pierce
   catalog no. 88702). To label free thiols, lysis buffer was supplemented
   with either 55 mM of stable isotope-labeled light (I^12C[2]ONH[4]) or
   heavy (I^13CD[2]^13CONH[2]) iodoacetamide (Sigma-Aldrich, catalog no.
   721328) for control or H[2]S-treated cells, respectively. Equal amounts
   of heavy or light labeled proteins from three biological replicates
   were combined and reducible thiols were reduced with 70 mM
   dithiothreitol for 45 min at 25°C. After diluting with 50 mM ammonium
   bicarbonate, newly released thiols were alkylated with 80 mM
   N-ethylmaleimide (NEM) and incubated 2 hours at 25°C. Proteins were
   precipitated by adding 6 volumes of cold acetone, incubated overnight
   at −20°C, then centrifuged at 16,000g for 15 min at 4°C. The pellets
   were dissolved in an 8 M urea in 50 mM ammonium bicarbonate and the BCA
   protein assay was repeated. The concentration of urea was diluted to <2
   M with ammonium bicarbonate and the proteins were digested with 1:33
   (enzyme: protein) of Lys-C for 2 hours at 25°C and subsequently with
   1:33 trypsin for 16 hours at 37°C while mixing at 300 rpm. The
   resulting peptides (100 μg) from three combined samples were each
   fractionated into six fractions using the high-pH RP spin columns
   according to the manufacturer’s protocol (Thermo Fisher Scientific,
   Pierce catalog no. 84868). Eluted peptides were dried by vacuum
   centrifugation. Two-microgram aliquots of peptides from each fraction,
   obtained using ZipTips with 0.6 μl of C18 resin (Millipore, Burlington,
   MA catalog no. ZTC18S096), were analyzed by LC-MS/MS on an Easy-nLC
   1200 coupled to a Orbitrap Fusion Lumos MS (Thermo Fisher Scientific,
   Waltham, MA). Pressure-loaded peptides were chromatographically
   separated on a 75 μm × 50 cm column (Acclaim PepMap RSLC C18, 2 μm, 100
   Å Thermo Fisher Scientific catalog no. 164540) thermostated at 45°C
   with a gradient of 5 to 35% solvent B in 180 min (solvent A: 5%
   acetonitrile, 0.2% formic acid and solvent B: 80% acetonitrile, 0.2%
   formic acid) at 300 nl/min. Mass spectra were acquired in
   data-dependent mode with a 3-s cycle between each MS1 acquisition. The
   FTMS survey MS scan mass range was m/z 375 to 1575. A quadrupole
   isolation window of 1.6 was used for precursor ion selection. Tandem
   mass spectra (MS/MS) were acquired following higher-energy collisional
   dissociation of peptide precursors with 35% collision energy. Ions were
   detected in the orbitrap. The automatic gain control (AGC) target value
   was 4 × 10^5 for the survey MS scan at a resolution of 60,000 at m/z
   400. The AGC target value for the MS/MS scan was 5 × 10^4 with a
   maximum injection time of 22 ms. Precursors with charge states 2 to 7
   were selected for fragmentation. Dynamic exclusion was enabled with a
   repeat count of 1, an exclusion duration of 25 s, and 10 parts per
   million mass tolerance.

   To control for changes in protein expression, an aliquot of each
   uncombined, labeled protein sample was analyzed using a label-free
   proteomic approach (MaxQuant LFQ). Proteins were digested as above, and
   peptides were analyzed using a U3000 nano LC system coupled to an
   Orbitrap Elite MS (Thermo Fisher Scientific). Peptides were loaded onto
   a C18 PepMap 100 (300 μm × 5 mm) trap column for 10 min at 30 μl/min of
   solvent A and separated using a gradient of 5 to 40% solvent B in 180
   min at 200 nl/min with a 75 μm × 35 cm fused-silica column
   [ReproSil-Pur 120 C18 AQ 1.9 μm at 50°C (ESI Source Solutions, Woburn,
   MA)] packed in-house. Mass spectra were acquired in data-dependent mode
   using a top 10 method. Each FTMS survey scan was acquired with a mass
   range of m/z 400 to 1500 in the Orbitrap followed by acquisition of the
   tandem mass spectra in the ion trap. A normalized collision energy of
   35% was used with a 10-s activation time. The minimal signal for
   triggering acquisition of MS/MS was 500. Dynamic exclusion was enabled
   with a repeat count of 1, a repeat duration of 30 s, and an exclusion
   duration of 180 s.

   The combined biological replicate samples were searched in MaxQuant
   v2.0.1.0 (Max Planck Institute) using a reviewed mouse database
   containing 17,090 protein sequences downloaded from UniProt on 30
   November 2021. A strategy similar to that used for Stable Isotope
   Labeling by Amino Acids in Cell Culture (SILAC) experiments was used to
   set up the search. The labels introduced during sample lysis [either
   carbamidomethylation of Cys with light iodoacetamide (L-IAA) or heavy
   iodoacetamide (H-IAA)] were created in Andromeda within the MaxQuant
   platform to obtain ratios of heavy/light for the combined samples. The
   search was set up with a multiplicity of 2 (L-IAA and H-IAA). A
   database of contaminants was included in the search and a maximum of 2
   trypsin missed cleavages were allowed. Methionine oxidation was used as
   a variable modification and no fixed modifications were defined. A
   decoy database strategy was used as a threshold for identifications
   with an FDR of 0.01 at the spectrum, peptide, and protein levels. The
   minimum peptide length was 7 and a minimum ratio count of 2 was
   required for quantitation. Match between runs was enabled. The entries
   from the “peptides” text file were processed in Perseus v1.6.15.0 (Max
   Planck Institute). The peptide list was filtered to eliminate common
   contaminants, entries from the reversed, decoy database, and peptides
   that did not contain cysteine. The H/L normalized ratios calculated by
   MaxQuant were log[2] transformed and median normalized. Entries were
   filtered to retain peptides with ratios in each of the three biological
   replicate experiments. A t test was performed to test the null
   hypothesis by comparing the ratios to zero. The threshold for change in
   reactive thiol status was a Benjamini-Hochberg adjusted P < 0.05.

   For the global proteomic analysis, each of the labeled, uncombined
   samples (n = 6) were analyzed in triplicate. The database search was
   performed as above with the exception of using a multiplicity of 1 and
   the label-free quantitation algorithm. Methionine oxidation and protein
   N-terminal acetylation were used as variable modifications; no fixed
   modifications were defined because the cysteines were modified with two
   distinct reagents (L-IAA and H-IAA) during cell lysis and NEM during
   sample preparation. The protein groups text file was processed in
   Perseus v.1.6.15.0. The list of proteins was filtered to eliminate
   common contaminants and entries from the reversed, decoy database. The
   protein intensities were log[2] transformed and filtered to retain
   proteins quantified in three biological replicate samples of either the
   H[2]S-treated or control group. Missing values were imputed from a
   normal distribution with a width of 0.3 downshifted by 1.8. A t test
   was performed to compare the mean log[2] intensities of proteins from
   H[2]S treated with control. To correct for multiple hypothesis testing,
   a permutation-based FDR of 0.01 was used as the threshold for change.
   Peptides and proteins were annotated with GO terms (downloaded from
   UniProt on 19 October 2021) and Reactome Pathway Names (downloaded on 2
   October 2020) using Perseus. For visualization of the data volcano
   plots of the log[2] fold change in peptide or protein abundance versus
   the −log[10] P value were generated in Perseus.

Statistical analysis

   All data reported are the arithmetic mean from at least three
   independent experiments performed in triplicate ±SD unless stated
   otherwise. The unpaired Student’s t test was used to evaluate the
   significance of differences observed between groups, accepting P < 0.05
   as a threshold of significance. Data analyses were performed using the
   Prism software (GraphPad, San Diego, CA), except for tumor control
   experiments where all calculations were performed in RStudio version
   2023.06.1 using R-4.1.3. For all in vivo experiments with survival
   outcomes as the primary outcomes, it is expected based on preliminary
   data that 70% of the mice in the control group will be euthanized by 4
   weeks. For a treatment to be successful, it would be expected that only
   10% would be euthanized. A sample size of nine mice per group provides
   80% power to detect this difference with a two-sided α of 0.05 using a
   log-rank test. For all survival outcomes, Kaplan-Meier curves were used
   to display the results. Median survival time and the corresponding 95%
   confidence interval were calculated for each experimental condition. A
   log-rank test was used to compare the outcomes across experimental
   conditions. For all continuous outcomes, graphical displays (e.g., bar
   charts) were used to demonstrate patterns of the outcomes within and
   across experimental conditions. Normality and variance homogeneity
   assumptions were assessed, and appropriate data transformations were
   used. All continuous outcomes were measured longitudinally from the
   same animal. These measures were modeled using linear mixed-effects
   regression, including fixed effects for experimental condition, time,
   and their two-way interaction; subject-specific random effects were
   incorporated to account for the correlation among measures obtained
   from the same subject over time. Linear contrasts were used to conduct
   group comparisons at each time point for which three or more mice were
   alive in each treatment group. For a given time point, P values were
   adjusted for multiple comparisons using Holm’s method of correction.
   Statistical analyses were performed in a blinded fashion with the
   statisticians only having access to coded group identifiers without
   knowledge of the corresponding treatment groups.

Key resources

   Reagent information is listed in [245]Table 1.

Table 1. Reagent information.

   Resource Source Identifier
   Antibodies
   APC/Cyanine7 anti-mouse CD8a antibody BioLegend 100714
   PE anti-mouse CD8a antibody BioLegend 100708
   PE anti-human CD3 antibody BioLegend 34406
   PE anti-mouse CD11c antibody BioLegend 117308
   PE anti-mouse CD38 antibody BioLegend 102708
   PE anti-mouse CD62L antibody BioLegend 104407
   PE anti-mouse CD223 (LAG-3) antibody BioLegend 125207
   Pacific Blue anti-mouse CD11c antibody BioLegend 117322
   Brilliant Violet 711 anti-mouse CD279 (PD-1) antibody BioLegend 135231
   Brilliant Violet 650 anti-human CD34 antibody BioLegend 561539
   Alexa Fluor 700 anti-mouse Ly-6A/E (Sca-1) antibody BioLegend 108142
   Pacific Blue anti-mouse/rat/human CD27 antibody BioLegend 124218
   Pacific Blue anti-mouse/rat/human CD27 antibody BioLegend 124217
   PerCP/Cyanine5.5 anti-mouse CD223 (LAG-3) antibody BioLegend 125212
   PE/Cyanine7 anti-mouse CD279 (PD-1) antibody BioLegend 109110
   APC anti-mouse CD25 antibody BioLegend 102012
   APC anti-mouse CD62L antibody BioLegend 104412
   APC anti-mouse I-A/I-E antibody BioLegend 107613
   APC anti-mouse CD366 (Tim-3) antibody BioLegend 134008
   CD107a (LAMP-1) monoclonal antibody [eBio1D4B (1D4B)], eFluor 660
   eBioscience 51-1071-82
   FITC anti-mouse CD3 antibody BioLegend 100204
   PE/Dazzle 594 anti-human CD62L antibody BioLegend 556419
   FITC anti-mouse/human KLRG1 (MAFA) antibody BioLegend 138410
   FITC anti-mouse CD86 antibody BioLegend 105110
   PerCP/Cyanine5.5 anti-T-bet antibody BioLegend 644806
   FOXP3 monoclonal antibody (FJK-16s), eFluor 450 eBioscience 48-5773-82
   PE anti-TCF1 (TCF7) antibody BioLegend 564217
   PerCP/Cyanine5.5 anti-mouse IFN-γ antibody BioLegend 505822
   Alexa Fluor 700 anti-human/mouse granzyme B recombinant antibody
   BioLegend 372222
   PE/Cyanine7 anti-mouse TNFα antibody BioLegend 506324
   PE anti-mouse IL-2 antibody BioLegend 503808
   APC anti-human IL-10 antibody BioLegend 506807
   PE anti-mouse perforin antibody BioLegend 154406
   PE anti-human CD8 antibody BioLegend 344706
   Recombinant Alexa Fluor 647 anti-GM130 antibody [EP892Y] Abcam ab195303
   FITC anti-human CD3 antibody BioLegend 317306
   FITC anti-mouse CD279 (PD-1) antibody BioLegend 135214
   APC anti-human CD19 antibody BioLegend 302212
   PE/Cyanine7 anti-human CD8 antibody BioLegend 344750
   PE/Cyanine7 anti-human CD34 antibody BioLegend 343516
   PE/Cyanine7 anti-human CD279 (PD-1) antibody BioLegend 329918
   PE/Cyanine7 anti-human CD62L antibody BioLegend 304822
   APC/Cyanine7 anti-human CD8 antibody BioLegend 344714
   APC/Cyanine7 anti-human CD366 (Tim-3) antibody BioLegend 345026
   Brilliant Violet 711 anti-human CD223 (LAG-3) antibody BioLegend 369319
   PRDX4 Rabbit PolyAb Proteintech 10703-1-AP
   Anti-Giantin mouse Abcam ab37266
   Anti-Giantin rabbit Abcam ab37266
   Anti-Tom20 rabbit Santa Cruz Biotechnology sc-17764
   Phospho-S6 ribosomal protein (Ser^235/236)—Alexa647 Cell Signaling
   Technology 4851
   Phospho-PERK (Thr^980) Cell Signaling Technology 3179
   Anti-rabbit PE Jackson ImmunoResearch Laboratories 111-116-144
   Anti-rabbit Alexa 647 Abcam [246]Ab150079
   Anti-rabbit Alexa 488 Abcam [247]Ab150077
   Fetal bovine serum Atlanta Biologicals [248]S11150
   Bacterial and virus strains
   CBS ([249]NM_000071) human tagged ORF clone Origene RC201755
   Biological samples
   Healthy human PBMCs Research Blood Components LLC
   Chemicals, peptides, and recombinant proteins
   GYY4137 Cayman 13345
   NaHS MilliporeSigma 161527
   H[2]O[2] Sigma-Aldrich H1009
   Gp100 GenScript RP20344
   Recombinant IL-2 NCI, Biological Resources Branch
   [250]https://ncifrederick.cancer.gov/research/brb/productDataSheets/cyt
   okineHumanInterleukins/IL-2Bulk.aspx
   Recombinant IL-4 R&D Systems 404-ML-025
   Recombinant IL-6 BioLegend 575702
   Recombinant IL-10 BioLegend 575802
   Recombinant IL-12 BioLegend 573004
   Recombinant IL-15 PeproTech 200-15
   Recombinant TGFb BioLegend 580702
   Recombinant GM-CSF BioLegend 576304
   Monensin BioLegend 420701
   2-Deoxy-d-glucose (2DG) Sigma-Aldrich D6134
   Antimycin A Sigma-Aldrich A8674
   Rotenone Sigma-Aldrich R8875
   Oligomycin Sigma-Aldrich O4876
   FCCP Sigma-Aldrich C2920
   IMDM GE Healthcare, HyClone SH30228.01
   PST3.1a (NSC-753859) Aobious [251]AOB13789
   Critical commercial assays
   TET Hydroxylase Activity Quantification Kit (Fluorometric) Abcam
   ab156913
   WSP-1 Cayman 11179
   CellTrace Violet Cell Proliferation Kit Thermo Fisher Scientific
   [252]C34557
   Click-iT Plus OPP Alexa Fluor 647 Protein Synthesis Assay Kit Thermo
   Fisher Scientific [253]C10458
   FITC annexin V Apoptosis Detection Kit I BD Pharmingen 556547
   7-AAD Thermo Fisher Scientific A1310
   MitoTracker Deep Red FM Thermo Fisher Scientific [254]M22426
   Tetramethylrhodamine, methyl ester, perchlorate (TMRM) Thermo Fisher
   Scientific T668
   MitoSOX mitochondrial superoxide indicators Thermo Fisher Scientific
   [255]M36008
   ThiolTracker Violet (glutathione detection reagent) Thermo Fisher
   Scientific [256]T10095
   Alexa Fluor 633 C5 maleimide Thermo Fisher Scientific [257]A20342
   BODIPY FL C5-ceramide
   (N-(4,4-difluoro-5,7-dimethyl-4-bora-3a,4a-diaza-s-indacene-3-pentanoyl
   )sphingosine) Thermo Fisher Scientific D3521
   H2DCFDA Thermo Fisher Scientific D399
   NaveniFlex Cell MR RED Cayman Chemical 39505
   iScript cDNA Synthesis Kit Bio-Rad 1708891
   SsoAdvanced Universal SYBR Green Supermix Bio-Rad 1725274
   NE-PER nuclear and cytoplasmic extraction reagents Thermo Fisher
   Scientific 78833
   Nucleofector kits for mouse T cells Lonza VPA-1006
   Fixation/Permeabilization Solution Kit BD Biosciences 554714
   Foxp3/Transcription Factor Staining Buffer Set Thermo Fisher Scientific
   00-5523
   LIVE/DEAD Fixable Yellow Dead Cell Stain Kit Thermo Fisher Scientific
   [258]L34959
   RT^2 profiler PCR arrays Qiagen PARN-034Z, PAZF-014Z
   Tumor dissociation kit, mouse Miltenyi 130-096-730
   Deposited data
   Raw and analyzed RNA-seq data This paper GEO: GSE249287
   Raw and analyzed proteomics data This paper PRIDE: PXD048141
   Experimental models: Cell lines
   Jurkat ATCC TIB-152
   Raji ATCC CCL-86
   B16-F10 ATCC CRL-6475
   Experimental models: Organisms/strains
   Mouse: C57BL/6 000664
   Mouse: B6-Rag−/− 034159
   Mouse: Pmel 005023
   Mouse: NSG 005557
   Mouse: Cbs−/− 002461
   Oligonucleotides
   Prdx4 siRNA Ambion AM16706
   Primer: CBS, forward: GATCGCCAGAAAGCTGAAGGAG, reverse:
   CCACCTCATAGGCTGTTTGCTC Invitrogen A15612T
   Primer: CSE, forward: GTGGGACAAGAGCCTGAGCAAT, reverse:
   GGATTTCCAGAGCGGCTGTATTC Invitrogen A15612T
   Primer: GCLM, forward: CCTCGCCTCCGATTGAAGATG, reverse:
   AAAGGCAGTCAAATCTGGTGG Invitrogen A15612T
   Primer: GCLC, forward: GGACAAACCCCAACCATCC, reverse:
   GTTGAACTCAGACATCGTTCCT Invitrogen A15612T
   Prdx4 shRNA: pLV[shRNA]-Puro-U6>hPRDX4[shRNA#1] VectorBuilder
   MaxH(VB230622-1500vrc)
   Prdx4 parent plasmid: pLV[Exp]-Neo-CMV>EGFP/hPRDX4[[259]NM_006406.2]
   VectorBuilder MaxH(VB230622-1499hxg)
   Prdx4 mutant primer: forward: CGGACTCGCGAAGAGGAGGCCCACTTCTAC, reverse:
   CCACCCGCGTAGAAGTGGGCCTCCTCTTC Invitrogen A15612T
   Primer: MGAT1, forward: CCTATGACCGAGATTTCCTCGC, reverse:
   TGAAGCTGTCCCTGCCCGTATA Invitrogen A15612T
   Primer MGAT5A, forward: AGGCAGAACCAGTCCCTTGTGT, reverse:
   CTTTGTGCTGGAGCCATAAACAG Invitrogen A15612T
   Primer MGAT5B, forward: CTCTTACCGCAGCCTGAGTTTC, reverse:
   GCAGGAAGATGCAACCATTGGC Invitrogen A15612T
   Software and algorithms
   FlowJo 10.2 TreeStar, OR
   [260]https://flowjo.com/solutions/flowjo/downloads/
   Prism 8 GraphPad [261]https://graphpad.com/scientific-software/prism/
   Agilent Seahorse Wave 2.4 Agilent
   [262]http://agilent.com/en-us/products/cell-analysis-(seahorse)/seahors
   e-wave-software
   CFX Manager 3.1 Bio-Rad
   [263]https://bio-rad-cfx-manager.software.informer.com/3.1/
   MaxQuant (MQ) v. 2.0.1.0 Max Planck Institute PRIDE PXD048141
   Perseus v. 1.6.15.0 Max Planck Institute PRIDE PXD048141
   [264]Open in a new tab

Acknowledgments