Terminal T-cell exhaustion predicts tumor control independent of immunotherapy in ovarian cancer

Abstract

T-cell exhaustion is typically studied in the context of immune checkpoint blockade, where proliferation and reinvigoration of exhausted cells drives therapeutic responses. However, terminal exhaustion may also represent a marker of chronic tumor-specific activation, raising the possibility that exhausted T cells reflect ongoing endogenous tumor control. Here, we sought to evaluate T-cell exhaustion as a prognostic marker using high-grade serous ovarian cancer (HGSC), an immunotherapy-resistant malignancy, as a model. In a cohort of 80 patients with stage III/IV HGSC, we assessed T-cell infiltration and exhaustion according to homologous recombination (HR) deficiency status. While overall immune infiltration was comparable between HR-deficient and proficient tumors, terminally exhausted CD8 and conventional CD4 T cells were enriched in HR-deficient tumors, where their presence correlated with improved progression-free survival. These findings suggest that exhausted T cells may indicate protective immunity even outside the context of immunotherapy and underscore their prognostic relevance in solid tumors.

Introduction

T-cell exhaustion, characterized by the expression of immune checkpoints (ICs), has been extensively studied since the advent of immune checkpoint blockade (ICB). We and others have shown that tumor antigen-specific CD8 T cells in cancer patients exist at various stages of exhaustion, with circulating cells exhibiting features of early exhaustion, and tumor-infiltrating cells often reaching a terminally exhausted state^1–4. In contrast, bystander T cells specific for viral antigens at the tumor site typically lack IC expression¹. CD4 T cells exhibit a similar pattern, with tumor antigen-specific cells progressing toward terminal exhaustion within tumors⁵. As the primary targets of ICB, exhausted T cells have been largely studied in the context of treatment response, and PD-1 blockade has been shown to enhance their proliferation and to reinvigorate their effector functions. Accordingly, the presence of exhausted CD8 T cells has been associated with response to ICB^1,2,4,6–9. Yet beyond their therapeutic relevance, the accumulation of exhausted T cells may serve as a sentinel of active, tumor-specific immunity and, as such, their contribution to tumor control may extend beyond responsiveness to ICB.

High-grade serous ovarian cancer (HGSC) is the most lethal gynecologic malignancy^10,11. Response rates to ICB in ovarian cancer have remained below 15% in clinical trials evaluating single-agent immunotherapy, administered alone or with chemotherapy, in both primary and recurrent settings^12–15. Nonetheless, rare but durable responses, uncommon with chemotherapy, have been observed in ICB-treated patients¹⁶, suggesting the existence of meaningful antitumor immune responses in a subset of cases. Indeed, the seminal study by Zhang and colleagues demonstrated the prognostic value of intraepithelial T cells in ovarian cancer¹⁷, though later studies failed to consistently link clinical outcome with overall T-cell infiltration^18,19. We and others have shown that circulating and tumor-infiltrating CD4 and CD8 T cells specific for neoantigens and cancer-testis antigens are frequently detectable in ovarian cancer patients^1,5,20–24. Moreover, adoptive transfer of vaccine-primed T cells has shown early signs of efficacy²⁵. Collectively, these findings underscore the immunogenic potential of HGSC, suggest that adaptive immunity is part of its pathophysiology, and support its use as a model to explore the prognostic value of exhausted T cells outside the context of immunotherapy.

The introduction of PARP inhibitors (PARPi) as maintenance therapy for HGSC patients with homologous recombination deficient (HRD) tumors^10,11 has improved clinical outcomes and opened new avenues for combinatorial strategies. It also compels a reassessment of HGSC pathophysiology through the lens of homologous recombination (HR) status. Immune profiling has revealed distinct characteristics between HRD and homologous recombination proficient (HRP) tumors. In multiple cancer types, defects in DNA damage repair, as seen in HRD tumors, have been associated with accumulation of cytoplasmic DNA, activation of the cGAS/STING pathway, and induction of type I interferons^26–28, promoting T-cell priming and infiltration. While the overall tumor mutational burden (TMB) of ovarian tumors is moderate²⁹, HRD tumors tend to exhibit higher TMB³⁰ and increased neoantigen load^31,32, both features consistent with greater immunogenicity. CD3 tumor-infiltrating lymphocytes (TILs) are enriched in HRD tumors³³, yet CD3 density alone does not predict survival within HR subgroups, indicating the need for deeper characterization of the T-cell compartment. In particular, bulk TIL quantification may obscure underlying immune complexity, including immune evasion processes such as T-cell exhaustion^34,35. Recent multiomic studies of advanced disease highlight immune escape as a key feature of tumor evolution³⁶. Together, these observations underscore the importance of refined immune profiling performed in the context of HR status to clarify the prognostic relevance of T-cell exhaustion, a dual-feature marker capturing both antitumor immunity and immune evasion, in ovarian cancer.

Here, using HGSC as a model and considering HR status, we assessed the prognostic significance of CD4 and CD8 T-cell exhaustion.

Results

Patient characteristics

Eighty patients diagnosed with stage III/IV HGSC were included in the study (Table S1). Forty-three presented with HRD tumors, of which 29 were BRCA1 or BRCA2 mutated. All patients received the standard of care with carboplatin-based chemotherapy following ESGO recommendations. Sixty-nine patients underwent debulking surgery, and complete cytoreduction was achieved in 67 patients, with 28 (96.6%) in the HRP and 39 (97.5%) in the HRD group. Thirty-five patients (43.8%) received neoadjuvant chemotherapy (NACT), with 18 patients (48.6%) in the HRP group and 17 (39.5%) in the HRD group. Maintenance treatment with Bevacizumab and PARPi was administered to 36 (45%) and 29 (36.3%) patients, respectively (Table S2). There were no significant differences in age at diagnosis, initial peritoneal cancer index (PCI) or the proportion of patients receiving NACT and adjuvant chemotherapy, between the HRD and HRP groups (Table S1). Baseline CA125 levels were significantly higher in patients with HRD tumors, with a median CA125 of 947.0 U/mL (range 53–4393), compared to 258.5 U/mL (range 0–8737) in patients with HRP tumors (p = 0.012).

T-cell infiltrate is not associated with tumor homologous recombination status

FFPE tumor samples, obtained prior to any chemotherapy, either at diagnostic laparoscopic surgery or at upfront debulking surgery, were used to assess the immune infiltrate by mIF using antibodies specific for CD3, CD8, CD4 and FOXP3. Based on these markers, the overall T-cell infiltrate was evaluated and used to determine the immune phenotype of tumors, that were classified as infiltrated, excluded or desert (Figure S1A). The majority of tumors exhibited an infiltrated immune phenotype (n = 39, 50.6%), followed by a desert phenotype (n = 28, 36.4%), while a smaller subset displayed an excluded phenotype (n = 10, 13.0%). No significant differences in immune phenotype were observed between HRD and HRP tumors (Figure S1B), nor among HRD tumors when comparing those with BRCA mutations to BRCA wildtype (HRDwtBRCA) cases (Figure S1C). While no statistically significant difference was observed in immune phenotype distribution, a desert phenotype was present in 58.3% of HRDwtBRCA tumors, versus 24.1% within the BRCA mutated group (Figure S1C).

We then quantified the total as well as the CD8 and CD4 T-cell intraepithelial and stromal infiltrates and analyzed their association with HR status. The total intraepithelial and stromal T-cell infiltrates, measured as the number of CD3⁺ cells/mm² in each area, were found at similar levels in HRD and HRP tumors (Fig. 1A–C). Further analysis of the infiltrating T-cell populations showed no significant differences in the numbers of CD8 T cells (CD3⁺CD8⁺) and CD4 T cells, both conventional (CD4 Tconv, CD3⁺CD4⁺FOXP3⁻) and T regulatory (CD4 Treg, CD3⁺CD4⁺FOXP3⁺), or in the CD8 T cells/CD4 Treg ratio, between HRD and HRP tumors (Fig. 1A–C). Among HRD tumors, however, there was a significantly higher density of intraepithelial T cells and of stromal and intraepithelial CD8 T cells and an increased stromal and intraepithelial CD8 T cells/CD4 Treg ratio, in tumors with BRCA mutations compared to HRDwtBRCA (Fig. 1D, E).

Fig. 1. T-cell infiltrate is not associated with tumor HR status.

Abundance of T-cell subpopulations was determined using quantitative mIF in pre-treatment samples from HGSC patients. A Representative images of HRD and HRP tumors stained with H&E, anti-CD3, -CD8, -CD4 and -FOXP3 mAbs are shown. A brightness correction is applied to the representative images to make the stainings visible. B–E Numbers/mm² of Total T cells (CD3⁺), CD8 T cells (CD3⁺CD8⁺), CD4 Tconv (CD3⁺CD4⁺FOXP3^-) and CD4 Treg (CD3⁺CD4⁺FOXP3⁺) as well as CD8 T cells/CD4 Treg ratio, in HRD (n = 41) and HRP (n = 36) tumors (B, C) and in BRCA1/2 mutated (n = 29) or HRDwtBRCA (n = 12) tumors (D, E), in the intraepithelial (B, D) and stromal (C, E) areas are shown. For CD8 T cells/CD4 Treg ratio, HRD (n = 40 intraepithelial, 41 stromal), HRP (n = 34) (B, C) and BRCA1/2 mutated (n = 28 intraepithelial, 29 stromal), HRDwtBRCA (n = 12) (D, E). Individual values and mean ± SD are shown. Normality was assessed using Shapiro-Wilk test. Statistical significance was assessed using Mann-Whitney test. *p < 0.05.

To expand the appraisal of the overall immune landscape in our cohort, we analyzed, by IF, the expression of PD-L1 (Figure S2A), which may reflect an ongoing immune response³⁷. We found low numbers of intraepithelial and stromal PD-L1⁺ cells (Figure S2B), in agreement with previous studies in ovarian cancer³⁸. No significant differences in the numbers of PD-L1⁺ cells were observed according to HR status (Figure S2B, C).

Altogether, these results did not reveal major differences in the immune phenotype, overall T-cell infiltrate, or PD-L1 expression according to the HR status, underscoring the need for a more detailed characterization of functionally relevant T-cell subpopulations.

HR deficiency is associated to higher infiltration by terminally exhausted T cells

To further characterize the T-cell infiltrate, we performed ex vivo phenotyping of CD8 and CD4 T cells, isolated from tumor single cell suspensions, using flow cytometry with an antibody panel designed to assess T-cell differentiation and exhaustion. We found higher proportions of memory (CD45RA^– CCR7^+/-) CD8 T cells and CD4 Tconv in HRD compared to HRP tumors (Figure S3A, B). Of note, although the observed differences were statistically significant, memory T cells represented the most abundant subpopulation regardless of HR status and T-cell populations. Further analyses were therefore focused on memory T cells.

We and others have shown that terminally exhausted CD8 T cells at the tumor site, which encompass tumor-specific T cells, are characterized by the co-expression of high PD-1 levels and of TIM-3 and CD39^1,3,39. We therefore set out to assess the expression of these markers in tumor-infiltrating memory CD8 T cells according to HR status. We assessed the proportions of cells expressing combinations of these markers and the mean fluorescence intensity (MFI) of PD-1 staining as an indicator of its expression levels. We observed that the levels of expression of PD-1 in CD8 T cells were increased in HRD compared to HRP tumors (Fig. 2A). Moreover, the proportions of PD-1^highTIM-3⁺, but not PD-1^highTIM-3^-, memory CD8 T cells were higher in HRD than HRP tumors (Fig. 2B). In HRD tumors, the PD-1^highTIM-3⁺ subset expressed higher levels of PD-1 (Fig. 2C), and a larger proportion of these cells expressed CD39 compared to HRP tumors (Fig. 2D). Lastly, PD-1^highCD39⁺ cells were more abundant in HRD compared to HRP tumors, while no differences were uncovered when PD-1^highCD39^- cells were considered (Fig. 2E).

Fig. 2. HR deficiency is associated to higher infiltration by terminally exhausted CD8 T cells.

T cells were isolated from HRD and HRP tumors, stained with mAbs and analyzed by flow cytometry. A–I Comparisons of HRD and HRP tumors for the indicated parameters. A Representative histograms and MFI of PD-1 staining in memory CD8 T cells are presented (left). n = 32 tumors from 25 HRD patients and 25 tumors from 24 HRP patients. B Representative dot plots, proportions of cells in each quadrant and MFI of PD-1 in PD-1^highTIM-3⁺ memory CD8 T cells, between brackets, are shown (left). n = 44 tumors from 32 patients and 26 tumors from 25 HRP patients. C n = 32 tumors from 25 HRD patients and 24 tumors from 23 HRP patients. D Representative histograms and proportions of CD39⁺ expression in PD-1^highTIM-3⁺ memory CD8 T cells are shown (left). n = 39 tumors from 29 HRD patients, 25 tumors from 24 HRP patients. E Representative dot plots, proportions of cells in each quadrant and MFI of PD-1 in PD-1^highCD39⁺ memory CD8 T cells, between brackets, are presented (left). n = 39 tumors from 29 HRD patients, 26 tumors from 25 HRP patients. F Representative histograms and MFI of PD-1 staining are presented (left). n = 32 tumors from 25 HRD patients, 27 tumors from 25 HRP patients. G Representative dot plots, proportions of cells in each quadrant and MFI of PD-1 in PD-1^highCD39⁺ memory CD4 Tconv, between brackets, are shown (left). n = 36 tumors from 29 HRD patients, 28 tumors from 26 HRP patients. H n = 32 tumors from 25 HRD patients, 27 tumors from 25 HRP patients. I Representative histograms and proportions of TIM-3⁺ cells in PD-1^highCD39⁺ memory CD4 Tconv are shown (left). n = 36 tumors from 29 HRD patients, 28 tumors from 26 HRP patients. Individual values and mean ± SD are shown in summary graphs. Multiple tumors from the same patient were analyzed when available. Normality was assessed using Shapiro-Wilk test. Statistical significance was assessed using Mann-Whitney in (A, B, D–G). T-test was used in (C, H, I). Welch’s correction was applied in (H). *p < 0.05; **p < 0.01.

Terminal CD4 T-cell exhaustion is also characterized by the expression of high PD-1 levels associated to CD39 and TIM-3 expression⁵. We therefore assessed the expression of these markers in the CD4 Tconv infiltrate. In memory CD4 Tconv, PD-1 expression levels were significantly higher in HRD than in HRP tumors (Fig. 2F), whereas no differences were observed in the proportions of PD-1^highCD39⁺ or PD-1^highCD39^- CD4 Tconv in HRD tumors (Fig. 2G). Both the level of expression of PD-1 in PD-1^highCD39⁺ CD4 T conv and the proportion of cells co-expressing TIM-3 were significantly higher in HRD tumors (Fig. 2H, I). In agreement with their role at the effector phase as enhancers of tumor-specific CD8 T cells⁴⁰, the proportions of terminally exhausted CD4 Tconv (PD-1^highCD39⁺) were positively correlated to those of terminally exhausted CD8 T cells (PD-1^highTIM-3⁺ and PD-1^highCD39⁺), in particular within the HRD group (Figure S4A, B). In addition, a positive correlation between the proportions of tumor infiltrating CD4 Treg and those of terminally exhausted CD8 T cells was found, particularly in the HRD group (Figure S4C). Finally, within the HRD group, no differences in the proportions of terminally exhausted T cells were observed between BRCA mutated and HRDwtBRCA tumors (Figure S5A–I).

Taken together, these results highlighted that HRD tumors harbored a T-cell infiltrate enriched in terminally exhausted CD8 and CD4 Tconv.

Terminally exhausted T cells are associated with clinical outcome in patients with HRD tumors

We have previously shown that terminally exhausted T cells infiltrating epithelial tumors are enriched in cells at late differentiation stages, characterized by tissue residency markers and high effector potential¹. Consistently, analysis of scRNA sequencing data from CD8 T cells infiltrating ovarian tumors⁴¹ confirmed findings from our group and others^1,3,4, showing that the terminally exhausted cluster, defined by expression of CXCL13, CD39, and the immune checkpoints PD-1, TIGIT, CTLA-4, and TIM-3, was enriched in cells lacking CCR7 and CD127 (Figure S6A–C). These cells also expressed tissue residency markers, including CD103, CD69, and CD49a, and lacked expression of S1PR1 (Figure S6D). They expressed high levels of granzymes and perforin, consistent with retained effector potential (Figure S6C) supporting the relevance of assessing the association between terminally exhausted T-cell infiltration and clinical outcome, which we evaluated using univariable and multivariable analyses.

The median follow up of patients in the cohort was 30.8 months (95%CI: 26.3; 39.7). Median PFS was 26.0 months (95% CI: 20.1; NR) and there were a total of 38 events (47.5%) (Fig. 3A). No association was observed between PFS and clinical features at diagnosis, except for the tumor load measured by PCI, associated with worse PFS (Fig. 3B, C). Landmark analysis based on maintenance treatment showed improved clinical outcome in patients treated with PARPi, whereas no significant benefit was provided by the addition of bevacizumab (Fig. 3D). When we explored the overall immune infiltrate, we found no significant prognostic value whether the infiltrate was evaluated by the immune phenotype (inflamed, excluded and desert) or numbers of T cells and CD8 T cells per mm² (Fig. 3E).

Fig. 3. Univariable analyses of PFS.

A Kaplan-Meier curve illustrating the PFS of the cohort. B, E Forest plots of univariable analyses of PFS using Cox model for clinical parameters (B) and imaging parameters (E, upper panel). C–E Forest plots of univariable analyses of PFS using LogRank test for HRD status and treatment strategies (C, D) and imaging parameters (E, lower panel). 9-month Landmark approach was applied in (D). Parameters with p < 0.05 are depicted in green and the corresponding p-values are annotated in bold (B–E).

We then explored the prognostic value of CD8 T-cell exhaustion according to exhaustion features found as significantly increased in HRD compared to HRP tumors (Fig. 2). Univariable analyses revealed significant association between terminal CD8 T-cell exhaustion and PFS in the overall cohort (Fig. 4A). Notably, this association was maintained in patients with HRD tumors but was absent in those with HRP tumors (Fig. 4B). Exhaustion of CD4 Tconv did not appear as strongly associated to PFS, except for one parameter (PD-1 MFI in PD-1^highCD39⁺) (Fig. 4A, B). Multivariable analyses adjusted on HRD status, NACT and age at diagnosis, performed on the overall cohort, confirmed the PFS benefit of CD8 T-cell as well as CD4 Tconv terminal exhaustion (Fig. 4C). We then performed PFS analyses according to T-cell exhaustion by dividing the cohort in two groups, high and low, based on the median value of all exhaustion parameters that showed significant prognostic value in multivariable analyses (Fig. 4C). Patients whose tumors had high exhaustion features displayed longer PFS (Fig. 4D–J).

Fig. 4. Terminally exhausted T cells are associated with clinical outcome in patients with HRD tumors.

A, B Forest plots of univariable analyses of PFS using the Cox model for exhaustion parameters in the entire cohort (A) and in HRD (upper) and HRP (lower) subgroups (B). C Multivariable analyses adjusted on HRD status, neaodjuvant chemotherapy and age at diagnosis in the entire cohort. D–J Kaplan-Meier curves of PFS according to exhaustion parameters (high versus low values: cut-off at median): MFI of PD-1 in Memory CD8 T cells (D) and in PD-1^highTIM-3⁺ CD8 T cells (E); proportions of CD39⁺ cells in PD-1^highTIM-3⁺ memory CD8 T cells (F); proportions of PD-1^highCD39⁺ cells in memory CD8 T cells (G) and in memory Tconv (I); MFI of PD-1 in memory CD4 T conv (H) and in PD-1^highCD39⁺ memory Tconv (J). Comparison between groups (high versus low) was performed with the log-rank test. Parameters with p < 0.05 are depicted in green and the corresponding p-values are annotated in bold (A–C).

Discussion

In pretreatment samples from a clinically homogeneous cohort of patients with stage III/IV HGSC, we found that terminally exhausted CD8 and conventional CD4 T cells were enriched in HRD tumors and associated with longer PFS in univariate and multivariate analyses. In contrast, overall T-cell infiltration and PD-L1 expression did not differ significantly between HRD and HRP tumors, nor did they correlate with clinical outcome. These findings support our hypothesis that T-cell exhaustion may serve as a marker of chronic, tumor-specific immune activation, rather than mere dysfunction. By demonstrating that exhausted T cells carry prognostic value in a setting where ICB is largely ineffective, our study broadens the biological and clinical interpretation of exhaustion. It positions terminally exhausted T cells as indicators of ongoing endogenous tumor control, even in the absence of therapeutic immune modulation.

Exhausted T cells, characterized by the expression of ICs, have been extensively studied since the advent of ICB, where they represent direct therapeutic targets. The early definition of exhaustion as a terminal loss of effector function has been challenged⁴². Instead, exhaustion is now understood as a dynamic continuum, reflecting an adaptive response to chronic antigen stimulation. Early exhausted T cells retain proliferative capacity and contribute to the maintenance of the antigen-specific T-cell pool⁴². In contrast, the role of terminally exhausted T cells, such as those analyzed in our study, remains under debate. Recent work has shown that the loss of their epigenetic integrity leads to apoptosis of terminally exhausted T cells⁴³, supporting the concept that exhaustion enables the persistence of immune responses to restrain chronic infections and cancer when elimination is not achievable, rather than simply representing a nonfunctional end state. We and others have demonstrated that exhausted T cells can recover effector functions and contribute to tumor control under ICB^1,4,44, alongside the replenishment of the T-cell pool by proliferating early exhausted counterparts^1,2,6. Our current findings, showing an association between terminally exhausted T cells and improved clinical outcome in HRD HGSC patients not receiving immunotherapy, further support the role of these cells as active participants in endogenous cancer control.

The prognostic impact of exhausted T cells in ovarian cancer has been investigated in previous studies^45,46. Using a machine-learning approach applied to tumor RNA-seq data, Chen et al. developed a T-cell exhaustion score that predicted clinical outcome, with higher exhaustion associated with increased immune infiltration and improved OS⁴⁵. Using immunohistochemistry, mIF, and RNA-seq, Fucikova et al. reported that longer relapse-free survival (RFS) and OS were associated with increased CD8 T-cell infiltration and higher numbers of PD-1⁺ and LAG-3⁺ cells in tumors⁴⁶. Despite methodological differences, these studies converge with our findings in supporting a positive association between T-cell exhaustion markers and favorable disease outcome. Notably, neither study accounted for tumor HRD status. The use of flow cytometry in our study enabled precise discrimination between terminally exhausted T cells, both CD8 and CD4 Tconv, and T cells expressing PD-1 in the absence of TIM-3 or CD39, a phenotype compatible with activation or earlier stages of exhaustion, the latter showing no association with HR status or clinical outcome. In the study by Fucikova et al., TIM-3⁺ CD8 T cells were, however, associated with poorer OS⁴⁶, a result that contrasts with our observations. Such discrepancies may reflect differences in analytical approaches (mIF versus flow cytometry) and/or clinical characteristics of the cohorts, including higher complete cytoreduction rates in our study and disease stage distribution (stage III–IV versus I–IV). The importance of exhausted T cells in HRD HGSC has also been highlighted in patients receiving a combination of ICB and PARPi, where T-cell exhaustion was associated with response to therapy⁹.

The NACT setting could represent an opportunity to assess the impact of chemotherapy on terminally exhausted T cells. This could not be evaluated in our cohort due to the limited availability of post-NACT tumor samples. The effect of NACT on exhausted PD-1⁺ CD8 T cells has previously been assessed in a cohort of 162 patients with HGSC⁴⁷. Using IHC, this study reported no statistically significant differences in PD-1⁺ CD8 T-cell infiltration following NACT. While tumor HRD status was not addressed, the authors notably observed that PD-1⁺ CD8 T cells were positively associated with survival in pre-NACT samples, but not in post-NACT specimens. It would nevertheless be of interest to assess the effects of NACT, as well as neoadjuvant immunotherapy, not only on tumor-infiltrating terminally exhausted T cells^47,48 but also on peripheral early exhausted T cells, which may represent a reservoir capable of replenishing tumor-specific T-cell populations in residual or recurrent disease⁴⁹.

The fact that exhausted CD8 T cells are indicative of an ongoing antitumor response was supported by the concomitant infiltration of HRD tumors by terminally exhausted CD4 Tconv and by CD4 Treg. The positive correlation between the proportions of tumor infiltrating Treg and those of terminally exhausted CD8 T cells is in agreement with our previous work showing that Treg accumulation in ovarian tumors was associated to that of type 1 helper CD4 Tconv⁵⁰. Using terminal exhaustion as surrogate marker of ongoing antitumor adaptive immunity, we could assess differences according to HR status in a large cohort. These analyses further those assessing directly tumor antigen-specific T cells in ovarian cancer. Indeed, those studies uncovered circulating and tumor infiltrating tumor antigen-specific T cells in treatment-naïve patients^1,23, but did not, given the necessary cumbersome methodologies, address their quantitative differences according to HR status.

HR deficiency has been associated with higher TMB and increased activation of innate immune pathways, both features potentially leading to the priming of antitumor adaptive immune responses. Previous work from our group and others highlighted that only tumor-specific, but not bystander, T cells reach terminal exhaustion at the tumor site^1,51. Considering terminally exhausted T cells as a signature of an antitumor adaptive immune response, the results of our study support the higher immunogenicity of HRD tumors. These results are in agreement with those showing the enrichment of HRD tumors in exhausted T cells using transcriptomic approaches^35,52. A previous report showed higher proportions of exhausted T cells in BRCA-mutated tumors in comparison to BRCAwt ones, thus considering in the same group HRP and HRDwtBRCA tumors⁵³. We found no difference in the proportions of exhausted T cells between BRCA-mutated and HRDwtBRCA tumors, thus emphasizing the importance of considering a comprehensive definition of HR deficiency. It is noteworthy that not all HRD tumors exhibited high proportions of terminally exhausted T cells, a phenomenon that has also been reported in highly immunogenic malignancies such as melanoma and lung and microsatellite instable colorectal cancers^3,54. This heterogeneity reflects the fact that the accumulation of terminally exhausted T cells depends on multiple factors⁴². These include the priming of a tumor-specific immune response, which itself relies on the availability of tumor antigens, the presence of a competent T-cell repertoire, and adequate innate immune activation signals. Once primed T cells infiltrate the tumor bed, progression toward terminal exhaustion further depends on sustained antigenic stimulation by tumor cells or antigen-presenting cells, as well as on the local cytokine milieu.

In line with a previous study based on the ICON 7 clinical trial and data from the TCGA database³⁴, our results did not uncover an association between HR status and immune phenotype. In addition, overall T-cell infiltration was similar between HRD and HRP tumors. Whereas greater T-cell infiltration of HRD compared to HRP tumors was described³³, the HR signature in the study focused solely on mutations in HR genes, unlike the Myriad signature. In our study, numbers of intraepithelial and stromal PD-L1⁺ cells were not associated with HR status and were generally low, in agreement with previous studies³⁸. Of note, however, the reported expression of PD-L1 in ovarian cancer varies among studies and its prognostic value remains controversial^38,55–57. Taken as a whole, these discrepancies, added to those between studies assessing the prognostic value of the overall T-cell infiltrate^17–19, support the in depth assessment of specific immune processes, as performed here.

Beyond overall immune infiltration, increasing evidence indicates that the functional impact of tumor-infiltrating T cells critically depends on their spatial organization and interactions with other components of the tumor microenvironment. In ovarian cancer, the contribution of exhausted CD8 T cells to clinical outcome has been shown to be conditioned by their proximity to PD-L1⁺ tumor cells and myeloid cells, highlighting the importance of myeloid-T cell interactions during the effector phase of the immune response⁹. These considerations may partly explain why the rate of response to ICB is lower than the proportion of patients exhibiting high terminally exhausted T-cell infiltration, including among patients with HRD tumors. More broadly, distinct immune phenotypes integrating T-cell and myeloid cell states have been described in ovarian cancer and shown to associate with patient outcome⁵⁸. In addition, the presence of tertiary lymphoid structures (TLSs) has been linked to improved survival and enhanced T-cell activation in ovarian cancer⁵⁹. Together, these observations underscore the importance of considering the spatial and cellular organization of the tumor immune microenvironment, in addition to T-cell exhaustion states.

Our study has several limitations. Terminal T-cell exhaustion was assessed exclusively by flow cytometry and not by spatially resolved approaches such as mIF. While flow cytometry enabled a precise and biologically grounded definition of terminally exhausted CD8 and CD4 Tconv populations based on the co-expression of PD-1 with TIM-3 and/or CD39, it did not provide information on their spatial organization or proximity to tumor cells, myeloid cells, or TLSs, thereby limiting direct comparisons with studies relying on in situ technologies and precluding an integrated analysis of T-cell exhaustion within the full tumor microenvironment^9,58,59. Our analyses were further restricted to pretreatment samples, as post-NACT specimens were not sufficiently available to assess the impact of chemotherapy on terminally exhausted T cells. Finally, a proportion of patients with HRD tumors did not receive PARPi maintenance therapy due to the retrospective determination of HRD status.

Our findings underscore the critical importance of precisely characterizing T-cell subpopulations that drive immune processes in the tumor microenvironment. Indeed, here we showed that beyond their value as biomarkers of response to immunotherapy, exhausted T-cell subpopulations emerge as prognostic biomarkers. This aligns with previous findings showing an association between tissue-resident memory T cells, which include tumor antigen-specific T cells, and overall survival in HGSC patients^19,60. This is also in line with the expression of tissue residency markers by terminally exhausted CD8 T cells infiltrating HGSC^1,41. The prognostic significance of exhausted T cells revealed here, alongside that of tissue-resident memory T cells^19,60, ascertains the physiological relevance of adaptive immunity in ovarian cancer and advocates for the development of combination immunotherapies that can either harness and amplify these spontaneous immune responses or provide them de novo through vaccination or adoptive T-cell transfer. In addition, our study highlights a broader concept: that terminally exhausted T cells, often considered dysfunctional, may in fact serve as markers of ongoing tumor-specific immune surveillance. This raises the possibility that similar prognostic value may apply to other immunogenic solid tumors, warranting investigation of exhaustion-related features across cancer types, even in the absence of immunotherapy.

Methods

Patients and specimens

Tumor samples were collected from 80 HGSC patients at the time of upfront diagnostic laparoscopic or debulking surgery for primary disease at the Institut universitaire du cancer de Toulouse – Oncopole (IUCT-O) between 2017 and 2022. Tumor specimens were obtained in the context of primary surgical management, at diagnosis, from treatment naive patients. Patient samples were collected in accordance with the Declaration of Helsinki, upon written informed consent and approval by the Comité de Protection des Personnes Sud Est III n° 2019-A01700-57 (DECIdE, NCT03958240) and the Comité de Protection des Personnes Sud Ouest et Outremer I n° DC-2016-2656. All patients had histologically documented tumors, were ≥ 18 years old at inclusion and were followed within a standard of care procedure. Exclusion criteria were: seropositivity for hepatitis B, hepatitis C, human immunodeficiency virus or hantavirus; any condition contraindicated with blood sampling procedure; pregnancy or breastfeeding; and suspected or documented active autoimmune disease or use of immunosuppressive drugs. The following parameters were collected at diagnosis: age, ECOG Performance status, mutation status, blood CA125 level, FIGO score, PCI, ascitic fluid volume.

For multiplex immunofluorescence (mIF), the most representative FFPE specimens were selected during the hematoxylin and eosin (H&E) review. Samples were handled by the Biopathological Support Platform for Clinical Studies (Oncopole Claudius Regaud (OCR), IUCT-O).

For flow cytometry analyses, tumor samples were rapidly transported to the research facility on ice. On arrival, samples were rinsed with Dulbecco’s Phosphate Buffered Saline (DPBS, Sigma-Aldrich, USA), subsequently minced on ice to smaller pieces, between 2 and 4 mm, and dissociated using gentleMACS Octo Dissociator (Miltenyi Biotec, Germany). Samples were then filtered using a 40 µm nylon mesh (BD Biosciences, USA). Tumor single-cell suspensions were cryopreserved in fetal bovine serum (FBS, Gibco, USA) containing 10% dimethyl sulfoxide (DMSO, Sigma-Aldrich, USA). Multiple samples issued from different lesions of a same patient have been analyzed.

H&E histological review

The representative H&E slides were scanned into whole-slide images using a 3DHITECH scanner at x20 magnification and used for histological review by two gynecological pathologists, including the assessment of histological type according to WHO 2020 classification (Vol.4⁶¹), SET architecture (solid, pseudo-endometrioid and transitional cell-like patterns), and necrosis⁶².

Assessment of HRD status

The presence of BRCA mutations was assessed by the Oncogenetics Laboratory as described in Al Saati et al. ⁶³. Briefly, genomic DNA was extracted from FFPE samples using the Maxwell FFPE plus Lev DNA purification kit (Promega, USA). DNA concentration was measured using QuBIT DS DNA BR Assay Kit (Thermo Fisher Scientific, USA). Capture was performed using a home-designed panel (Roche, Switzerland) including BRCA1 and BRCA2 genes and sequenced on an Illumina NextSeq500 sequencer (Illumina, USA) according to the manufacturer’s instructions. SNV detection was performed using HaplotypeCaller from GATK 3.3.0 and VarScan 2.3.7 software by the OCR bioinformatics platform.

Molecular HRD status was performed for patients without detectable BRCA mutations using the Myriad MyChoice^® CDx HRD test on genomic DNA extracted from FFPE specimens. If a pathogenic BRCA mutation was detected or the resulting HRD score was greater or equal to 42, the sample was deemed to be HRD.

mIF staining

mIF analyses were performed on FFPE tissue sections. The Discovery ULTRA (Ventana Medical Systems, USA) was used to automate the staining procedure. After dewaxing, the tissue slides were heat pre-treated using a CC1 (pH8) buffer (Roche, Switzerland). The slides were then stained for mIF using the RUO Discovery Universal procedure (v0.00.0370) in a 4-step protocol with sequential denaturation (CC2 buffer (pH6), at 100 °C, Roche, Switzerland) after each step. The fluorochrome sequence recommendations were respected⁶⁴.

Tissue slides were subsequently incubated using the primary monoclonal antibodies (mAbs) CD8, CD4, CD3 and FOXP3 as indicated (Table S3). Anti-CD8 mAb was diluted in Envision Flex diluent (Agilent Technologies, USA).

Targets were then linked using the OmniMap anti-rabbit and OmniMap anti-mouse HRP conjugated secondary antibodies (Table S4) (Roche, Switzerland). Visualization of the different targets was finally established using the Rhodamin6G, RED610 Cy5, and FAM detection kits (Roche, Switzerland).

For PD-L1 staining, slides were processed as previously indicated and stained with anti-PD-L1 (Table S3) mAb in Envision Flex diluent (Agilent Technologies, USA). Target was then linked using the OmniMap anti-rabbit HRP conjugated secondary antibody (Table S4). Visualization of the target was established using the Cy5 detection kit (Table S4).

For all experiments, the tissue slides were counterstained using Hoechst 33342 (Thermofisher, USA) in Discovery Diluent (Roche, Switzerland) and mounted with gelatin mounting medium (Sigma-Aldrich, USA).

Tumor annotation and digital image analysis

mIF stained slides were digitized in 16 bits using the Zeiss AxioScan.Z1 (Zeiss, Germany) whole-slide scanner equipped with a Colibri 7 (7-channels) solid-state light engine and appropriate custom high performance bandpass filter cubes (IDEX Health & Science, USA). Automated whole-slide image analysis was performed with the IndicaLabs Halo® platform (IndicaLabs, USA). Halo^® image analysis algorithms were run on a 56 CPU cores computing cluster (Dell R740XD equipped with 2x Intel Xeon Platinum 8280 L 2.7 G 28 C/56 T 10.4GT/s 38.5 M Cache Turbo HT / 512 GB RAM DDR4 / 30 To attached storage in RAID 5) and analyzed data were stored in the Halo^® MySQL database installed on a different server (Dell R640 equipped with 2x Intel Xeon Gold 5222 3.8 G 4 C/8 T 10.4GT/s 16.5 M Cache Turbo HT / 256 GB RAM DDR4).

Using the HighPlex FL module, the tumor microenvironment was studied in stromal and tumor areas annotated on each slide by a pathologist. Areas presenting artefacts such as tissue folds or degraded tissue fragments were excluded from the analysis. Nuclei were then segmented based on the Hoechst 33342 staining and cells cytoplasm and membranes were simulated by nuclei dilation. Representative images were extracted using ZEN Blue (Zeiss, Germany) and a brightness correction was applied to make the stainings visible.

CD8 T cells were defined as CD3⁺CD8⁺, CD4 T cells as CD3⁺CD4⁺CD8^-, conventional (CD4 Tconv) as CD3⁺CD4⁺CD8^-FOXP3^- and regulatory (CD4 Treg) as CD3⁺CD4⁺CD8^-FOXP3⁺. Each marker was analyzed by fluorescence intensity thresholding in the corresponding cell compartments.

Data were exported as CSV files and processed using R v4.3 to count each cell type in each sample, following the fluorescence intensity thresholds defined above. In parallel, analysis of infiltrated, excluded, and desert phenotypes was performed based on the CD8 infiltrate using R-generated images.

Cell purification and phenotypic assessment

CD3⁺, CD4⁺ or CD8⁺ cells were enriched by magnetic positive selection (CD3 MicroBeads, CD8 Microbeads or CD4 Microbeads, Miltenyi Biotec) from thawed tumor single-cell suspensions. Cells were stained to assess viability using BD Horizon Fixable Viability Stain 700 (BD Biosciences, USA) or Fixable Viability Dye eFluor 506 (eBioscience, USA), in DPBS for 20 min at 4 °C. Cells were then assessed phenotypically by surface staining for 15 minutes at 4 °C in DPBS containing 5% FBS with fluorochrome-labeled mAbs specific for CD3, CD4, CD8, CD45RA, CCR7, CD25, CD127, PD-1, TIM-3, CD39, as indicated (Table S5). Intranuclear staining was performed using the FOXP3/Transcription Factor Staining Buffer Set (eBioscience, USA). Briefly, cells were fixed and permeabilized for 45° minutes at 4 °C and stained with anti-FOXP3 mAb for 45 minutes at 4 °C. Cells were analyzed using BD LSRFortessa™ X20 (BD Biosciences, USA) flow cytometer and data were analyzed using DIVA software (BD Biosciences, USA) or FlowLogic (Inivai Technologies, Australia). For MFI quantification, only samples acquired with the same cytometer settings were analyzed.

Single cell RNA-Seq analyses

Raw count RNA-seq data from tumor-infiltrating CD8 T cells isolated from seven HGSC patients were obtained from the Gene Expression Omnibus (GSE184880)⁴¹ and analyzed using Seurat v5.4.0⁶⁵ following recommended guidelines. Briefly, the seven samples were merged, RNA counts were log-normalized, and the 2000 most variable genes were selected for principal component analysis (PCA). Data integration was performed using Harmony⁶⁶, and the first 50 principal components were used for Uniform Manifold Approximation and Projection (UMAP) dimensional reduction.

Cell types were annotated using CellTypist⁶⁷. Innate lymphoid cells (ILCs) and T cells were retained based on CellTypist annotations and quality-control criteria, including < 10% mitochondrial gene expression and > 10% ribosomal gene expression, and a new UMAP was generated as described above. From this ILC/T-cell subset, CD8 T cells were identified based on CellTypist annotations and expression of CD3 and CD8 genes, while cells expressing CD4 and/or FOXP3 were excluded. The final CD8 T-cell UMAP was generated using the same pipeline, and eight clusters were identified (resolution = 0.5) and annotated based on differential gene expression between clusters. Data representation was performed using R v4.5.2.

Statistical analyses

Patients’ characteristics were described using median and range (minimum-maximum) for quantitative variables and number and percentage for qualitative variables. The Kruskall-Wallis test was used to compare continuous variables in Table S1 and Chi-2 or exact Fisher test to compare categorical variables between groups.

Representation of flow cytometry and mIF data (Figs. 1 and 2 and Figures S1–S5) and statistical analyses of these data were performed using GraphPad Prism software (v 9.5.0, USA). Normality was assessed using the Shapiro-Wilk test. For normally distributed values, the two-tailed Student t-test was used to compare variables and Welch’s correction was applied to correct for variances, when needed. The Mann-Whitney test was used to compare variables in absence of normal distribution. Pearson correlation was used to compare variables in Figure S4.

Progression-Free Survival (PFS) was defined as the time between initial diagnosis and progression or death from any causes. Patients alive and without progression were censored at their last reported date. PFS rates were estimated using Kaplan-Meier and presented with their 95% Confidence Interval (95% CI). Univariable analyses were performed using the Log-rank test and univariable and multivariable analyses using Cox proportional hazards model. Variable selection for multivariable analyses was based on clinical relevance. For exhaustion parameters that were statistically significant in multivariable analyses (p < 0.05), patients were divided into 2 groups (high versus low) based on median parameter values. This provided a graphical representation of univariable PFS outcomes according to these exhaustion parameters.

To compare PFS according to maintenance treatment, a 9-month Landmark approach was used considering the immortal time bias introduced by the time between initial diagnosis and initiation of maintenance treatment, defining a maintenance treatment group with patients treated within 9 months after diagnosis, excluding patients who died, recurred or were lost to follow-up before 9 months.

For Figs. 3, 4 and Tables S1, S2, statistical analyses were conducted using STATA version 18 software (StataCorp LLC, USA).

Statistical tests were two-sided and p-values < 0.05 were considered significant.

Author contributions

M.A. and A.M. supervised the study. A.S., M.M., M.D., P.V., C.M.S., G.C.L., N.V.A., F.X.F., V.F., M.M., L.S., S.G., C.D., A.A.S., C.T., and G.B. performed experiments. A.S., B.S., N.T., B.C., and T.F. analyzed data. M.D., C.I., S.B., C.C., C.M.G., G.B., and J.P.D. provided materials. A.S., M.A., and A.M. wrote the manuscript.

Data availability

The biological data generated in this study are available from the corresponding authors on reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41698-026-01412-2.