GSK-3 regulates CD4-CD8 cooperation for super-armed CD8+ cytolytic T cells in immunotherapy against tumors

GSK-3 KD mice show increased T-cell memory

To more fully define the role of GSK-3 in T-cell immunity, we generated mice with markedly reduced levels of GSK-3α and GSK-3β in mature T cells (referred to GSK-3 KD mice). This was achieved by crossing mice carrying the distal promoter lck-Cre (dlck-Cre) with mice carrying floxed alleles for gsk-3α (gsk-3αflox/flox) and gsk-3β (gsk-3βflox/flox). The distal lck promoter drives Cre recombinase expression in peripheral T cells, allowing for gene deletion specifically in mature T cells after their development in the thymus.⁶⁴ Immunoblotting with an anti-GSK-3α/β antibody showed a reduced expression of both the GSK-3α and GSK-3β isoforms in splenic CD8+ and CD4 + T cells (Fig. 1a). Although the expression of both isoforms was reduced, the GSK-3β isoform expression was completely lost, while the expression of the GSK-3α isoform was partially reduced.

The alternative text for this image may have been generated using AI.

GSK-3 knockdown regulates CD8⁺ T-cell progenitor maintenance and limits exhaustion during chronic viral infection. a Immunoblot analysis of GSK-3α and GSK-3β protein expression in splenic CD4⁺ and CD8⁺ T cells isolated from wild-type (WT) control and GSK-3 KD mice (n = 3). b Body weight curves following intravenous infection with LCMV Clone 13 (Cl-13; 2 × 10⁶ PFU). GSK-3 KD mice recovered more rapidly and maintained significantly higher body weights from day 10 onwards (n = 14 per group; data pooled from three independent experiments). c Splenic LCMV Cl13 viral titres measured by plaque assay in kidney homogenates from WT and GSK-3 KD mice at day 30 p.i. (n = 7–8). d Absolute counts of splenic CD8⁺ T cells in LCMV Cl13-infected mice at days −1, 7, and 30 p.i. (n = 4). e Number of GP33-specific (H-2D^b–KAVYNFATM–BV421 tetramer-positive) CD8⁺ T cells in the spleen at days 7 and 30 p.i. (n = 4). f Absolute numbers of GP33⁺CD8⁺ memory-precursor effector cells (MPECs; KLRG1⁻CD127⁺), short-lived effector cells (SLECs; KLRG1⁺CD127⁻), effector memory (TEM; CD44⁺CD62L⁻), central memory (TCM; CD44⁺CD62L⁺), and terminal effector (TE) cells in the spleen at day 30 p.i. (n = 6–8; data pooled from two independent experiments). GSK-3 KD mice exhibited significantly expanded MPEC, TEM, and TCM populations relative to WT controls. g Frequencies of MPECs, SLECs, TEM, and TCM as a percentage of GP33⁺CD8⁺ splenic T cells at day 30 p.i. (n = 6–16; data pooled from three independent experiments). h Percentage of GP33⁺CD8⁺ T cells expressing EOMES, FOXO1, CD127 (IL-7Rα), or KLRG1 at day 30 p.i. (n = 5–8). GSK-3 KD CD8⁺ T cells exhibited elevated FOXO1 and CD127 expression and reduced KLRG1 relative to WT counterparts. i Representative flow cytometry dot plots showing MPEC (upper panel) and SLEC (lower panel) cells among GP33⁺CD8⁺ T cells at day 30 p.i. (representative of n = 7–12; data pooled from two independent experiments). j Ki-67 expression in GP33⁺CD8⁺ SLEC and MPEC populations. Left panel: quantification of Ki-67⁺ cells in each subset (n = 6–7). Right panel: representative flow cytometry histograms for WT (upper) and GSK-3 KD (lower) cells. Elevated Ki-67 was observed selectively within the GSK-3 KD MPEC population. K TCF-1 expression (%) in GP33⁺CD8⁺ SLEC and MPEC subsets. Left panel: percentage of TCF-1⁺ cells among MPECs in WT and GSK-3 KD mice (n = 7–12; data pooled from two independent experiments). Right panel: representative flow cytometry profiles. l Percentage of TCF-1⁺ cells within CD44⁻, CD44⁺, GP33⁺TCM, and GP33⁺TEMCD8⁺ T-cell subsets at day 30 p.i. (n = 5–7; representative of two independent experiments). m Expression of exhaustion markers PD-1, TIM-3, and CD101 on GP33⁺CD8⁺ T cells at day 30 p.i. (n = 12–14; data pooled from two independent experiments). TIM-3 expression was significantly reduced on GSK-3 KD GP33⁺CD8⁺ T cells. n Absolute numbers of progenitor stem-like (TIM-3⁻TCF-1⁺), transitionally exhausted (CX3CR1⁺TIM-3⁺CD101⁻), and terminally exhausted (CX3CR1⁻TIM-3⁺CD101⁺) cells within PD-1⁺GP33⁺CD8⁺ T cells (n = 12–14; pooled from ≥2 independent experiments). o Corresponding frequencies of the three exhaustion subsets as a percentage of PD-1⁺GP33⁺CD8⁺ T cells (n = 12–14; pooled from ≥2 independent experiments). p Percentage of PD-1⁺TIM-3⁺TOX⁺ GP33⁺CD8⁺ T cells in spleens of WT mice treated with the GSK-3 inhibitor SB415286. q Ki-67 expression across progenitor stem-like, transitional, and terminally exhausted GP33⁺CD8⁺ T-cell subsets in WT and GSK-3 KD mice. Elevated Ki-67 was observed selectively in the TCF-1⁺ progenitor subset of GSK-3 KD mice. r–s IFN-γ production in GP33⁺CD8⁺ T cells at day 30 p.i., expressed as frequency (°r) and mean fluorescence intensity (MFI; s) (n = 5). t Frequency of dual IFN-γ⁺TNF-α⁺ GP33⁺CD8⁺ T cells at day 30 p.i. (n = 5). u–v Granzyme B (GZMB) expression in GP33⁺CD8⁺ T cells, shown as percentage (°u) and MFI (v) (n = 5). w Percentage of GP33⁺CD8⁺ T cells in the spleen at day 7 p.i. (n = 4). x MPEC and SLEC frequencies among GP33⁺CD8⁺ T cells at day 7 p.i. (n = 5). y Te^m and Tc^m frequencies among GP33⁺CD8⁺ T cells at day 7 p.i. (n = 4). z TCF-1 expression (%) in GP33⁺CD8⁺ T cells at day 7 p.i. (n = 4). No significant difference in TCF-1 expression was observed at this early timepoint. Data are presented as mean ± s.e.m. Statistical comparisons were performed by unpaired two-tailed Student’s t-test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; NS not significant

Initially, we examined the differentiation status of peripheral CD4⁺ and CD8⁺ T cells in naïve, unchallenged mice (Supplementary Fig. 1). T-cells from mesenteric lymph nodes (LNs) and spleen of mice born and raised in sterile cages were gated on CD4+ and CD8+ markers (Supplementary Fig. 1a, b, respectively). GSK-3 KD mice displayed a marked reduction in naïve T cells (CD44⁻CD62L⁺), particularly within the CD8⁺ compartment. In contrast, both CD4⁺ and CD8⁺ T cells from GSK-3 KD mice exhibited increased frequencies of central memory (T_CM; CD44⁺CD62L⁺) and effector memory (T_EM; CD44⁺CD62L⁻/lo) subsets. This shift was compartment-specific, with lymph nodes demonstrating a more pronounced expansion of CD8⁺ T_EM cells, whereas spleens showed a greater increase in T_CM cells. Together, these findings identify GSK-3 as a key homeostatic regulator of T-cell fate decisions, constraining memory differentiation and maintaining the naïve pool.

Given this, we next infected mice with Lymphocytic choriomeningitis virus Clone 13 (LCMV Cl13), a prototypic model for chronic viral infection (Fig. 1) (see Materials and Methods), as described.^42,65 Both cohorts showed an initial reduction in body weight; however, GSK-3 KD mice recovered more rapidly, with significantly greater body weight by day 10–30 (Fig. 1b). Correspondingly, LCMV Cl-13 viral titres were lower in GSK-3 KD mice compared to WT mice (Fig. 1c). In contrast, no differences in weight were observed with acute LCMV Armstrong infection at either day 7 (Supplementary Fig. 2a), or day 30 (Supplementary Fig. 2b).

Temporal analysis revealed a shift in cellular dynamics where although splenic GSK-3 KD T cell levels started at a lower level, they surpassed WT mice numbers during late-stage infection (Fig. 1d). Specifically, GSK-3 KD CD8⁺ T-cell numbers were lower when measured one day before infection (i.e., day −1) and at day 7 of infection, a pronounced expansion was evident during the chronic phase of infection at day 30, with a two-fold increase compared with WT controls. Consistent with this, GP33-specific CD8⁺ T cells, identified using an H-2Dᵇ–KAVYNFATM–BV421 tetramer, were also significantly more abundant in GSK-3 KD mice at day 30 (Fig. 1e). Together, these findings indicate that reduced GSK-3 expression promotes enhanced expansion and/or persistence of virus-specific CD8⁺ T cells under conditions that typically favor T-cell exhaustion.

To investigate the cellular mechanisms underlying this effect, we next examined the differentiation status of the CD8⁺ T-cell compartment. GSK-3 KD mice exhibited increased numbers of GP33 tetramer–positive CD8⁺ memory-precursor effector cells (MPECs; CD8⁺KLRG1⁻CD127⁺) (Fig. 1f). The number of MPECs increased from a mean of 0.5 × 10⁶ in WT mice to 1.4 × 10⁶ in GSK-3 KD mice. By contrast, the number of short-lived effector cells (SLECs; CD8⁺KLRG1⁺CD127⁻) was reduced. The number of GP33⁺CD8⁺ terminal effector T cells was also reduced in GSK-3 KD mice as defined by CD8 + KLRG1⁺CD39⁺CD127⁻ CD44⁺CD62L⁻TCF-1⁻GZMB⁺ (Fig. 1f) or by CD8⁺ KLRG1⁺ CD127⁻ CD44⁺ CD62L⁻ TCF-1-GZMB⁺ (Supplementary Fig. 2c). KLRG1⁺ is a marker for effector CD8⁺ T cells and is lost in memory T-cells.^66,67

An increase in the GSK-3 KD MPEC population and reduction in SLECs was also evident when expressed as a percentage of GP33-specific CD8⁺ T cells (Fig. 1g). These data showed that in the context of LCMV Cl13 infection, GSK-3 KD skews CD8⁺ responses toward a memory-precursor and long-lived memory phenotype at the expense of terminal effector differentiation.

In parallel, we observed elevated expression of FOXO1⁺ and CD127⁺ (i.e., IL-7R) CD8⁺ cells, accompanied by a marked reduction in KLRG1-expressing CD8⁺ T cells (Fig. 1h). Eomesodermin (EOMES) expression remained unchanged. CD127 is a canonical marker of memory-precursor and memory CD8⁺ T cells, while the transcription factor FOXO1 is essential for memory establishment, sustaining TCF-1 and IL-7Rα expression, while repressing terminal effector programs such as T-bet.⁶⁸ Although EOMES contributes to both effector and memory traits, it typically remains stable during late memory differentiation. A representative FACS profile demonstrated increased CD127 and reduced KLRG1 expression, confirming the expansion of MPECs in GSK-3 KD mice (Fig. 1I).

Consistent with this, Ki-67 expression, a marker for cell division,⁶⁹ increased selectively in the GSK-3 KD CD8+ GP33+ MPEC population (Fig. 1j, left panel). While 9.7% of GP33-CD8⁺WT cells expressed Ki-67, 21% of GP33-CD8⁺ GSK-3 KD cells expressed the marker. By contrast, the proliferation in the SLEC population was unaffected by the reduced GSK-3 expression. This differential effect on the MPEC cells was also shown in a representative FACs plot (right panel). These results showed increased cell division was induced by the reduction in GSK-3 expression selectively on MPEC GP33-CD8 + T-cells. Taken together, these findings show that GSK-3 downregulation selectively promotes the differentiation of memory-precursor CD8⁺ T cells early during infection, biasing the response toward a durable memory fate. Under physiological conditions, GSK-3 expression would therefore normally act as a negative regulator of CD8⁺ T-cell persistence, instead, promoting terminal effector differentiation while restraining the formation of long-lived memory precursors.

As a further control, acute infection with the LCMV Armstrong strain also increased T_EM cells in GSK-3 KD mice relative to wild-type (WT) mice at day 7 (Supplementary Fig. 2a, right panel). By day 30, consistent with a contraction/memory phase, the same Armstrong strain infection showed a relative shift toward an increase in T_CM GP33-CD8 T-cells compared to T_EM cells (Supplementary Fig. 2b right panel). There were also fewer GSK-3 KD TE cells upon LCMV Cl-13 infection when compared to WT mice (Supplementary Fig. 2c). In addition, we observed an increase in the percentage of CD44⁺CD4⁺ T cells at day 30 post-LCMV cl-13 infection and in the absolute numbers of CD4+ and CD44⁺CD4⁺ T cells (Supplementary Fig. 2d). By contrast, while the percentage of Th1, Tfh, and Treg subsets were similar in the spleen of GSK-3 KD and WT mice, a clear increase in the number of these cells was seen in the spleen of GSK-3 KD mice (Supplementary Fig. 2e). Similarly, the percentage and numbers of B-cells and CD44 + B-cells were indirectly enhanced, as were the frequency and absolute number of germinal-center (CD19⁺CXCR5⁺BCL6⁺) B cells (Supplementary Fig. 2e). These findings indicate that GSK-3 downregulation augments both CD4⁺ T-cell activation and germinal-center B-cell responses without altering the proportional balance among helper T-cell subsets.

GSK-3 KD promotes TCF-1 expression in CD8 + T-cells

Importantly, and central to this paper, within these subsets, GSK-3 KD mice also consistently showed an increased expression of the progenitor stem-like transcription factor TCF-1 in multiple CD8⁺ subsets. The increase in TCF-1 expression in GSK-3 KD was seen in both the SLEC and MPEC cells, but was at its highest level in the MPEC population (i.e., 28% in WT to 56% in GSK-3 KD T cell) (Fig. 1k, left panel). A representative FACs plot of GP33 + CD8 + T-cells showed the same pattern (i.e., 33.6% in WT to 55.8% for GSK-3 KD) (right panel).

Further, there was an increase in the percentage of TCF-1⁺ cells in the CD8⁺CD44+ population and GP33⁺ CD8⁺ CM and EM subsets in GSK-3 KD mice (Fig. 1l). These findings showed that GSK-3 KD promotes the expanded presence of the progenitor marker in a range of CD8⁺ subsets at day 30 of chronic LCMV Cl-13 infection.

GSK-3 KD CD8 + T-cells show reduced exhaustion

We next examined whether reduced GSK-3 expression affected T-cell exhaustion. While the MFI of PD-1 expression was similar on T cells of both genotypes, TIM-3 expression was significantly reduced on GSK-3 KD GP33⁺ CD8⁺ T cells relative to the WT equivalents (Fig. 1m). In this context, PD1 and TIM-3 expression can mark exhausted CD8 + T-cells with reduced effector functions.^70,71 In keeping with this, we also observed an increase in the number of progenitor stem-like TCF-1⁺ GP33⁺ CD8⁺ T cells (i.e., lacking TIM-3) (Fig. 1n). The number of transitory exhausted (PD-1⁺TIM-3⁺CX3CR1⁺) was reduced while terminally exhausted (PD-1⁺TIM-3⁺CX3CR1-CD101⁺) cells were expressed at low levels. Similarly, we observed an increase in percentage of stem-like TCF-1⁺ progenitors accompanied by a decrease in transitory and terminally exhausted cells (Fig. 1o). On average, the presence of GP33⁺ CD8⁺ PD1⁺ TIM1- cells increased from 7 percent in the WT spleen to 16 percent in the GSK-3 KD mice. Using the GSK-3 inhibitor SB415286, we also noted a decrease in PD-1⁺TIM-3⁺TOX⁺ exhausted cells in the GP33⁺ CD8⁺ population from infected mice (Fig. 1p). Importantly, we also saw an increase in Ki-67 expression preferentially in the GP33⁺ CD4⁺ TCF-1+ progenitor stem-like subset (Fig. 1q). Representative FACS profiles of GSK-3 KD versus WT GP33⁺CD8⁺PD-1⁺ splenic T cells also showed an increase in the frequency of TCF-1⁺TIM-3⁻ T cells (i.e., from 12.7 to 39.3%) as well as a decrease in the presence of transitional CX3CR1⁺ and terminally exhausted CD101⁺ T cells (Supplementary Fig. 3).

Consistent with reduced exhaustion, the GSK-3 KD GP33⁺CD8⁺ T-cells involving a major loss of GSK-3β and a partial reduction in GSK-3α expression showed a an increase in the frequency (Fig. 1r), and the MFI of IFN-γ production (Fig. 1s). The frequency of IFN-γ⁺ GP33⁺ CD8 + T-cells at day 30 of infection increased from 24 to 82% indicating that the reduction in GSK-3 reversed T-cell exhaustion. There was also a higher percentage of dual IFN-γ⁺ x TNF-α⁺ cells (Fig. 1t). Granzyme B expression was likewise elevated in both the percentage and MFI among GP33⁺CD8⁺ T cells (Fig. 1u, v), indicating reinforced cytotoxic potential. Together, these data demonstrate that partially-reduced GSK-3 expression preserves effector competence and cytokine polyfunctionality under chronic antigen exposure. Moreover, using a moderate genetic knock-down model, we show that GSK-3 constrains the generation of TCF-1⁺ stem-like progenitors, thereby limiting exhaustion and sustaining functional GP33-specific CD8⁺ T-cell responses at day 30.

As a control, we examined GSK-3 KD T cells at an earlier timepoint (day 7 post-infection). At this stage, GSK-3 KD mice displayed a reduced frequency of T cells (Fig. 1w), accompanied by an increase in MPECs and a reduction in SLECs (Fig. 1x). While there was a decrease in T_EM cells at day 7 (Fig. 1y), no increase in TCF-1 expression was yet observed (Fig. 1z). This early skewing toward an MPEC fate—coupled with the later expansion of TCF1+ progenitors by day 30—suggests that reduced GSK-3 expression initiates an intrinsic differentiation trajectory that favors progenitor maintenance and limits terminal exhaustion, ultimately supporting durable antiviral immunity.

GSK-3 acts as a metabolic gatekeeper of T cells

The increased frequency of TCF-1⁺ T cells in GSK-3 KD mice suggested that GSK-3 may regulate the activation threshold of T cells in response to TCR engagement. To assess proliferative responses, splenic T cells were labeled with Tag-it Violet™ (a proliferation-tracking dye) and stimulated in vitro with graded concentrations of soluble anti-CD3 (0.005–2 µg ml⁻¹) for 72 h (Fig. 2a). Flow-cytometric analysis was then performed on gated CD4⁺ and CD8⁺ T-cell populations. Both WT and GSK-3 KD T cells exhibited an array of distinct division peaks corresponding to successive rounds of proliferation across the 0.01–2.0 µg ml⁻¹ anti-CD3 range. However, at the lowest concentration (0.005 µg/ml), GSK-3 KD T cells showed three distinct division peaks relative to unstimulated controls (n = 8, arrows). This contrasted with WT CD8⁺ T cells which displayed a single division peak with only a faint indication of a second. A similar effect was observed in CD4 and CD8 + T-cells where GSK-3 KD cells showed a greater number of cell divisions at the lower and higher anti-CD3 concentrations (Supplementary Fig. 4). In some experiments, GSK-3 KD T-cells were seen to proliferate at concentrations as low as a remarkable 0.0025 ug/ml. These observations indicate that GSK-3 downregulation confers heightened proliferative competence, allowing T cells to undergo successive divisions even under sub-optimal TCR stimulation.

The alternative text for this image may have been generated using AI.

GSK-3 acts as a metabolic checkpoint governing bioenergetic reprogramming and mitochondrial fitness in CD8⁺ T cells. a Representative flow cytometry histogram of CellTrace™ Violet (Tag-it Violet™)-labelled splenic CD8⁺ T cells from WT (CTRL) and GSK-3 KD mice stimulated with graded concentrations of soluble anti-CD3ε (0.005–2 µg ml⁻¹) for 72 h (n = 4). Progressive dilution of dye fluorescence reflects successive cell divisions (arrows). GSK-3 KD CD8⁺ T cells underwent multiple rounds of division at the lowest antibody concentration (0.005 µg ml⁻¹), at which WT cells showed minimal proliferation. b Quantification of Ki-67⁺ cells among CD8⁺ T cells stimulated as in a. GSK-3 KD T cells exhibited significantly higher proliferation at the lowest anti-CD3ε concentration tested (n = 6; data pooled from two independent experiments). Two-way ANOVA with Šidák correction; *P < 0.05, ****P < 0.0001. c Representative extracellular acidification rate (ECAR) traces from splenic T cells activated with anti-CD3ε (1 µg ml⁻¹) for 24 h, with sequential injections of D-glucose, oligomycin, and 2-deoxyglucose (2-DG). Both basal and stimulated glycolytic activity were markedly elevated in GSK-3 KD T cells. d–f Quantification of glycolytic parameters: glycolysis (d), glycolytic capacity (e), and glycolytic reserve (f), derived from ECAR traces (n = 3). g Representative oxygen consumption rate (OCR) traces from T cells activated with anti-CD3ε (1 µg ml⁻¹) for 24 h, with sequential injections of oligomycin, FCCP, and rotenone/antimycin A (Rot/AA). Basal and stimulated OCR values were elevated in GSK-3 KD T cells. h–k Quantification of respiratory parameters: OCR in response to FCCP (h), maximal respiration (i), spare respiratory capacity (j), and ATP production rate (k) (n = 3). l Ratio of maximal OCR to maximal ECAR in activated T cells (n = 3). m–n Glucose uptake assessed by 2-NBDG incorporation (m) and GLUT1 expression by flow cytometry (n) in CD8⁺ and CD4⁺ T cells, either unstimulated or activated with anti-CD3ε (1 µg ml⁻¹) for 24 h (n = 5–6). Increased glucose uptake and GLUT1 expression were selectively observed in CD8⁺, but not CD4⁺, T cells from GSK-3 KD mice. o–p GLUT1, CD44, and GZMB MFI in GP33⁺CD8⁺ MPECs (o) and SLECs (p) at day 30 p.i. with LCMV Cl13. GLUT1 and GZMB were selectively elevated in GSK-3 KD MPECs. q GLUT1 expression (MFI) on progenitor stem-like, transitionally exhausted, and terminally exhausted GP33⁺CD8⁺ T-cell subsets at day 30 p.i. (n = 5–7). GLUT1 was predominantly expressed on TCF-1⁺ progenitor cells and was selectively increased in the GSK-3 KD progenitor subset. r–s SCENITH analysis of GP33⁺CD8⁺CD44⁺ T cells at day 30 p.i. with LCMV Cl13. r shows puromycin incorporation following 2-DG blockade of glycolysis (reflecting OXPHOS compensatory capacity); s shows puromycin incorporation following oligomycin blockade of mitochondrial ATP synthase (reflecting glycolytic compensatory capacity) (n = 5). t SCENITH-based quantification of basal protein synthesis (MFI) in PD-1⁺TCF-1⁺ progenitor-exhausted (TPEX) versus PD-1⁺TCF-1⁻ effector/terminally exhausted (TEFF/TEEX) GP33⁺CD8⁺ T-cell subsets. u–v SCENITH analysis of OXPHOS activity (u) and OXPHOS dependency (v) in TPEX versus TEFF/TEEX subsets at day 30 p.i. w–x SCENITH analysis of glycolytic capacity (w) and glycolytic dependency (x) in TPEX versus TEFF/TEEX subsets at day 30 p.i. GSK-3 KD TPEX cells exhibited reduced glycolytic dependence and enhanced OXPHOS engagement, consistent with a metabolically flexible progenitor phenotype. y Mitochondrial content, assessed by MitoSpy™ Green FM staining (MFI), in GP33⁺CD44⁺CD8⁺ T cells from WT and GSK-3 KD mice at day 30 p.i. (n = 5). z Mitochondrial mass (MitoSpy MFI) stratified by differentiation state: TCF-1⁺ progenitor, TIM-3⁺CX3CR1⁺CD101⁻ transitionally exhausted, and TIM-3⁺CX3CR1⁻CD101⁺ terminally exhausted GP33⁺CD8⁺ T-cell subsets (n = 5). Increased mitochondrial content was observed across all differentiation states in GSK-3 KD mice. Data are presented as mean ± s.e.m. unless otherwise stated. Statistical comparisons were performed by unpaired two-tailed Student’s t-test unless otherwise stated. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; NS not significant

A similar trend was observed when monitoring the marker of cellular proliferation, Ki-67 (Fig. 2b). Following low-dose anti-CD3 stimulation (0.005 µg/ml), approximately 70% of GSK-3 KD T-cells expressed Ki-67 compared to 30% of WT cells. A more modest but still significant increase was also detected at higher anti-CD3 concentrations, indicating that reduced GSK-3 expression enhances proliferation across a range of activation thresholds. A comparable effect on both CD4⁺ and CD8⁺ T-cells was confirmed in independent experiments as anti-CD3 concentrations as low as 0.0025ug/ml (Supplementary Figs. 5a, b), whereas no consistent difference was observed in the frequency of apoptotic cells or in Bcl-2 expression (Supplementary Fig. 6a–bn). GSK-3 KD CD4⁺ and CD8⁺ T-cells also exhibited increased β-CATENIN expression (Supplementary Figs. 6d) with only marginal effects on AKT phosphorylation (Supplementary Fig. 6c).

In this context, metabolism plays key role in T-cell activation.^72,73,74,75 Upon activation, T-cells undergo a metabolic switch from oxidative phosphorylation (OXPHOS) to glycolysis.^72,73,74,75 The role of GSK-3 in regulating T-cell metabolism has not been fully explored. To assess this, we first measured the oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) in resting and anti-CD3–stimulated T cells after 24 h using a Seahorse XF24 Extracellular Flux Analyzer (Fig. 2c–l). Cells were sequentially treated with mitochondrial perturbing agents, including oligomycin (to inhibit ATP synthase), FCCP (to uncouple the electron transport chain), and rotenone plus antimycin A (to inhibit complexes I and III, respectively).

ECAR, a proxy for glycolytic activity, was measured in a representative experiment (Fig. 2c). Anti-CD3 activated GSK-3 KD T cells exhibited significantly higher basal ECAR levels compared to WT controls, a disparity that was further amplified following the addition of glucose. To assess maximal glycolytic capacity, we administered oligomycin to inhibit mitochondrial ATP synthase. This shift toward compensatory glycolysis further widened the gap, highlighting a robust glycolytic program in GSK-3 KD T cells.

Pooled data from multiple experiments confirmed the enhancement of glycolysis in the GSK-3 KD T-cells (Fig. 2d). Moreover, the same cells exhibited higher glycolytic capacity (Fig. 2e) and an elevated glycolytic reserve, defined as the difference between maximal ECAR after glucose + oligomycin and after glucose alone (Fig. 2f). Collectively, these data suggest that the loss of GSK-3 primes T cells for an enhanced metabolic state with enhanced aerobic glycolysis.

Despite this, GSK-3 levels also influenced mitochondrial respiration, as reflected by changes in OCR profiles (Fig. 2g–k). Both basal and anti-CD3–induced OCR levels were higher in GSK-3 KD T cells compared to wild-type controls (Fig. 2g), as was the response to FCCP (Fig. 2h). FCCP reveals the maximal uncoupled respiratory capacity—the highest achievable oxygen consumption rate when mitochondrial activity is disengaged from ATP production—providing an insight into a T-cell’s ability to meet elevated energetic demands. The maximal OCR did not differ significantly between groups (Fig. 2i). Notably, the spare respiratory capacity, defined as the difference between maximal and basal OCR, was reduced in anti-CD3–stimulated GSK-3 KD T cells (Fig. 2j). This is in line with the higher basal OCR levels. Further, consistent with these observations, GSK-3 KD T cells exhibited increased ATP production Fig. 2k), but a decreased OCR/ECAR ratio (Fig. 2I). Together, these findings indicate that loss of GSK-3 enhances mitochondrial output while simultaneously biasing T-cell metabolic programming toward glycolysis.

In support of this glycolytic shift, glucose uptake was elevated in CD8⁺, but not CD4⁺ T cells from GSK-3 KD mice (Fig. 2m, left and right panels). This was accompanied by increased expression of the glucose transporter GLUT-1 in CD8 + T-cells (Fig. 2n, left panel).

While GSK-3 KD CD4 T cells exhibited a similar upward trend in GLUT-1 expression, it did not reach the same magnitude as their CD8 counterparts (Fig. 2n right panel). GLUT-1 plays a crucial role in mediating glucose uptake in activated T-cells.^76,77,78

In the context of LCMV Cl-13 infection, it was difficult to extract sufficient numbers of cells to conduct Seahorse analysis. Nevertheless, we found that GSK-3 KD GP33⁺ CD44⁺ CD8⁺ MPECs express a higher MFI of expression of GLUT1 and GZMB than WT cells, an effect not evident on SLECs (Fig. 2o, p, respectively). Intriguingly, GLUT1 expression was selectively upregulated on GSK-3 KD stem-like progenitor exhausted (TPEX) cells (Fig. 2q) but remained negligible on transitory and terminally exhausted subsets. This profile aligns with the increased frequency of Ki67 TCF1⁺ progenitor cells observed 30 days post-infection (Fig. 1o, q). Overall, our findings indicated that GSK-3 affected both the glycolysis and OXPHOS pathways of cellular metabolism, and in particular, by augmenting glycolysis in progenitor CD8⁺  TCF-1⁺ T-cells.

To complement these findings, we also used SCENITH analysis to profile GP33 + CD8 + T cells at day 30 of infection (Fig. 2r–x). SCENITH is a flow-cytometry–based assay that quantifies cellular dependence on glycolysis versus mitochondrial respiration by measuring changes in protein synthesis following selective metabolic inhibition.⁷⁹ Upon glycolysis blockade with 2-DG, GSK-3 KD CD8⁺CD44⁺ T cells maintained significantly higher puromycin incorporation, demonstrating enhanced mitochondrial compensation and greater metabolic flexibility to sustain OXPHOS under conditions of limited glucose flux (Fig. 2r). Conversely, when OXPHOS was inhibited with oligomycin, GSK-3 KD CD8⁺ T cells exhibited a stronger glycolytic compensatory response compared with WT cells (Fig. 2s). This reflects an increased capacity to support ATP production via glycolysis when mitochondrial ATP generation is impaired. Collectively, these in vivo data confirm that the GSK-3 deficiency promotes enhanced engagement of both glycolytic and mitochondrial metabolic programs in CD8⁺ T cells during chronic LCMV Cl13 infection, conferring metabolic flexibility that preserves protein synthesis under metabolic stress.

We also compared these effects with a focus on TCF-1⁺ stem-like T cells. PD-1⁺TCF-1⁺ progenitor (TPEX) cells were compared to the remaining PD-1⁺TCF-1⁻ CD8⁺ T-cell compartment comprising effector (TEFF) and terminally exhausted (TEEX) cells (Fig. 2t–x). GSK-3 KD CD8⁺ T cells exhibited higher basal translation in both the TCF1⁺ TPEX and TCF1^–TEFF/TEEX subsets compared with WT cells (Fig. 2t). In addition, GSK-3 KD cells maintained elevated OXPHOS during 2-DG-mediated glycolytic inhibition, outperforming WT and effector populations at 30 days post-infection (Fig. 2u). An increase in the percentage dependency on OXPHOS was observed in the GSK-3 KD cells effector and terminally exhausted TEFF/TEEX cells (Fig. 2v). Interestingly, despite their lower basal biosynthetic activity, the TEFF/TEEX CD8⁺ T cells exhibited greater OXPHOS dependence than progenitor-like cells, consistent with a metabolically rigid phenotype.

Regarding glycolysis, both GSK-3 KD TPEX and TEFF/TEEX cells exhibited reduced glycolytic capacity compared to WT T cells when OXPHOS was inhibited by oligomycin (Fig. 2w), reflecting a greater reliance on OXPHOS-derived ATP in GSK-3 KD mice. GSK-3 KD TPEX cells also showed slightly lower glycolytic dependence, whereas TEEX cells exhibited a small increase (Fig. 2x). Overall, GSK-3 KD reduces dependence on glycolysis—particularly in TPEX cells—and endows them with balanced metabolic plasticity, allowing efficient engagement of either glycolysis or OXPHOS depending on environmental stress. In contrast, TEFF/TEEX CD8 T-cells from 30-day-old infected mice remain more metabolically constrained, characterized by greater OXPHOS reliance and limited flexibility.

GSK-3 deficiency drives expansion of mitochondrial biomass

Because the concurrent enhancement of glycolysis and OXPHOS suggested a fundamental shift in cellular bioenergetic capacity, we next quantified mitochondrial content in GSK-3–deficient T cells. Using a MitoSpy fluorescent probe, we observed a significant increase in mitochondrial mass within GP33⁺CD44⁺ CD8⁺ T cells from GSK-3 KD mice (Fig. 2y). This increase was evident across multiple differentiation states, including a marked enrichment in TCF-1⁺ progenitor-like CD8⁺ T cells, as well as in TIM-3⁺CX3CR1⁺ transitionally exhausted and surprisingly, in TIM-3⁺CD101⁺ terminally exhausted CD8⁺ T-cells (Fig. 2z). Together, these data indicate that the GSK-3 knockdown promotes increased mitochondrial mass across the CD8⁺ T-cell differentiation stages supporting enhanced energy-generating capacity in progenitor, effector, and exhausted populations. This suggests that GSK-3 acts as a central metabolic brake, and its removal enables a robust mitochondrial program regardless of the T cell’s differentiation state.

GSK-3 KD empowers PD-1 blockade to overcome ICB tumor resistance

Given these effects, we next assessed the role of GSK-3 in regulating T-cell responses against tumors (Fig. 3). For this, we derived and used a B16-F10 melanoma subclone (B16-F10 R1) which is resistant to anti-PD-1 treatment. B16-F10 R1 cells were injected intradermally followed by intra-peritoneal injections of the GSK-3 competitive inhibitor, SB415286 (GSK-3i),^42,57 anti-PD-1 or a SB415286/anti-PD-1 combination, every 4 days (Fig. 3a). The concentrations and properties of the inhibitor and antibody have been previously described^42,58 Neither anti-PD-1 nor SB415286 monotherapy affected tumor growth of this non-immunogenic tumor (Fig. 3b). However, by contrast, the combination of anti-PD-1 and SB415286 synergized to reduce tumor growth from days 8-17 based on tumor volume (Fig. 3b) and tumor weight when measured at day 15 (Fig. 3c). Spider-graphs show reduced volume growth of individual tumors in mice (Fig. 3d, left panels). In addition, 65% of mice bearing B16-F10 R1 tumors became responsive to combination therapy as opposed to none with monotherapy (right panel). Kaplan-Meier survival analysis demonstrated a survival advantage with combination therapy (Supplementary Fig. 7). Further, the anti-PD-1/SB415286 combination also significantly enhanced tumor control compared to monotherapy in the colorectal MC38 model (Supplementary Fig. 8).

The alternative text for this image may have been generated using AI.

GSK-3 inhibition synergizes with PD-1 blockade to overcome tumor resistance and remodel the tumor immune microenvironment. a Schematic of the experimental design. C57BL/6 WT mice were implanted intradermally with B16-F10 R1 melanoma cells (a PD-1-resistant subclone). Mice received intraperitoneal injections of vehicle, SB415286 (GSK-3i), anti-PD-1 antibody, or the combination every 4 days beginning on day 1 post-implantation. b,c Tumor volume over time (b) and tumor weight at day 15 (c) for each treatment group (n = 5–10; data pooled from ≥2 independent experiments). Combination therapy significantly reduced tumor growth compared to either monotherapy alone. d Left panels: spider plots showing the growth trajectory of individual tumors per group. Right panel: pie charts depicting the proportion of responders (defined as ≥50% reduction in tumor volume) in the anti-PD-1 and combination treatment groups (data pooled from four independent experiments). 65% of mice in the combination group responded to therapy. e Phenograph visualization of 20 distinct immune cell clusters identified by mass cytometry (CyTOF) in anti-TCRβ-gated tumor-infiltrating lymphocytes (TILs) (n = 5–10). f Heatmap displaying the relative expression of 15 surface markers across the 20 CyTOF-defined immune clusters. Cluster 20 (★), which expanded significantly with combination therapy, corresponded to a CD8⁺TEM phenotype (CD44⁺CD62L⁻IA-IE⁺PD-L1⁺PD-1⁻LAG-3⁻). Cluster 1 (★), which was diminished, corresponded to immunosuppressive Tregs (CD44⁺CD25⁺FOXP3⁺CD73⁺PD-1⁺LAG-3⁺). g Box plots showing the distribution of each immune cluster across treatment groups as a proportion of anti-TCRβ⁺ TILs. h Frequency of CD8⁺TEM TILs as a proportion of total TILs (left panel) and correlation with tumor mass (right panel) (n = 9). TEM frequency correlated inversely with tumor weight. i Frequency of CD8⁺ effector (TEFF) TILs (left) and their correlation with tumor mass (right) (n = 9). No significant association between Tᴇ^ff frequency and tumor regression was detected. j Overall CD8⁺ TIL abundance across treatment groups (n = 9). Both anti-PD-1 monotherapy and combination therapy increased total CD8⁺ TIL infiltration. k,l Frequency of total CD4⁺ TILs (k) and CD4⁺FOXP3^+hiCD25⁺ Tᴀᴇᶣ (l) across treatment groups (n = 9). Combination therapy was associated with a marked and significant reduction in intratumoral Tᴀᴇᶣ frequency. Data are presented as mean ± s.e.m. Statistical comparisons were performed by one-way or two-way ANOVA with post-hoc correction. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; NS not significant. n = 3 independent experiments with 4–5 mice per group

To define the nature of TILs, we used Time-of-Flight (CyTOF) mass cytometry with a panel of 15 monoclonal antibodies (Fig. 3e-m). From this, 20 distinct cellular subsets within the anti-TCRβ positive population became apparent as shown with phenograph illustrations (Fig. 3e), heat maps (Fig. 3f) and boxplot representations (Fig. 3g). Although there were various statistically non-significant upward and downward trends, only two clusters achieved statistical significance (<0.05) (Supplementary Fig. 9). Cluster 20 was expanded and corresponded to a CD8⁺ T-cell subset characterized by the phenotype CD44⁺ CD62L⁻IA-IE⁺CD11b⁻CD73⁻ PD-L1⁺, consistent with a CD8⁺T_EM population (P = 0.013). Notably, this subset lacked expression of PD-1 and LAG-3, indicating an absence of both transitory and terminal exhaustion phenotypes (Fig. 3f, see asteriks). Conversely, Cluster 1 in the CD4⁺ population was diminished and comprised cells expressing CD44⁺CD25⁺FOXP3⁺ CD73⁺PD-1⁺LAG-3⁺IA-IE⁺CD49f⁺PD-L1⁺, a phenotype characteristic of immunosuppressive regulatory T cells (Tregs) (P = 0.044). The reduction of this cluster suggests an attenuated suppressive compartment in GSK-3 KD mice.

Using a less stringent threshold, there was a trend toward reduced frequencies of Cluster 19 CD4⁺ T-cells (P = 0.09) (Fig. 3f, see asterisks and Supplementary Fig. 9). This cluster shared many phenotypic markers with the Treg-enriched Cluster 1 but lacked expression of PD-1 and LAG-3, suggesting a less suppressive phenotype. Similarly, the FoxP3⁺ Cluster 12 resembled Cluster 19 but lacked CD49f (integrin α6) and exhibited lower levels of the inhibitory ectoenzyme CD73 (P = 0.127). CD73 marks a subset of Tregs with enhanced immunosuppressive capacity.^80,81 Furthermore, CD8+ Cluster 4 showed an increase with a p value of 0.149 resembled Cluster 20, but withexhaustion markers PD-1, LAG3, and CD73. Overall, combination therapy was associated with an increase in CD8⁺ TILs and a decrease in Treg-like CD4⁺ suppressor cells.

We next employed conventional flow cytometry to validate the CyTOF-based findings (Fig. 3h–l). This complementary analysis confirmed a significant increase in CD8⁺ T_EM TILs with combination therapy (Fig. 3h, left panel). Thisfrequency of CD8⁺ T_EM cells exhibited a strong inverse correlation with tumor mass (right panel). A similar increase was observed for CD8⁺ effector (T_EFF) cells (Fig. 3i, left panel); however, itdid not reach statistical significance and was not associated with tumor regression (right panel). In line with previous reports,⁸² anti–PD-1 monotherapy increased CD8⁺ TIL abundance which was unaffected by the presence of SB415286 (Fig. 3j). Together, these data indicate that while PD-1 blockade broadly expands CD8⁺ T-cell infiltration, concomitant GSK-3 inhibition selectively promotes the relative presence of tumor-rejecting cytotoxic CD8⁺ T_EM cells.

Conventional flow cytometric analysis also showed a decrease in the presence of CD4⁺ T-cells (Fig. 3k). This effect was most pronounced on suppressor T-cells with a marked reduction in CD4⁺FOXP3^hiCD25⁺ Tregs in response to GSK-3i x anti–PD-1 combination therapy (Fig. 3l). FACs staining demonstrated that anti–PD-1 treatment decreased the frequencies of PD-1⁺FOXP3⁺ and TIM-3⁺FOXP3⁺ Tregs (Supplementary Fig. 10), with a modest but reproducible trend to further reduce with combination therapy. Together, these data indicate that GSK-3 inhibition cooperates with PD-1 blockade to constrain immunosuppressive Treg subsets within the tumor microenvironment, thereby favoring a more permissive milieu for effective antitumor immunity

GSK-3KD licenses PD-1 blockade to induce perforin and 7/9 granzymes for enhanced CD8 + CTL TILs

We next assessed the transcriptional programs induced by anti–PD-1 and GSK-3 inhibition, alone or in combination, using bulk RNA-seq analysis of tumor-infiltrating lymphocytes (Fig. 4a–c). Differentially expressed genes were initially defined by a ≥ 2-fold change with p < 0.01 relative to untreated controls. A Venn diagram (Fig. 4a) illustrates the distinct and overlapping gene sets modulated by each treatment. Anti–PD-1 monotherapy altered 162 genes, GSK-3i 109 genes, with only three genes shared between the two monotherapies. In contrast, combination therapy elicited a markedly broader transcriptional program, inducing 910 unique genes in tumors from highly responsive mice (defined as >50% tumor reduction relative to untreated controls) and 176 unique genes in non-responders. There was a clear synergy with GSK-3i licensing anti-PD-1 induced alterations in gene expression within T-cells. Network analysis revealed enrichment for genes associated with immune activation and inflammatory response pathways (Supplementary Fig. 11).

The alternative text for this image may have been generated using AI.

Transcriptomic profiling reveals a GSK-3–PD-1 synergistic program driving coordinated granzyme and perforin induction in tumor-infiltrating CD8⁺ T cells. a Venn diagram of differentially expressed genes (DEGs) in CD8⁺ TILs isolated from B16-F10 R1 tumors following treatment with anti-PD-1 monotherapy, GSK-3i monotherapy, or combination therapy, relative to untreated controls (≥2-fold change, P < 0.01). Anti-PD-1 and GSK-3i monotherapies regulated largely non-overlapping gene sets (162 and 109 DEGs, respectively, with only 3 shared), whereas combination therapy uniquely induced 910 genes in high responders (>50% tumor volume reduction) and 176 genes in non-responders. b Volcano plots of DEGs in each treatment group relative to untreated controls, with a stringent threshold (≥2-fold change, P < 0.001). GSK-3i monotherapy selectively induced Fasl. Anti-PD-1 monotherapy predominantly downregulated immune adhesion and chemotaxis genes (Vcam1, Cxcl9, Cxcl11). Combination therapy elicited a robust, synergistic transcriptional response dominated by upregulation of cytolytic effector genes. c Heatmap of the 37 genes significantly upregulated by combination therapy (stringent threshold). Key upregulated genes include the T-cell signaling kinase Itk, the transcription factor Tox2, perforin (Prf1), and seven of the nine known murine granzymes: Gzmb, Gzmc, Gzmd, Gzme, Gzmf, Gzmg, and Gzmk. d Heatmap focusing on cytolytic effector genes comparing expression levels across GSK-3i monotherapy, anti-PD-1 monotherapy, and combination therapy. The coordinated upregulation of perforin and multiple granzymes was exclusively observed with combination treatment. e Heatmap of genes upregulated (≥2-fold, P < 0.001) in TILs from combination therapy responders versus non-responders. Gzmb, Gzmk, and Prf1 were significantly elevated in responders, underscoring the importance of a broad cytolytic repertoire for effective tumor clearance. f Cytotoxicity score computed for each treatment group by averaging normalized expression values of a pre-defined cytolytic gene signature (Gzma, Gzmb, Prf1, Fasl, Ifng, Cd3e, Nkg7, Lgals1, Fgl2, Cxcl11, Nr4a3, Spp1, Icos, Klrg1, Cd8a, and Cd8b1). Combination therapy significantly increased the cytotoxicity score relative to anti-PD-1 monotherapy. Statistical significance assessed by Student’s t-test. g Quantification of TCF-1⁺ cells within the CD8⁺TCM TIL compartment by flow cytometry (n = 4). TCF-1⁺ frequency was elevated following combination therapy. h Tcf7 (TCF-1) mRNA expression (reads per million, RPM) in CD8⁺ TILs across treatment groups. Combination therapy was associated with increased Tcf7 transcript levels. Data are presented as mean ± s.e.m. Statistical comparisons were performed by unpaired two-tailed Student’s t-test unless otherwise stated. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; NS not significant. n = 3 independent experiments with 4–5 mice per group

To increase analytical stringency, a stricter threshold (≥2-fold change, p < 0.001) was applied for a more refined comparison. Volcano profiling showed that the combination of GSK-3i and anti-PD-1 blockade induced a distinct immunogenic signature (Fig. 4b). GSK-3i monotherapy induced the expression of Fasl, encoding the death ligand FasL (left panel). Anti-PD-1 monotherapy mostly downregulated genes included genes involved in immune adhesion and chemotaxis (Vcam1, Cxcl9, Cxcl11) while upregulating only Fasl and Gm8369 (a predicted non-coding transcript with no currently defined function) (middle panel). By contrast, GSK-3i in combination with anti-PD-1 elicited a robust synergistic transcriptional response, characterized by coordinated up- and down-regulation of gene programs (right panel). A control volcano plot comparing GSK-3i monotherapy to combination therapy is shown in Supplementary Fig. 12. Differential expression analysis revealed induction of cytolytic genes (e.g., Gzmf, Serpinb9b) in the combination treatment relative to GSK-3 alone. GSK-3 inhibition acted as a licensing signal that unlocked the capacity of anti-PD-1 to drive transcriptional changes in TILs.

Genes shown were then further filtered to exclude low-expression transcripts and retain those with consistent expression across samples. Among the 37 genes upregulated by combination therapy versus untreated controls under the more stringent threshold, several were directly linked to T cell activation and effector differentiation, as visualized by heatmap analysis (Fig. 4c). These included interleukin-2–inducible T cell kinase (ITK) and the transcription factor Tox2. ITK is a central mediator of T cell receptor signaling, regulating downstream activation pathways that control cytokine production and lineage commitment,⁸³ whereas Tox2 supports metabolic adaptation and tissue residency of effector T cells.⁸⁴

However, most strikingly, combination therapy resulted in robust induction of genes for cytolytic machinery, including perforin (Prf1) and a remarkable seven of the nine known murine granzymes (Gzmk, Gzmg, Gzme, Gzmf, Gzmd, Gzmb, and Gzmc) (Fig. 4c, d). The coordinated upregulation of this range of target-killing components has not previously been described in the context of pathways that synergize with anti-PD-1 to potentiate CTL effector function. Perforin is a pore-forming cytolytic protein that facilitates the delivery of granzymes into target cells, thereby enabling apoptotic cell death.^85,86 Granzymes Gzmb, Gzmg, Gzmd, Gzme, Gzmf and Gzmk have reported cytotoxic functions.^{87,88,89,90,91} Notably, Gzmk is associated with early-stage memory CD8⁺ T-cell differentiation.⁹² Together, these findings position GSK-3 as a unique upstream amplifier of cytolytic programs for enhanced anti–PD-1–driven CTL immunity within the tumor microenvironment.

To further distinguish the transcriptional signatures associated with treatment responsiveness, we compared tumors from highly and poorly responsive mice. It was designated as a responder is there was a > 50% tumor volume reduction and a non-responder with a < 50% tumor volume reduction (otherwise a poor responder). TILs from both responders and non-responders expressed Gzmg, Gzme, Gzmf, and Gzmd (Fig. 4e). However, although Gzmk, Gzmb, and Prf1 were expressed in non- or poorly responsive mice, the increase did not reach statistical significance. These results underscore the importance of a broad repertoire of cytolytic mediators for optimal tumor clearance. Among these, perforin, a pore-forming effector essential for CTL-mediated killing and is, in part, also regulated by PD-1 internalization,²⁵ while GZMB is well established as a key executor of CD8⁺ T-cell effector function in TILs.⁹³ Together, these data indicate that effective tumor regression requires the coordinated induction of the perforin–granzyme cytolytic program.

Gene set enrichment analysis (GSEA) also revealed a trend toward an enrichment of gene sets associated with T cell–mediated cytotoxicity and xenogeneic target cell killing (Supplementary Fig. 13). Consistent with this, analysis of an independently curated, unbiased cytolytic gene signature—including Gzma, Gzmb, Prf1, Fasl, Ifng, Cd3e, Nkg7, Lgals1, Fgl2, Cxcl11, Nr4a3, Spp1, Icos, Klrg1, Cd8a, and Cd8b1—demonstrated a significantly enhanced cytotoxicity score following combination therapy (Fig. 4f). Collectively, again, these findings indicate that GSK-3 inhibition cooperates with PD-1 blockade to reinforce transcriptional programs associated with cytolytic effector function.

Notably, transcriptional profiling also revealed an elevated frequency of TCF-1⁺ cells within the CD8⁺ T_CM compartment (Fig. 4g) as well as an increase in Tcf7 (TCF-1) expression following combination therapy (Fig. 4h). These findings suggest that GSK-3 inhibition, in concert with PD-1 blockade, both amplifies cytolytic effector programs and sustains the progenitor-like memory pool for a durable anti-tumor immunity.

GSEA also suggested an increase in glycolysis-associated genes with combination versus either monotherapy (Supplementary Fig. 14a, b). Consistently, an a priori pathway-defined gene set (GO/KEGG/Reactome) encompassing glycolysis and glucose import (Slc2a1, Hk2, Pfkl, Aldoa, Pkm, Ldha, Hif1a, Slc16a3) indicated a higher glycolysis score with combination therapy (Supplementary Fig. 14c, d). A pre-defined glucose import gene panel (Slc2a1, Slc2a3, Akt1, Akt2, Pik3cd, Pik3ca, Mtor, Prkaa1, Tbc1d4, Insr) showed an increased glucose import score in the combination group (Supplementary Fig. 14c, d).

GSK-3 KD licenses anti-PD-1 induction of CD8 multi-granzyme expression

These findings were also validated at the genetic level by analyzing the response of GSK-3 KD mice to anti–PD-1 therapy (Fig. 5a, b). Whereas anti–PD-1 monotherapy failed to significantly inhibit B16-F10 R1 tumor growth in WT mice, it produced a marked (> 75%) reduction in tumor burden in GSK-3 KD mice. This effect was evident by a measure of tumor volume (Fig. 5a) and tumor weight at day 17 post-implantation (Fig. 5b). These data confirmed that reduced GSK-3 expression synergizes with PD-1 blockade to suppress the growth of tumors otherwise resistant to immune-checkpoint monotherapy.

The alternative text for this image may have been generated using AI.

Genetic GSK-3 knockdown licenses PD-1 blockade to generate super-armed, multi-granzyme-expressing CD8⁺ cytolytic T cells in the tumor microenvironment. a,b Tumor volume over time (a) and tumor weight at day 16 post-implantation (b) in WT and GSK-3 KD mice bearing B16-F10 R1 tumors, treated with or without anti-PD-1 antibody. Anti-PD-1 monotherapy was ineffective in WT mice but produced a > 75% reduction in tumor burden in GSK-3 KD mice. c,d Percentage of CD8⁺ cells among TILs (c) and absolute CD8⁺ TIL counts per gram of tumor (d) across treatment groups (n = 3–5). e Representative flow cytometry dot plots of CD4⁺ and CD8⁺ TIL fractions across treatment groups, illustrating the marked increase in CD8⁺ TIL proportion in anti-PD-1-treated GSK-3 KD mice. f Frequency of TEM cells within the CD8⁺CD44⁺ TIL compartment (n = 4). TEM TILs were significantly expanded in anti-PD-1-treated GSK-3 KD mice. g TCF-1 expression (MFI) within CD8⁺CD44⁺ TILs (n = 3–4). GSK-3 KD sustained elevated TCF-1 MFI, which was further increased by anti-PD-1 treatment. h Frequency of CD8⁺CD44⁺ TILs co-expressing TCF-1 and TIM-3 (n = 3), reflecting the emergence of a hybrid progenitor-exhausted state in GSK-3 KD mice. i GZMB expression (MFI) within CD8⁺CD44⁺TCF-1⁺ TILs (n = 3–4). Combined anti-PD-1 and GSK-3 KD significantly enhanced GZMB expression in TCF-1⁺ TILs. j,k Frequencies of TCM (j) and TEM (k) cells within the CD8⁺TRP2⁺ TIL population, identified using H-2K^b TRP-2 BV421 tetramers (n = 3–4). GSK-3 KD mice showed a shift toward central memory within the TRP-2-reactive TIL compartment. l Proportion of TCF-1⁺ cells within the CD8⁺TRP2⁺TEM TIL subset (n = 3–4). m,n GZMB co-expression within CD8⁺TRP2⁺TCF-1⁺ TILs, shown as frequency (m) and MFI (n) (n = 3–4). o Intracellular expression of perforin, GZMA, GZMB, GZMC, and GZMF in CD44⁺CD8⁺ TILs from WT and GSK-3 KD mice treated with or without anti-PD-1 (n = 3–5), shown as MFI (upper panels) and frequency of positive cells (lower panels). Anti-PD-1 treatment in GSK-3 KD mice markedly increased both the proportion and per-cell intensity of all cytolytic molecules assessed. Data are presented as mean ± s.e.m. Statistical comparisons were performed by unpaired two-tailed Student’s t-test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; NS not significant. n = 3 independent experiments with 4–5 mice per group

Further, anti–PD-1 synergy with the GSK-3 KD increased the presence of CD8⁺ TILs as a percentage (Fig. 5c) and in the number of CD8⁺ TILs/gm of tumour (Fig. 5d). While untreated GSK-3 KD mice and anti-PD-1 treated WT mice showed an increase in CD8⁺ TIL frequency (Fig. 5c, rising from 24% to 42–44%), the combination of the two treatments yielded a synergistic effect to 60%. Likewise, the number of tumoral CD8 + T-cells increased from 2.4 × 10⁶ cells/gm in the WT control to 9.2 × 10⁶ cells/gm in GSK-3 KD mice. The increased presence of CD8 TILs was also clearly visible in a representative FACS plot (Fig. 5e).

Furthermore, combination therapy in mutant mice significantly enriched the T_EM TIL population (Fig. 5f). This expansion was accompanied by an increase in TCF-1 expression (MFI) in the CD8⁺ CD44⁺ TIL compartment of untreated GSK-3 KD mice- a trend further amplified by anti–PD-1 (Fig. 5g). Consistent with this, GSK-3 KD also increased the percentage of TCF-1⁺ TIM-3⁺ TILs within the CD8⁺ CD44⁺ TIL population (Fig. 5h). At the same time, CD8⁺CD44⁺TCF-1⁺ TILs showed an increase in GZMB expression from GSK-3 KD mice treated with anti–PD-1 (Fig. 5i). Collectively, these findings describe a hybrid phenotype in which GSK-3 deficiency bypasses the standard progression toward terminal exhaustion, instead preserving a TCF-1-high, effector-competent state in TIM+ exhausted cells, altering the differentiation landscape of the TILs.

For the identification of tumor-reactive TILs from B16-F10 tumors, we used H-2K^b TRP-2 BV421 tetramers (mouse MHC class I; TRP-2_180–188 (SVYDFFVWL)) to identify TRP-2–specific TILs. TRP-2 (tyrosinase-related protein 2) is a melanocyte differentiation antigen expressed by B16 melanoma cells. TILs isolated from B16-F10 tumors at day 18 were assessed for differentiation status and expression of TCF-1 and GZMB. GSK-3 KD mice showed an increase in the percentage of CD8⁺CD44⁺TRP2⁺ T_CM cells in untreated and anti–PD-1-treated mice (Fig. 5j). While CD8⁺CD44⁺TRP2⁺ TILs predominantly displayed a T_EM phenotype in both genotypes, GSK-3 KD mice showed a reduction in T_EM cells that was roughly commensurate with the increase in T_CM cells (Fig. 5k). Importantly, the proportion of the CD8⁺TRP2⁺ T_EM subset expressing TCF-1⁺ was elevated in GSK-3 KD mice (Fig. 5l). Anti–PD-1 treatment increased the frequency of TCF-1⁺ CD8⁺CD44⁺TRP2⁺ TILs in WT mice from 21 to 39%, approximating the levels observed in untreated and anti-PD-1 treated GSK-3 KD mice. Collectively, these findings indicate that GSK-3 KD enriches the TRP2⁺ CD8⁺ TIL compartment for central memory differentiation and promotes a TCF-1⁺ memory-like transcriptional program within the T_EM subset.

Moreover, within the TCF-1⁺ compartment, co-expression of GZMB was significantly enriched in CD8⁺CD44⁺TRP2⁺TCF1+TILs from anti-PD-1 treated GSK-3 KD mice relative to the untreated GSK-3 mice and anti-PD-1 treated WT mice (Fig. 5m). Further, this was accompanied by an increase in the GZMB MFI in anti-PD-1 treated GSK-3 KD mice (Fig. 5n). GSK-3 down-regulation favors the accumulation of TCF-1⁺, memory-biased CD8⁺ TILs, while PD-1 blockade enhances their cytolytic competence, giving rise to a distinct TCF-1⁺GZMB⁺ effector–memory hybrid state within the tumor.

Further underscoring this, flow cytometric analysis of intracellular granzymes revealed that anti–PD-1 treatment in GSK-3 KD mice markedly upregulated the expression of perforin and multiple granzymes—GZMA, GZMB, GZMC, and GZMF—as determined using validated anti-granzyme antibodies (Fig. 5o). Other granzymes could not be reliably assessed due to the lack of antibodies suitable for flow cytometry detection. While anti–PD-1 monotherapy modestly increased the frequency of GZMB- expressing CD8⁺ T cells in WT mice, the combination of anti–PD-1 and GSK-3 KD significantly enhanced both the mean fluorescence intensity (MFI) of perforin and granzyme expression (upper panels) and the percentage of CD8⁺ T cells expressing perforin and multiple granzymes GZMA, GZMB, GZMC and GZMF (lower panels). Consistent with our findings using pharmacological GSK-3 inhibition, these results demonstrate that genetic down-regulation of GSK-3 synergizes with PD-1 blockade to increase the cytolytic capacity of CD8⁺ TILs with the coordinated upregulation of perforin and a broad repertoire of multiple granzymes.

GSK-3 regulates Treg/Th balance and CD4 + T-helper dependency

A key next question was whether the induction of granzymes by combination therapy was driven by a role for GSK-3 in regulating CD4⁺ T-cell help for CD8⁺ T cells (Fig. 6). CD4⁺ T-cell help facilitates CD8⁺ T-cell differentiation and effector function.^4,5,6 We therefore first examined the impact of GSK-3 KD on CD4⁺ T-cell populations in the TME.

The alternative text for this image may have been generated using AI.

GSK-3 coordinates CD4⁺ T-cell helper–Treg balance and is required for CD4⁺ help-dependent CD8⁺ cytolytic function during PD-1 blockade. a Total CD4⁺ TIL frequency in WT and GSK-3 KD mice with or without anti-PD-1 treatment. No significant difference in overall CD4⁺ TIL abundance was observed between genotypes. b,c Frequencies of CD4⁺FOXP3⁻ helper T cells (b) and CD4⁺FOXP3⁺ Tregs (c) within the CD4⁺CD44⁺ TIL population (n = 3). Representative flow cytometry plots (c) demonstrate that anti-PD-1 treatment increased helper T cells to 73.8% of CD4⁺ TILs in GSK-3 KD mice (versus 40.2% in WT), while concomitantly reducing Tᴀᴇᶣ to 26.2% (versus 59.8% in WT). d,e GSK-3 expression is needed for Treg mediated suppression of the proliferation of conventional T-cells. In vitro Treg (T_reg) suppression assay profiles using different Tconv:Treg ratios where the proliferation of prelabelled conventional T-cells (T_conv) was quantified by dye dilution (d; n = 3). CD69 expression on conventional T-cells was measured as a secondary readout (e). f–h TIM-3 expression is reduced on GSK-3 KD CD44⁺CD4⁺ TILs: frequency of TIM-3⁺ cells (f; n = 3–4), representative flow cytometry dot plots (g), and TIM-3 MFI (h; n = 3–5). Anti-PD-1 treatment in GSK-3 KD mice reduced both the frequency (from 60.4 to 47.0%) and the intensity of TIM-3 expression on CD4⁺ TILs. i Representative flow cytometry dot plot confirming efficient in vivo CD4⁺ T-cell depletion following anti-CD4 antibody administration (300 µg i.p. at days −1, 7, and 14). j,k Tumor volume (j) and tumor weight (k) following CD4⁺ T-cell depletion in WT and GSK-3 KD mice treated with anti-PD-1. In WT mice, CD4⁺ depletion enabled anti-PD-1-mediated tumor regression (consistent with removal of T_reg mediated suppression). Conversely, CD4⁺ depletion completely abrogated the anti-tumor efficacy of anti-PD-1 in GSK-3 KD mice, resulting in rapid tumor outgrowth. l CD8⁺ TIL counts per gram of tumor following CD4⁺ depletion. CD4⁺ depletion prevented anti-PD-1-driven CD8⁺ TIL expansion in GSK-3 KD mice. m Expression of perforin, GZMA, GZMB, GZMC, and GZMF in CD44⁺CD8⁺ TILs following CD4⁺ T-cell depletion in GSK-3 KD mice treated with anti-PD-1, shown as MFI (upper panels) and frequency of positive cells (lower panels). CD4⁺ depletion markedly reduced granzyme and perforin expression, demonstrates the indispensability of CD4⁺ helper signals for the super-armed CTL phenotype. n Representative flow cytometry dot plot confirming in vivo T_reg depletion following anti-CTLA-4 antibody treatment (200 µg i.p. at days 4, 7, and 10). o Tumor volume in WT and GSK-3 KD mice following anti-CTLA-4-mediated T_reg depletion (n = 3–4). Anti-CTLA-4 promoted tumor regression in WT mice but conferred no additional benefit in GSK-3 KD mice, which already achieved effective tumor control. p,q Frequency of T_regs (p) and CD4⁺FOXP3⁻ helper T cells (q) among CD4⁺ TILs following anti-CTLA-4 treatment (n = 3). Anti-CTLA-4 selectively depleted Tregs while expanding the helper T-cell fraction. r,s IFN-γ production by CD4⁺ TILs in WT versus GSK-3 KD mice, expressed as frequency of IFN-γ⁺ cells (r) and MFI (s). GSK-3 KD CD4⁺ TILs produced significantly more IFN-γ, providing a potential mechanistic basis for enhanced CD4⁺-mediated support of CD8⁺ cytolytic responses. Data are presented as mean ± s.e.m. Statistical comparisons were performed by unpaired two-tailed Student’s t-test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; NS not significant. n = 4–5 mice per group

Firstly, we examined the status of CD4 TILs in WT versus GSK-3 KD mice. The down regulation of GSK-3 did not significantly alter the overall frequency of CD4⁺ TILs (Fig. 6a). However, TILs in the GSK-3 KD mice showed an increase in the presence of CD4⁺FOXP3⁻ helper T cells (Fig. 6b). This was especially evident in a comparison of the effects of anti-PD-1 in GSK-3 KD vs WT mice (Fig. 6b). A representative FACS plot shows that anti–PD-1 reduced helper T cells to 40.2% of total CD4⁺ TILs in WT mice, while dramatically increasing to 73.8% in GSK-3 KD mice (Fig. 6c). Conversely, the frequency of CD4⁺FOXP3⁺ Tregs increased from 36.8 to 59.8% in WT mice treated with anti–PD-1, as in previous reports.^94,95,96,97 However, their presence was markedly decreased to 26.2% in GSK-3 KD mice commensurate with the increase in CD4+ helpers. These results indicated that GSK-3 down regulation shifts the intratumoral CD4⁺ T-cell balance away from suppressive Tregs and toward helper subsets in the context of anti-PD-1 immunotherapy, thereby potentially enhancing the available CD4-dependent support for CD8⁺ effector T-cell responses.

To determine whether GSK-3 is required for Treg suppressor function, we compared the inhibitory capacity of WT versus GSK-3 KD Tregs in vitro (Fig. 6d, e). Purified CD4⁺CD25⁺ Tregs were co-cultured with increasing numbers of Tag-it–labelled wild-type conventional CD4⁺CD25⁻ responder T cells (Tconv) in the presence of soluble anti-CD3 and mitomycin C–inactivated splenocytes as feeder cells (see Materials and Methods). After 72 h, Tresp proliferation was quantified to assess Treg-mediated suppression.

Wild-type Tregs robustly suppressed responder T-cell proliferation, reducing the proliferation index across all Tconv:Treg ratios tested (black columns, Fig. 6d). Proliferation was reduced by 80% at a 2:1 ratio and by 50% at an 8:1 ratio. In striking contrast, GSK-3 KD Tregs were profoundly impaired in exercising suppression, achieving only 10% inhibition at a 1:1 ratio. Consistent with this loss of function, WT Tregs reduced expression of the activation marker CD69 on Tconv cells by over 50%, whereas GSK-3 KD Tregs had no significant effect (Fig. 6e). Together, these findings establish GSK-3 as an essential regulator of Treg suppressive function, highlighting its critical role in sustaining immune regulation and peripheral tolerance

Anti–PD-1 treatment in GSK-3 KD mice also reduced both the frequency and surface expression intensity of the exhaustion marker TIM-3 on CD4⁺CD44⁺ TILs (Fig. 6f). A representative flow-cytometry plot demonstrates a decrease in TIM-3⁺ cells from 60.4% in WT mice to 47.0% in GSK-3 KD mice (Fig. 6g). The MFI of TIM-3 expression was likewise diminished in GSK-3 KD mice treated with anti–PD-1 (Fig. 6h). The same treatment had no effect on TIM-3 expression in WT mice. Together, these data indicate that anti–PD-1 synergizes with reduced GSK-3 activity to limit TIM-3 upregulation—an inhibitory receptor commonly associated with T-cell dysfunction in chronic settings—thereby potentially promoting improved CD4⁺ helper function and downstream CD8⁺ effector activity.

Given the pronounced alterations in the CD4⁺ T-cell compartment, we next assessed the functional contribution of CD4⁺ T cells by depleting them at multiple time points before and during anti–PD-1 therapy (Fig. 6i, j). Anti-CD4 depletion efficiently reduced the presence of CD4⁺ T cells from 45.6% in untreated controls to 0.1% in CD4-depleted mice (Fig. 6i and Supplementary Fig. 15a). Strikingly, CD4 depletion enabled anti–PD-1 therapy to induce robust rejection of B16-F10 R1 tumors in Ctrl (WT) mice, as reflected by significant reductions in both tumor volume and tumor weight (Fig. 6j, k, respectively). This contrasted with the lack of an effect of anti-PD-1 on tumor growth in normal non-depleted WT mice (Fig. 5a). These findings suggested that the anti-CD4 depletion alleviated immunosuppression, mostly likely via the depletion of CD4+ Tregs, effectively restoringPD-1–mediated tumor control.

Conversely, in GSK-3 KD mice, which displayed an increased frequency of CD4⁺ helper T cells, CD4 depletion generated a different result by completely abrogating the therapeutic efficacy of anti–PD-1. This loss of response was evident from the resumption of rapid tumor growth and the marked increase in tumor burden (Fig. 6j, k). In contrast to WT mice—where anti–PD-1 treatment promoted robust CD8⁺ TIL accumulation—the same therapy failed to expand the CD8⁺ TIL population in GSK-3 KD mice following CD4 depletion (Fig. 6l).

Importantly, the absence of CD4⁺ T-cell help, in GSK-3 KD mice, also abrogatedthe anti–PD-1–induced upregulation of cytotoxic effector molecules within the CD8⁺ TIL compartment. Expression of perforin, GZMA, GZMB, GZMC, and GZMF was markedly reduced, as reflected by decreases in both the MFI (upper panel) and the percentage of granzyme-expressing cells (lower panel) (Fig. 6m). These results indicate that GSK-3 deficiency enhances CD8⁺ effector responses through CD4⁺ T-cell help, and that this helper function is indispensable for sustaining PD-1 blockade–mediated tumor control.

To further confirm the contribution of Treg-mediated suppression to the anti-tumor response, we examined the effects of Treg depletion using anti–CTLA-4 treatment (Fig. 6n, o). Administration of anti–CTLA-4, which effectively depleted CD4⁺FOXP3⁺ Tregs (Fig. 6n; Supplementary Fig. 15b), resulted in a pronounced reduction in tumor size in both WT and GSK-3 KD mice Fig. 6o). However, while anti–CTLA-4 facilitated tumor regression in WT mice, it produced no additional benefit in GSK-3 KD mice, where tumors were efficiently rejected. In other words, GSK-3 KD mice rejected B16-F10 R1 tumors irrespective of Treg depletion, whereas WT mice required Treg removal to achieve comparable tumor control. This finding contrasts with the earlier observation that anti-CD4 treatment impaired tumor regression in GSK-3 KD mice (Fig. 6j). As a control, anti–CTLA-4 treatment significantly reduced the number of CD4⁺FOXP3⁺ Treg TILs (Fig. 6p), while increasing the presence of CD4⁺FOXP3⁻ conventional T cells (Fig. 6q), confirming the selectivity of Treg depletion.

Collectively, these results indicate that while both anti-CD4 and anti–CTLA-4 target CD4⁺ subsets, their functional outcomes differ fundamentally. Loss of CD4⁺ helper cells in GSK-3 KD mice abolished anti–PD-1–mediated tumor regression, whereas Treg depletion had no additional effect, suggesting that reduced GSK-3 expression heightens CD8⁺ T-cell dependence on CD4⁺ helper signals. As expected, depletion of CD8⁺ T cells impaired tumor control in both WT and GSK-3 KD mice (Supplementary Fig. 16), confirming the central role of cytotoxic T cells in mediating tumor rejection.

Finally, to elucidate potential mechanisms underlying CD4⁺ helper activity, we analyzed cytokine production in GSK-3 KD CD4⁺ T cells. These cells exhibited a striking increase in IFN-γ production, as demonstrated by elevations in the frequency of IFN-γ–expressing CD4⁺ T cells (Fig. 6r) while also increasing the level of IFN-γ expression as expressed as an MFI (Fig. 6s). Enhanced IFN-γ secretion by CD4⁺ T cells promotes CD8⁺ T-cell activation, survival, and cytolytic function, providing a potential mechanism by which CD4⁺ help sustains effective antitumor CTL responses.

Together, these findings demonstrate that CD4⁺ T-cell subsets exert context-dependent effects on PD-1 blockade: in WT mice, CD4⁺ Tregs constrain anti-tumor immunity, whereas in the GSK-3 KD setting, CD4⁺ helper T cells play the dominant role for maintaining CD8⁺ T-cell cytotoxicity and achieving durable tumor control.