Gpx4 ^fl/fl (027964), ERT2Cre (010688), and wild-type (WT) CD45.2⁺ and CD45.1⁺ C57BL/6J (002014) congenic mice were obtained from the Jackson Laboratory. P14 transgenic mice with CD8⁺ T cells specific for LCMV GP_33–41 epitope and LCMV strains Armstrong (Arm) and LCMV Clone-13 (Cl13) were provided by R. Ahmed (Emory University). Intraperitoneal (i.p.) injection of 2 × 10⁵ plaque-forming units (PFU) of the LCMV Arm established acute viral infection in mice. Intravenous (i.v.) injection of 2 × 10⁶ PFU of LCMV Cl13 established chronic viral infection in mice. Mice were infected at 6–10 weeks of age unless otherwise stated. All mice were housed in specific pathogen-free conditions at 25°C and 55% humidity on a 12-h dark/12-h light cycle.

For acute viral infection, a total of 2 × 10⁴ naive P14 (CD45.1⁺) cells were adoptively transferred into naive wild-type (WT) congenic recipient mice (CD45.2⁺). On the following day, the recipients were infected i.p. with 2 × 10⁵ PFU of the LCMV Arm. For chronic viral infection, a total of 2 × 10³ naive P14 (CD45.1⁺) cells were adoptively transferred into naive WT congenic recipient mice (CD45.2⁺). On the following day, the recipients were infected i.v. with 2 × 10⁶ PFU of LCMV Cl13. In some cases, Gpx4 ^fl/fl-ERT2Cre-P14 and WT-P14 cells were adoptively co-transferred at a ratio of 1:1 into congenic recipient mice and received LCMV Cl13 infection the next day.

Bone marrow (BM) cells were collected from WT B6 mice (CD45.2⁺) and Gpx4 ^fl/fl-ERT2Cre (KO) mice (CD45.1⁺). A total of 6 × 10⁶ mixed BM cells were intravenously co-transferred at a KO to WT ratio of 3:7 into lethally irradiated WT recipients (CD45.2⁺), which received two doses of sublethal irradiation (550 rad). Chimeras were fed with antibiotic-containing water during the first 4 weeks and allowed a total of 8–10 weeks for reconstitution before infection.

Tamoxifen (TAM, T5648, Sigma-Aldrich, Darmstadt, Germany) and its active metabolites, 4-hydroxy tamoxifen (4-OHT, H6278, Sigma-Aldrich, Darmstadt, Germany), were employed to inactivate GPX4 in T cells from Gpx4 ^fl/fl-ERT2Cre mice in vivo and in vitro, respectively. For in vivo GPX4 deletion, TAM was prepared and administrated as previously described (17). For in vitro GPX4 deletion, enriched CD8⁺ T cells from Gpx4 ^fl/fl-ERT2Cre-P14 mice were treated with 5 μM 4-OHT for 36 h before subjected to adoptive transfer.

Liproxstatin-1 (Lip-1; S7699, Selleck, Houston, TX, USA), a potent peroxyl radical scavenger to antagonize ferroptosis, was administrated i.p. at 30 mg/kg/day for 10 days. Mice in the control group received sham treatment with vehicle during the same period.

Flow cytometry data were acquired with a FACSCanto II or LSRFortessa (BD Biosciences, La Jolla, CA, USA) and were analyzed with Flowjo software (BD Biosciences, La Jolla, CA, USA). Major histocompatibility complex class I (H-2Db) tetramer specific for the LCMV epitope of GP_33–41 was provided by a tetramer core facility of the US National Institutes of Health (Emory, Atlanta, GA, USA). The antibodies used in flow cytometry are listed as follows: anti-CD8 (clone 53-6.7), anti-CD45.1 (clone A20), anti-CD45.2 (clone 104), anti-CD44 (clone IM7), anti-TCR Vα2 (clone B20.1), anti-PD-1 (clone 29F.1A12), anti-LAG3 (clone C9B7W), and anti-Ki-67 (clone 11F6) were purchased from Biolegend (San Diego, CA, USA); anti-TIM3 (clone RMT3-23) and anti-CD39 (clone 24DMS1) were purchased from eBioscience (San Diego, CA, USA); anti-GPX4 (clone EPNCIR144]) was purchased from Abcam (Cambridge, UK). Cell surface staining was performed in FACS buffer. Staining of GPX4 was performed with a BD Cytofix/Cytoperm kit (554714, BD Biosciences, La Jolla, CA, USA) after cell surface staining. Staining of Ki-67 was performed with a Foxp3/Transcription kit (554714, BD Biosciences, La Jolla, CA, USA) after cell surface staining. Staining of apoptosis was conducted using a commercial apoptosis detection kit (559763, BD Pharmingen, San Diego, CA, USA) following manufacturer’s instructions.

Cells were stained with cell surface markers before being subjected to molecular probe or mitochondrial dye staining in R10 (RPMI medium containing 10% fetal bovine serum) at 37°C under 5% CO₂ for 30 min, unless otherwise stated. To detect mitochondrial mass, cells were further stained with 100 nM MitoTracker Green (M7514, Invitrogen, Carlsbad, CA, USA) in PBS (calcium and magnesium free). To measure lipid peroxidation, cells were further stained with 5 μM BODIPY 581/591 C11 (BODIPY-C11, D3861, Invitrogen, Carlsbad, CA, USA) or 4 μM Liperfluo (L248, Dojindo, Mashiki, Tabaru, Japan) in PBS. In some scenario, cells were treated with PBS or hydrogen peroxide (H₂O₂, 1 mM) for 1 h at room temperature before staining with BODIPY-C11. For ROS or mtROS detection, cells were loaded with 5 μM CellROX Deep Red (C10422, Invitrogen, Carlsbad, CA, USA) or 5 μM MitoSOX (M36008, Invitrogen, Carlsbad, CA, USA). To determine mitochondria membrane potential changes, cells were stained with 100 nM TMRE (87917, Sigma-Aldrich, Darmstadt, Germany). Cells were then resuspended in FACS buffer and subjected to flow cytometric analysis.

Fluorescence-activated cell sorting (FACS) was performed on a BD FACSAria III cell sorter (BD Biosciences, La Jolla, CA, USA) at the Clinical Medical Research Center of Southwest Hospital, Third Military Medical University. T_N (CD44^loCD62L^hi, or CD44^loCD62L^hiVα2⁺), T_EFF and T_MEM (CD44^hi, or CD44^hiCD45.1⁺), and T_EX (CD44^hiPD-1^hi, or CD44^hiPD-1^hiCD45.1⁺) cells were sorted at indicated timepoints post LCMV Arm or Cl13 infection after CD8⁺ T cell enrichment via depletion of cells positive for lineage markers through the use of biotin-conjugated antibodies [anti-CD4 (RM4-5, Biolegend, San Diego, CA, USA), anti-B220 (RA3-6B2, Biolegend, San Diego, CA, USA), anti-CD25 (PC61, Biolegend, San Diego, CA, USA), anti-CD11b (M1/70, Biolegend, San Diego, CA, USA), anti-CD11c (N418, Biolegend, San Diego, CA, USA), anti-Gr-1 (RB6-8C5, Biolegend, San Diego, CA, USA), anti-TER119 (TER-119, Biolegend, San Diego, CA, USA), anti-CD49b (DX5, Invitrogen, Carlsbad, CA, USA), and anti-NK1.1 (PK136, Biolegend, San Diego, CA, USA)] coupled to BeaverBeads Mag500 Streptavidin Matrix (22302, Beaver, Suzhou, Jiangsu, China). The population purity after cell sorting was verified >95% in all experiments. Sorted CD8⁺ T cells were then subjected to RNA analysis, immunoblot analysis, and glutathione peroxidase activity assay.

FACS-sorted T_N, T_EFF, T_MEM, and T_EX cells were used to conduct glutathione peroxidase activity assay using a Glutathione Peroxidase Assay kit (ab102530, Abcam, Cambridge, UK) according to manufacturer’s instructions.

Total RNA was sorted from FACS-sorted T_N, T_EFF, T_MEM and T_EX cells using a Trizol-based approach, followed by reverse transcription (RT) via a commercial RT kit (R212, Vazyme, Nanjing, Jiangsu, China). The resulting cDNA was analyzed with a SYBR Green PCR kit (208054, Qiagen, Tegelen, Hulsterweg, Netherlands) on a CFX96 Touch Real-Time system (Bio-Rad, Hercules, CA, USA). Here are the primers used in the study: Gpx4 (Gpx4-F, 5′-CGCGATGATTGGCGCT-3′; Gpx4-R, 5′-CACACGAAACCCCTGTACTTATCC-3′) and house-keeping gene Hprt (Hprt-F, 5′-TCAGTCAACGGGGGACATAAA-3′; Hprt-R, 5′-GGGGCTGTACTGCTTAACCAG-3′).

Viral load in specific tissues (spleen and brain) of LCMV Cl13-infected mice were determined as previously described with minor modifications (30). Briefly, indicated tissues were weighted and homogenized with Trizol for RNA extraction, followed by RT using the aforementioned commercial RT kit (R212, Vazyme) with an LCMV-specific glycoprotein (GP) primer (GP-R, 5′-GCAACTGCTGTGTTCCCGAAAC-3′). The qPCR assay was conducted with LCMV GP-specific primer pair (GP-F, 5′-CATTCACCTGGACTTTGTCAGACTC-3′; GP-R, 5′-GCAACTGCTGTGTTCCCGAAAC-3′). The viral load was determined and presented as PFU per gram as previously described (31).

We retrieved publicly accessible RNA-sequencing data of T_EFF or T_EX P14 cells from the mice infected with LCMV Arm or Cl13 (32), from GEO database (GEO accessible number, GSE119942). The data were subjected to gene set enrichment analysis (GSEA) (Broad Institute) with gene sets from the Molecular Signatures Database or published data (17, 22, 33, 34) using the official GESA application or R script.

FACS-sorted T_EFF, T_MEM, and T_EX cells were subjected to Western blot (WB) as previously described with minor modifications (17). Briefly, sorted cells were lysed in RIPA buffer, and thereafter, extracted protein samples were boiled with SDS sample buffer. Equal amounts of protein were separated by 15% SDS-PAGE and transferred onto a PVDF membrane (ISEQ00010, Millipore, Burlington, MA, USA). After blocking with 5% bovine serum albumin, membranes were probed overnight with primary antibodies anti-GPX4 (ab125066, Abcam, Cambridge, UK) or anti-β-Actin (4970S, Cell Signaling Technology, Danvers, MA, US) at 4°C. The membranes were probed with horseradish peroxidase-conjugated antibody to rabbit IgG (7074, Jackson ImmunoResearch Laboratories, West Grove, PA, USA) after cleanup of unbounded primary antibody. Membranes were then washed four times before visualized with a SuperSignal West Femto Maximum Sensitivity Substrate (34094, Thermo Fisher Scientific, Waltham, MA, USA) using a Fusion imaging system (FUSION Solo S, VILBER, Marne-la-vallée, Ile-de-France, France).

Previously, retroviruses that overexpress any isoform of Gpx4 have been constructed based on their intracellular location, including MIGRA-cGPX4 (cGPX4-OE), MIGRA-mGPX4 (mGPX4-OE), and MIGRA-nGPX4 (nGPX4-OE) (17). Briefly, cGPX4 locates in the cytoplasm and nucleus, mGPX4 locates in the mitochondria, and nGPX4 locates in the nucleus (35). Retroviruses knocking down Gpx4 were constructed based on the backbone pMKO.1-GFP (shCTRL). Each construct incorporated the sequences of a unique shRNA targeting Gpx4 (pMKO.1-shGPX4-GFP, shGPX4 for short). Sequences of shRNA are as follows:

shGPX4-1: 5′-GACGTAAACTACACTCAGCTA-3′;

shGPX4-2: 5′-CAAATGGAACTTTACCAAGTT-3′;

shGPX4-3: 5′-CCCACTGTGGAAATGGATGAA-3′;

shGPX4-4: 5′-GCCAGGAAGTAATCAAGAAAT-3′.

All sequences were verified by DNA sequencing (Tsingke Biotech, Beijing, China). Retroviruses were packaged with 293T cells using package plasmid pCL-Eco and indicated retroviral constructs. In vivo activated P14 cells by GP_33–41 peptide were spin transduced for 120 min at 37°C by centrifugation (2,100 rpm) with freshly collected retroviral supernatants, 8 μg/ml polybrene (H9268, Sigma-Aldrich, Darmstadt, Germany), and 20 ng/ml IL-2 (130-098-221, Miltenyi Biotec, Bergisch Gladbach, Germany). Then, the transduced P14 cells were sorted and transferred into recipient mice, followed by infection of the hosts with LCMV Cl13 or Arm the next day.

Statistical analysis was conducted with Prism 9.2.0 software (GraphPad). Flow cytometric data analysis was conducted using FlowJo software (BD Biosciences, La Jolla, CA, USA). Quantifications of WB results were conducted using Fiji image analysis software (ImageJ 1.54g, NIH). For comparisons between two independent groups, significance was determined by two-tailed unpaired and paired Student’s t-tests as indicated in different settings. A p < 0.05 was considered statistically significant. Asterisks were used to indicate significance: p < 0.05*, p < 0.01**, p < 0.001***, and p < 0.0001****, and ns, not significant.

We first leveraged two LCMV strains, Cl13 and Arm, and P14 mice expressing TCR specific recognizing the LCMV GP_33–41 epitope to facilitate the investigation of virus-specific CD8⁺ T cells. CD45.1⁺ naive P14 cells isolated from P14 mice were adoptively transferred into congenic CD45.2⁺ recipient C57BL/6 (B6) mice, followed by infection with LCMV Cl13 or LCMV Arm 1 day later to establish either chronic or acute viral infection, respectively ( Figure 1A ). This experimental setup enables. At indicated timepoints post-infection (D8 and D15), parallel comparison of virus-specific T_EFF and T_EX cells was conducted via flow cytometric analysis. The gating strategy of P14 cells is illustrated as in Supplementary Figure S1 . Exhausted state of T_EX P14 cells during chronic infection was verified by significantly higher expression of PD-1 compared to that in T_EFF P14 cells ( Figure 1B ). As expected, T_EX P14 cells from chronically infected mice manifested increased apoptosis (Annexin V⁺ or Annexin V⁺7-AAD⁻, Supplementary Figures S2A, B ) and enhanced cell death (7-AAD⁺, Supplementary Figure S2C ) relative to T_EFF P14 cells in acutely resolved infected mice.

Furthermore, we applied BODIPY-C11, a molecular probe specifically detecting the accumulation of peroxided lipids, to examine the ferroptosis of virus-specific CD8⁺ T cells. Compared to functional T_EFF P14 cells, T_EX P14 cells displayed a pronounced accumulation of peroxided lipids, irrespective of whether they were under physiological conditions (PBS-treated) ( Figure 1C ) or under excessive oxidative stress (H₂O₂ treated) ( Figure 1D ), indicating the occurrence of ferroptosis in exhausted P14 cells at 8 and 15 days post-infection (8 dpi and 15 dpi). Furthermore, a progressively accumulated lipid peroxidation was observed in T_EX cells at 15 dpi in comparison to that at 8 dpi under excessive oxidative stress ( Figure 1D ), implying that terminally differentiated T_EX P14 cells may be more susceptible to ferroptosis. In addition, the intensity of Liperfluo in terminally exhausted (Tim3^hi) population was found to be notably higher than in the less exhausted (Tim3^lo) counterpart ( Figure 1E ). In this scenario, we observed that mice from the Cl13–D15 group displayed a significantly lower frequency and number of P14 cells in both PBMCs ( Figure 1F ) and spleens ( Figure 1G ), compared to those from the Arm-D15 group.

Moreover, using publicly available bulk RNA sequencing data (32), we performed GSEA to compare the transcriptional differences between virus-specific T_EFF and T_EX cells with respect to ferroptosis-associated signatures (17, 22, 33, 34). GSEA results revealed that gene signatures of ferroptosis were more enriched in T_EX cells relative to T_EFF cells. ( Figure 1H ).

To conclude, our findings indicate that in addition to apoptosis, ferroptosis may also be responsible for the clonal depletion of virus-specific T_EX cells.

To further confirm that ferroptosis accounted for the reduction in T_EX cells, we set out to testify whether pharmacological inhibition of ferroptosis could alleviate the decline of virus-specific T_EX cells. CD45.1⁺ naive P14 cells were adoptively transferred into congenic CD45.2⁺-recipient B6 mice, and the hosts were subsequently infected with LCMV Cl13 the following day to establish chronic viral infection. Recipient mice were administrated with either the ferroptosis inhibitor liproxstatin-1 (Lip-1) or vehicle via the i.p. route between 15 dpi and 24 dpi and were subsequently euthanized at 25 dpi ( Figure 2A ). We first verified that Lip-1 treatment could efficiently inhibit ferroptosis by reducing the accumulation of lipid peroxides ( Figure 2B ). Compared to vehicle treatment, Lip-1 treatment resulted in markedly elevated frequency of P14 cells in PBMCs (~ 2-fold, 25 dpi), which was comparable in both Lip-1 and vehicle groups prior to treatment (15 dpi) ( Figure 2C ). In comparison to vehicle treatment, a significantly higher proportion and larger absolute number of P14 cells were consistently observed in the spleens of mice from the Lip-1 treatment group (~1.5-fold, 25 dpi) ( Figure 2D ). Moreover, Lip-1 treatment significantly restrained the terminal differentiation of T_EX P14 cells by reducing the expression of inhibitory receptors, including TIM3, CD39, and LAG3 ( Figure 2E ). Accordingly, the administration of Lip-1 resulted in a notable reduction in the viral load in both lymphoid (spleen) and non-lymphoid (brain) tissues ( Figure 2F ).

Collectively, we corroborate the role of ferroptosis in driving the clonal deletion of T_EX cells and unravel that selective inhibition of ferroptosis considerably increases both the frequency and cell number of virus-specific T_EX cells in the settings of chronic viral infection and ultimately results in a better viral control of hosts.

Compared to functional T_EFF cells, T_EX cells generally manifested imbalanced redox states and dysfunctional mitochondria (4, 36). We speculated that aberrant redox balance and dysfunctional mitochondria may trigger ferroptosis in T_EX cells. To this end, we implemented the acute and chronic LCMV infection model as in Figure 1A and set out to analyze the differences in cellular oxidative stress and functional markers of mitochondria between T_EFF and T_EX cells using flow cytometry. In comparison with T_EFF P14 cells, T_EX P14 cells manifested a markedly elevated intensity of the fluorescent probe CellROX, indicative of a heightened level of intracellular ROS ( Figure 3A ). GSEA results also revealed enrichment of signatures associated with the cellular response to ROS in T_EX P14 cells ( Figure 3B ). Furthermore, as indicated by the fluorescent probe MitoSOX, an increased mtROS was observed in T_EX P14 cells rather than in T_EFF P14 cells ( Figure 3C ). Aberrant accumulation of ROS, especially mtROS, suggested a poor cellular fitness and mitochondrial dysfunction in T_EX P14 cells (36). Furthermore, we monitored the mitochondrial membrane potential (MMP) of T_EX P14 cells using the fluorescent probe tetramethylrhodamine ethyl ester (TMRE) and discovered a remarkably decreased MMP of T_EX P14 cells at later timepoints (Cl13 D15), echoing the increased apoptosis and the mitochondrial dysfunction in T_EX P14 cells ( Figure 3D ). In addition, we measured mitochondrial mass using the fluorescent probe MitoTracker Green, finding a marked accumulation of dysfunctional mitochondria in T_EX P14 cells ( Figure 3E ).

Taken together, we speculated that aberrant redox balance and accumulated dysfunctional mitochondria may contribute, at least in part, to the ferroptosis of virus-specific T_EX cells during LCMV Cl13 infection.

As a critical enzyme preventing the accumulation of lipid peroxidation, GPX4 plays a pivotal role in antagonizing ferroptosis (22, 23). We set out to evaluate the content and activity of GPX4 in T_EX cells in comparison to other T-cell subsets during viral infection. To this end, T_N, T_EFF (Arm-D8 or Arm-D15), T_MEM (Arm-D330), and T_EX (Cl13-D14) P14 cells were sorted to conduct real-time quantitively PCR (RT-qPCR) assay to detect the transcription levels of the Gpx4 gene. We observed that Gpx4 transcription was significantly repressed in T_EX P14 cells compared to those in T_N, T_EFF, and T_MEM P14 cells ( Figure 4A ). In addition, we examined the protein level of GPX4 using both flow cytometry and Western blot and discovered that the protein expression of GPX4 was severely decreased in T_EX cells compared to that in T_EFF or T_MEM cells ( Figures 4B–D ). Moreover, a significant reduction in GPX4 enzymatic activity was observed in T_EX cells (Cl13-D7 or Cl13-D15) in comparison to that in T_N cells ( Figure 4E ). In contrast, no significant difference was observed between T_EFF and T_N cells or between T_MEM and T_N cells ( Figure 4E ).

To conclude, the expression of Gpx4 and the enzymatic activity of GPX4 were markedly impeded in T_EX cells during chronic viral infection, which may contribute to the ferroptotic cell death of T_EX cells.

To assess the significance of GPX4 in the prolongation of T_EX cells during chronic viral infection, we utilized a retrovirus-mediated gene knockdown (KD) approach to specifically knock down GPX4 in P14 cells. Four retroviral constructs were created based on the backbone pMKO.1-GFP (shCTRL). Each construct carries a unique shRNA targeting Gpx4 (pMKO.1-shGPX4-GFP, shGPX4 for short). This methodology is illustrated in detail in Figure 5A . Both shCTRL-transduced P14 cells (CD45.1⁺) and shGPX4-transduced P14 cells (CD45.1⁺CD45.2⁺) were sorted and adoptively co-transferred at a ratio of 1:1 into the recipient mice (CD45.2⁺), which were subsequently infected i.v. with LCMV Cl13 1 day later. Both PBMCs and splenocytes from recipients were subjected to flow cytometric analysis at 7 dpi ( Figure 5B ). We first validated that all the four shRNAs can efficiently reduce Gpx4 transcription with a percent decrease ranging from 70% to 90% ( Figure 5C ). We also validated that the abovementioned four shRNAs can efficiently reduce Gpx4 protein levels from the in vitro results ( Supplementary Figures S3A, B ). We then confirmed that GPX4-KD resulted in a substantial reduction in P14 cell frequency in PBMCs ( Figures 5D, E ). Additionally, both the frequency and cell number of P14 cells were observed to decline consistently in spleens from recipients in GPX4-KD group at 7 dpi ( Figures 5F–H ). However, GPX4-KD did not alter the expression of inhibitory receptors, such as PD-1 and CD39 ( Figure 5I ).

We also checked whether knockdown of Gpx4 impaired cell proliferation in vitro. The ratio of shGPX4-transduced P14 cells to shCTRL-transduced P14 cells was significantly decreased at D4 compared to that at D0 ( Supplementary Figure S3C ). Moreover, GPX4-KD resulted in a decreased frequency of Ki-67⁺ P14 cells and a reduced expression of Ki-67 ( Supplementary Figure S3D ). To summarize, GPX4-KD significantly impaired cell proliferation of P14 cells in vitro.

In conclusion, we affirmed that the partial loss of GPX4 expression via GPX4-KD in P14 cells significantly exacerbated the decline of virus-specific T_EX cells.

To further substantiate the indispensability of anti-ferroptosis function of GPX4 in the persistence of virus-specific T_EX cells, Gpx4 ^fl/fl-ERT2Cre-P14 (KO-P14) mouse was bred to facilitate the tamoxifen (TAM)/4-OHT-inducible knockout of Gpx4 in virus-specific P14 cells. P14 cells from KO-P14 mice were treated with 5 μM 4-OHT in vitro for 36 h to acutely delete Gpx4. Meanwhile, P14 cells from WT-P14 mice were subjected to 4-OHT treatment as a control ( Figure 6A ). We adoptively co-transferred 4-OHT-treated WT-P14 cells and 4-OHT-treated KO-P14 cells at a ratio of 1:1 into the recipient mice, then infected these mice with LCMV Cl13. Mice were euthanized at 21 dpi and subjected to flow cytometric analysis ( Figure 6A ). Phenocopied with T_EX P14 cells by GPX4-KD, GPX4-KO markedly reduced both the frequency and cell number of T_EX P14 cells ( Figure 6B ). We testified that GPX4-KO significantly increased BODIPY-C11 intensity ( Supplementary Figures S4A, B ), indicative of an augmented ferroptosis. Consequently, a pronounced ferroptosis brought a more exhausted state of P14 cells, as evidenced by a heightened frequency of PD-1⁺ P14 cells and upregulated PD-1 expression ( Figure 6C ). In addition, we observed a significant reduction in cell proliferation in KO-P14 cells relative to WT cells, indicated by a diminished frequency of Ki-67⁺ P14 cells and deceased Ki-67 expression ( Figure 6D ), underscoring a vital role of GPX4 in the renewal and persistence of T_EX P14 cells.

Next, we sought to analyze the impact of deletion of GPX4 by means of TAM administration in established T_EX P14 cells during chronic viral infection ( Figure 6E ). We validated that GPX4-KO faithfully increased the intensity of BODIPY-C11 in T_EX P14 cells at 5 days post TAM treatment ( Supplementary Figure S4C ). As expected, the frequency of KO-P14 cells was markedly diminished in PBMCs relative to that in WT cells, resulting in a notable reduction in the KO to WT ratio ( Figure 6F ). Consistently, both the frequency and cell number of KO-P14 cells (KO) exhibited a pronounced decrease in spleens compared to those in WT counterparts ( Figures 6G, H ).

Moreover, to validate the role of GPX4 and ferroptosis in endogenous T_EX cells, we constructed bone marrow chimeras (BMCs) by transferring bone marrow cell mixtures from Gpx4 ^fl/fl-ERT2Cre (CD45.1⁺) and B6 (CD45.2⁺) mice into the lethally irradiated recipient mice (CD45.2⁺). Eight weeks post-reconstitution, BMCs were infected with LCMV Cl13, followed by administration of TAM from 8 dpi to 12 dpi, and mice were euthanized at 8 days post TAM treatment for flow cytometric analysis ( Supplementary Figure S5A ). We observed that the amount of BODIPY-C11 was more enriched in KO polyclonal PD1^hi CD44^hi T_EX cells than in WT ones ( Supplementary Figure S5B ), coinciding with a dramatically decreased ratio of KO to WT in PBMC-derived polyclonal PD1^hi CD44^hi CD8⁺ T cells and in endogenous virus-specific splenic CD44^hi GP33⁺ CD8⁺ T cells ( Supplementary Figures S5C, D ).

To summarize, we depicted that GPX4 was intrinsically indispensable for the persistence of T_EX cells by antagonizing ferroptosis.

The above results illustrated a profound role of GPX4 and ferroptosis in the clonal deletion of virus-specific T_EX cells, encouraging us to examine whether GPX4 overexpression (GPX4-OE) could mitigate the decline of T_EX P14 cells by preventing ferroptosis. We used a set of GPX4-OE retroviruses (17) to overexpress each of three GPX4 isoforms (cGPX4, mGPX4, or nGPX4) in P14 cells, and those transduced P14 cells (mAm⁺) were transferred separately into recipient mice, followed by LCMV Cl13 infection ( Figure 7A ). Mice were euthanized at the early (D8) and late phase (D20) post-infection, respectively, and splenocytes were subjected to flow cytometric analysis ( Figure 7A ). At 8 dpi, we verified that all of the three isoforms of GPX4 can significantly reduce the intensity of BODIPY-C11 on transduced mAm⁺ P14 cells ( Figure 7B ). However, no notable discrepancies in the frequency or cell number of P14 cells were discerned between any of the three GPX4-OE groups (cGPX4, mGPX4, or nGPX4) and the control group (Vector) at this timepoint ( Figure 7C ). We also observed a comparable expression level of inhibitor receptors (e.g., PD-1, TIM3, CD39, and LAG3) on P14 cells between GPX4-OE and vector group ( Figure 7D ). Correspondingly, no notable distinctions were observed in the viral load in both lymphoid (spleen) and non-lymphoid tissues (brain) ( Supplementary Figure S6A ).

At the late stage of infection (20 dpi), our results demonstrated that GPX4-OE markedly attenuated the accumulation of lipid peroxidation ( Figure 7E ). Notably, mGPX4-OE rather than cGPX4-OE or nGPX4-OE substantially increased both the frequency and cell number of P14 cells ( Figure 7F ). Interestingly, apart from mGPX4-OE, cGPX4-OE can also significantly reduce the viral load in both lymphoid and non-lymphoid tissues, compared to control treatment ( Supplementary Figure S6B ), hinting an effective improvement of effector function induced by cGPX4-OE despite of the unaltered cell numbers. Similar to that observed at D8, no notable differences were identified in the expression of aforementioned inhibitory receptors between GPX4-OE and vector group ( Figure 7G ), suggesting that GPX4-OE mitigates the decay of T_EX P14 cells primarily on the replenishment of P14 cell pool.

In summary, overexpression of mGPX4 can substantially safeguard virus-specific T_EX cells against ferroptosis, ensuring a better persistence of T_EX cells, and conferring a superior control of the virus.