Gastric cancer (GC) takes the fifth place on most common cancers worldwide and it is currently the third cause of death by malignant tumors (1–3). In Chile, GC is a principal cause of cancer death (2, 3), with most patients being diagnosed with advanced or metastatic disease (4, 5). While the five-year survival rate for GC patients is 30.4%, this rate falls to approximately 5% in metastatic GC patients, reducing life expectancy to less than a year (6). In the absence of a standard-of-care treatment for advanced or recurrent GC after chemotherapy failure, there is a clinical requirement for new treatment options.

Pembrolizumab, a PD-1 blocking antibody, was granted accelerated approval for the treatment of patients with recurrent locally advanced or metastatic gastric or gastroesophageal junction adenocarcinoma (7). Although this therapy showed a remarkable 56% overall response rate (ORR) on the KEYNOTE-052 trial, the most recent published study shows a 22.7% ORR (8). Currently the only companion diagnostic available is an FDA approved test to determine tumor PDL-1 expression, but despite this selection 80% of patients show no clinical benefit (8), losing their opportunity to receive an effective treatment and suffering the elevated economic burden of this treatment. Similar results have been shown on CHECKMATE trials for another αPD-1 antibody, Nivolumab. In a phase 1/2 study in chemotherapy-refractory GC patients, second line Nivolumab treatment delivered a 26% ORR (44% ORR in PD-L1⁺ tumors) (7). Thus, while checkpoint therapy is offering enhanced overall survival to a significant number of patients, a better knowledge of GC immune escape mechanisms is required to identify possible predictive biomarkers and design new therapeutic approaches to overcome resistance. Of note, most proposed biomarkers are based on static measures and therefore cannot be evaluated through time and hence fail to rule out acquiring resistance.

Tim-3 is expressed on the most dysfunctional subset among tumor-infiltrating CD8⁺PD1⁺ T cells in cancer (9–11), and its upregulation in T cells exacerbates tumor progression in murine models (12). Therefore, Tim-3 has been proposed as a marker that identifies both terminally differentiated effector cell and irreversibly exhausted T cells (13–17). Strikingly, Tim-3 upregulation can be observed after αPD1 therapy (18). Additionally, the simultaneous blockade of Tim-3 and PD1 increases T cell responses compared with anti-PD 1 monotherapy (9, 11, 19–22). On the other hand, Tim-3⁺Treg are enriched in the tumor and correlate with tumor aggressiveness and progression (23–25). Tim-3⁺T_reg display an enhanced immune-suppressive function that can be abrogated by Tim-3 blockade (23, 26–28). In fact, αTim-3 synergizes with αPD1 therapy to reduce T_reg infiltration and increase CD8⁺ T cell infiltration (9). Tumor Tim-3⁺T_reg accumulation precedes CD8⁺ T cell dysfunction, and that their early depletion prevents CD8⁺ T exhaustion (23).

Gal-9 expression is found in 57% of gastric tumors, a key Tim-3 ligand associated with poor prognosis (29–31). Moreover, Tim-3 has also been proposed as a prognostic marker for solid tumors including GC, with high levels of Tim-3 expression associated with poor survival (32). However, the presence of an immune-suppressive tumor microenvironment and redundancy between inhibitory immune checkpoints may be responsible for why a high percentage of GC patients do not show clinical benefit of ICB.

An emerging target is the inhibitory receptor Tim-3 which has been reported to serve as Galectin-9 receptor. Galectin-9 has been widely implicated in CD8+ T cell exhaustion; however, its role in gastric cancer has not been previously characterized. Our findings demonstrate that Gal-9 is sufficient to enhance immune checkpoint expression even in the presence of anti-PD-1 blockade ex vivo, suggesting that it may contribute to therapy resistance. Further clinical studies are required to validate its role and assess its potential as a predictive biomarker in a clinical setting.

Gal-9 role in GC biology has been understudied and there are contradicting studies. Gal-9 expression was associated with improved survival, but only when late stages were ruled out (36). Similarly, cytoplasmic Gal-9 in GC cell lines suppresses migration, invasion, and epithelial-mesenchymal transition and inhibits metastasis (37). Controversially, we observed that high levels of LGALS9 significantly associate with poor outcomes and that extracellular Gal-9 effectively induces cell migration and invasion through its sugar binding domains. Similarly, other authors found that Gal-9 associates with poor prognosis on an Asian cohort of patients (31). This increase in invasion was not only observed on cancer cells but also on gastric epithelial cells, suggesting that this could be in fact a physiological function of extracellular Gal-9. Taken together, it is likely that Gal-9 has different functions depending on its subcellular expression, where cytoplasmic and extracellular Gal-9 display different biological functions.

CD8⁺ T cells are key elements of the cancer immune response, and their dysfunctional states are widely associated with poor clinical outcomes and immunotherapy resistance (38). A reduced effector function, including the production of TNF, IL-2 and IFNγ has been previously reported in dysfunctional CD8⁺ T cells (39–42). It must be noted that dysfunctional T cells are a highly heterogeneous population. Taking this into consideration, it has been suggested that this T cell subset can be classified into ‘pre- dysfunctional’, ‘early dysfunctional’ and ‘late dysfunctional’ cells (43). However, it must be kept in mind that T cell dysfunction is not a binary state, but rather a continuum of different states that culminate into a terminally exhausted phenotype leading to senescence. The transcriptomic analysis brought to light that a molecular signature of dysfunctional CD8⁺ positively correlates with tumor LGALS9, supporting that Gal-9 is associated with CD8⁺ dysfunction in GC. To further evaluate this hypothesis, we performed in vitro experiments with T cell primary cultures. Gal-9 increased both PD-1⁺Tim-3⁺ and PD-1⁺Tim-3⁺Lag3⁺ populations, which have been respectively suggested as early dysfunctional and late dysfunctional T cells. Importantly, the appearance of these sub-populations was accompanied by a dose-dependent impairment of cell division. It has been proposed that CD8⁺ T cells retain proliferative capacity during their transition from the pre-dysfunctional state to an early dysfunctional state but lose this capacity at the stage of more profound, ‘late’, dysfunction, either because of an intrinsic block or because their high inhibitory receptor expression suppresses T cell activation (40, 43–47). Herein, we demonstrate that CD8⁺ T cell proliferative capacity is impaired by Gal-9, and that it further drives an exhausted-like phenotype, increasing the frequency of PD-1⁺Tim-3⁺ exhausted-like progenitors and terminally exhausted cells characterized by the expression of Lag3 and failed IFNγ production. Moreover, co-treatment with a Tim3- blocking antibody prevented this effect. Therefore, it is likely that Gal-9, acting through Tim-3 engagement, is sufficient to increase the frequency of these populations. Validating these findings, previous reports show that blockade Gal-9 or Tim-3 restores CD8⁺ T cell proliferation (48). A variety of transcription factors regulate CD8+ T cell dysfunctionality, including TOX and TCF1 [Reviewed in (43)]. TCF1 and TOX appear to be determinants for T cell fate regarding a commitment into a stem-like phenotype or a dysfunctional phenotype in murine models (49). Since our results indicate that Gal-9 is driving the differentiation process into late- dysfunctional cells, further studies should evaluate if this protein can regulate these transcription factors and furthermore and increase the expression of the late-dysfunction marker CXCL13 (40, 43).

Tumour-resident T_reg cells have been shown to counteract tumor-specific immune responses by suppressing the infiltration and anti-tumor activity of CD8+ T cells (50). Moreover, T_reg accumulation has been associated with immunotherapy resistance in mice and patients (51, 52). Therefore, we evaluated if LGALS9 levels were also associated with increased T_reg infiltration. Strikingly, transcriptomic analysis revealed a strong positive correlation, suggesting that Gal-9 could increase Treg infiltration, possibly by expanding this population or alternatively increasing their trafficking into the tumor. Seki et al. showed that Lgals9 knockout (KO) mice displayed a reduced frequency of T_reg (52). A separate study showed that this KO mouse had impaired Foxp3 expression, that exogenous Gal-9 could restore Foxp3 function, and furthermore, that TGFβ1 induces Gal-9 expression in a feedforward loop (53). In addition, Gal-9 promoted TGFβ1-dependent induction of T_reg via the TGFβ/Smad signaling pathway. Given the evidence from mice models and our bioinformatics results, as a proof of concept we evaluated if Gal-9 could expand T_reg, using T cell primary cultures from healthy donors. In accordance with our hypothesis, we observed that Gal-9 at the three concentrations used effectively increased T_reg frequency. Furthermore, Gal-9 was able to upregulate Foxp3 expression, enrich the Tim3⁺T_reg population and potentiate iT_reg suppressive capacity.

It has been shown that Tim-3 blockade relieves T_reg mediated immunosuppression. Using the Tgfbr1/Pten double KO mouse model, the application of an αTim- 3 antibody reduced T_reg count and restored IFNγ production haltering tumor growth (27, 28). Tim-3 appears not only to play a key role on Treg function in cancer, but also in autoimmune diseases such as osteoarthritis, where a reduction of Tim-3 expression on T_reg associates with a decreased production of IL-10 (26). Besides Tim-3, Gal-9 has been reported to act through other receptors such as CD44 and CD137 (54). In murine models where Gal-9 could expand Treg, this action was principally attributed to an interaction with the CD44 receptor that enhanced the stability and function of these cells (53). However, our observations in human peripheral T cells, Gal-9 action appears to be predominantly mediated through the Tim-3 receptor, although we cannot rule out that Gal-9 could also, albeit partially, mediate changes through alternative receptors. Mechanistically, it has been proposed that Tim3 can enhance Treg effector-like phenotype through glycolytic metabolism (PMID: 34525351). In accordance, Gal-9 has been proposed to regulate T cell function through the PI3K/AKT/mTOR pathway (PMID: 38141929). Taking together it is possible that Gal-9 signaling through Tim-3 modulates cell metabolism by fine-tuning of the PI3K/AKT/mTOR pathway. While this remains to be evaluated, we can conclude that Tim-3 is required for Gal-9 mediated T_reg expansion.

In the field of immunotherapy, there remains an open question of whether ICB acts by reinvigorating a tumor-resident T cell population or mobilization of T cells from outside the tumor also occurs. Studies in mice models revealed that stem-like T cells were critical for obtaining a positive tumor response upon application of single agent αPD-1 therapy or αPD-1 and αCTLA-4 combination therapy. However, by depleting stem-like T cells and thus leaving late-dysfunctional T cells, treatment with ICB still reduced tumor growth (14), thus indicating that this treatment not only prevents exhaustion but can also rescue exhausted cells. Other studies presented similar results (40, 46), supporting that early-dysfunctional T cells, but not late-dysfunctional T cells, may be responsible for the favorable responses observed with ICB therapies [Extensively reviewed in (43)]. Tim-3 targeting promotes IL-12-dependent antitumor immunity without altering immune infiltration (55), thus supporting a key role for this pathway in the peripheral activation of T cells and possibly early fate establishment during antigen presentation.

Herein we report that a CD8⁺ dysfunction molecular signature is positively correlated with LGALS9 levels. Importantly, the same signature applied to pre-treatment transcriptomic data from patients with melanoma who subsequently received ICIs, consistently out-performed all other candidate predictive biomarkers tested, including PD-L1 levels, tumor mutational burden, and an IFNγ signature (33). Given the results of our bioinformatic analysis, we evaluated if Gal-9 could provide a bypass to the presence of a PD1 blocking antibody. Even in presence of the αPD-1, Gal-9 increased the frequency of both the early (PD-1⁺Tim3⁺) and late (PD-1⁺Tim3⁺Lag3⁺) exhausted phenotypes. The participation of the Tim-3 receptor is highlighted by the effectively decreased frequency of these subsets. Of note, it was published that Gal-9 could also bind to PD-1, in a way that promoted a lattice formation between PD-1 and Tim-3, which was proposed to increase the stability of dysfunctional T cells, possibly by reducing Tim-3 mediated cell death (56). These results have clinical applicability as they suggest that the axis Gal-9/Tim-3 T cells can push T cells into a dysfunctional state even in the presence of an ICB antibody ( Figure 6 ). Interestingly these results indicate not only that Gal-9 presence could provide a bypass to αPD-1 therapy and thus contribute to ICB resistance, but also shine a light on αPD-1’ mechanism. In our experiments, αPD-1 treatment effectively reduced the amount of dysfunctional T cells and thus a principal modus operandi of αPD-1 therapies may be to obstruct the differentiation of T cells into a dysfunctional state. Although the main discussion within the field has been that a durable response to α-PD1 therapy requires the presence of tumor-specific T cells with low levels of dysfunction, our results point to another possibility. Whether it is merely the levels of exhaustion, or also the presence of inhibitory ligands that are the determinants is yet to be ascertained. As T cell exhaustion is led by a robust system of inhibitory pathways, under the activation of other co-inhibitory pathways such as Gal-9/Tim-3, T cells may undergo this differentiation regardless of the αPD-1 treatment. In line with this idea, an immuno-predictive score for neuroblastoma based on the levels of inhibitory and activating immune-checkpoint genes, IMPRES, was found to be associated with immunologically hot tumors and longer overall survival in patients with untreated metastatic melanoma (57). In predicting responses to ICB in this setting, IMPRES strikingly achieved an overall accuracy of AUC = 0.83, outperforming existing predictors and capturing almost all true responders while misclassifying less than half of the non-responders. This work strengthens the hypothesis that a robust network of regulatory checkpoints is in operation to maintain the tumor immunosuppressive microenvironment ( Figure 6 ). Noteworthy, our results showed that a combinatory strategy using αPD-1 together with αTim-3 effectively prevented the effects of Gal9 on T cell fate, further supporting the idea of combining ICB for cancer treatment. Encouragingly, there are 63 clinical trials of antibodies targeting Tim-3 either alone or in combination with PD-1 blockers. Moreover, Gal-9 is a secreted protein that can be detected in plasma, and thus future studies should examine the clinical benefit of evaluating Gal-9 and intra-tumor Tim-3 levels as a companion diagnostic for αPD-1 + αTim-3 combinatorial immunotherapy. Furthermore plasmatic Gal-9 may prove value as a resistance risk indicator for patients receiving ICB monotherapy and could signal when to begin combinatorial therapy.

In summary, this study combines transcriptomic analysis with ex vivo functional assays to uncover a role for Galectin-9 in promoting CD8⁺ T cell dysfunction and Treg expansion in gastric cancer. A major strength is the integration of patient-derived data with mechanistic in vitro validation. However, reliance on the TCGA database, which lacks treatment metadata and may introduce selection bias, limits the ability to generalize findings to treated populations. Further in vivo and clinical studies are warranted to confirm these observations.