Spatial Profiling of Ovarian Carcinoma and Tumor Microenvironment Evolution under Neoadjuvant Chemotherapy

High-grade epithelial ovarian carcinoma (EOC) represents the majority of ovarian cancers and is a leading cause of women’s mortality worldwide, accounting for more than 200,000 deaths annually (1). Around 70% of patients are diagnosed with advanced disease (FIGO III) or distant metastasis (FIGO IV).

Benefits from immuno-oncology (IO) approaches have been very limited in EOC. Studies to date evaluating agent PD1/PDL1 immune checkpoint inhibitors (ICI) have yielded disappointing results. ICIs aimed at restoring CD8⁺ T-cell function have reshaped therapeutic paradigms in several tumor types, including melanoma, kidney cancer, and non–small cell lung cancer (2). In contrast, single-agent PD1/PDL1 inhibitors resulted in disappointing response rates (<10%) in EOC (3, 4); however, the association of anti-PDL1/PD1 agents with chemotherapy in patients with advanced EOC demonstrated no significant benefit compared with chemotherapy alone (5, 6).

Enhancing benefit from immune strategies in EOC will likely require a better understanding of the cellular heterogeneity within the immune tumor microenvironment (iTME) and its evolution following treatment. Ovarian tumors were initially described as immunogenic, capable of stimulating an antitumor immune response. The presence of CD8⁺ tumor–infiltrated lymphocytes (TIL) has been detected in half of EOC at diagnosis and is associated with significantly longer overall survival (OS) in EOC (7, 8). However, cancer cells can often escape immune surveillance by expressing several inhibitory ligands that interact with receptors located in immune cells (IC) and/or tumor cells (TC). In fact, today, the iTME of EOC is increasingly described as highly heterogeneous and immunosuppressive (9). A better understanding of the iTME is required to improve patient outcomes by identifying other IO agents beyond PDL1/PD1 inhibitors that may be more relevant to EOC (10, 11). As EOC iTME contains ICs interacting with stromal and/or TCs both via coregulators, cytokines, and other enzymes, each of these actors could represent potential targets for future IO strategies. We previously reported that the majority of EOC tumors express multiple inhibitory molecules, with TIM3 nearly ubiquitous (12). In addition, we showed significant changes in the iTME composition before and after neoadjuvant chemotherapy (NACT). Importantly, the effects of platinum-based chemotherapy on iTME composition demonstrated significant interpatient variability, increasing cytotoxic T-cell influx in some, promoting suppressive IC infiltration in others (13–17). It seems crucial to take into account the iTME plasticity during treatment to propose individualized IO strategies.

To address this question, we performed spatial multiplexed immune profiling and cell clustering analyses to define the EOC iTME composition in EOC samples pre- and post-NACT from patients enrolled in the GINECO-sponsored CHIVA clinical trial (NCT01583322).

This translational research study was a post hoc exploratory analysis based on patient tumor samples from the CHIVA clinical trial. This research was conducted in accordance with the recommendations outlined in the Helsinki Declaration and approved by the medical ethics boards of all participating institutions. All the patients provided “written informed consent” to translational research substudies, and the protocol was authorized by the local medical ethical committee (March 2012) and approved by the Agence Nationale de Sécurité du Médicament (ANSM; June 2012).

The CHIVA trial recruited patients with inoperable FIGO IIIC/IV EOC who were candidates to receive NACT. Between January 2013 and May 2015, 188 patients were randomized to 3 cycles of carboplatin AUC5 and paclitaxel 175 mg/m² with an anti-VEGF tyrosine kinase inhibitor (nintedanib 200 mg twice a day) or placebo before debulking surgery if possible. This initial strategy was followed by postoperative chemotherapy with or without nintedanib. Per protocol, formalin-fixed, paraffin-embedded tumor samples from diagnostic laparoscopy, essentially peritoneal metastases, and from interval cytoreductive surgery if feasible were prospectively collected and centralized in an academic tumor bank.

Response rates post-NACT were assessed radiologically according to Response Evaluation Criteria in Solid Tumors 1.1 and categorized into two groups: objective response (OR⁺), when a partial/complete response was observed as the best response; and no objective response (OR⁻), when stability/progressive disease was observed as the best response. Complete cytoreduction (CC0; no macroscopically visible residual disease) at interval debulking surgery was also assessed. The trial was negative: the addition of nintedanib did not improve progression-free survival (PFS), CC0, or objective response rate, attributable to increased toxicity and lower chemotherapy exposure. Both OR⁺ and CC0 were independent factors associated with improved outcomes (18). However, no significant differences were detected in immune densities between samples obtained post-nintedanib versus postplacebo (Supplementary Fig. S1). Therefore, for the current study, samples were analyzed regardless of the treatment arm.

Samples were selected based on their viable tumor cellularity, excluding necrotic samples. In a few patients with a complete pathologic response, immune parameters post-NACT could not be evaluated. The most representative regions of each tumor were chosen by a pathologist on a hematoxylin and eosin–stained slide to construct several tissue microarrays (TMA), with three 1.2-mm cores from each tumor sample. Overall, 124 tumors were interpretable at diagnosis and 107 tumors after NACT and interval surgery. Among the NACT-treated patients, 41 received placebo and 66 received nintedanib. In addition, 89 pre/post-NACT-paired tumors were available.

For immunofluorescence (IF)- and IHC-based staining, we used the following Abs: CD3, CD4, CD8, Foxp3, CD20, CD68, CD163, DC-lamp, cytokeratin, and granzyme B (GrzB). Samples were stained sequentially with the validated Ab, generating multiplex staining panels (Supplementary Table S1).

We used the following Abs to generate IHC-monoplex panels: IDO1, TIM3, LAG3, PDL1, and HLA I. For each tumor sample, immune coregulator expression was scored by an expert pathologist, who reported the average of TC and IC with moderate/strong staining in three TMA cores. For each sample, we calculated the percentage of positive IC in the stroma (stromal+, s⁺) and the percentage of positive tumor, lymphocytes, and macrophage cells in the tumor area (intraepithelial+, ie⁺). A tumor sample was considered positive for an immune checkpoint marker if ≥1% of cells (ie⁺ and/or s⁺) stained positive. Overall, samples from 117 and 103 patients pre- and post-NACT, respectively, were available for the evaluation of at least one immune coregulator. For HLA I scoring, an H score (range: 0–300) was used to quantify the percentage and intensity of HLA I⁺ TC. The median H score was used to define low– (≤median) or high–HLA I expression (>median). The number of samples evaluable for each marker varied because of crushed artifacts or lost cores during TMA sectioning. Paired samples were not always available for each patient because of these technical reasons (Supplementary Table S2).

For IF and chromogenic IHC analysis, the slides were scanned at high resolution (20×) using a multispectral imaging system (Zeiss Axio Scan.Z1). Data from the multispectral camera were accessed using imaging Visiopharm software. The different IC populations were characterized and quantified in the intratumoral region using the cell segmentation and cell phenotype tool (Visiopharm). A mean score from the three cores was given for each population. CD8⁺, CD4⁺, NK (CD3⁺CD8⁻), Foxp3⁺, and CD20⁺ cells were expressed as the density of cells per mm². The tumor area was defined by cytokeratin⁺-staining, and the abundance of T and B cells was quantified by IF. Cytotoxic CD8⁺ T, CD4⁺ T_helper, regulatory Foxp3⁺ T (Foxp3⁺ Treg), and CD20⁺ B lymphocytes were evaluated in pre- and post-NACT tumors. However, the presence of GrzB⁺, CD68⁺ (M1), CD163⁺ (M2), and DC-lamp⁺ cells was expressed and quantified as the percentage of positive-stained surface in pre/post-NACT tumors.

The assessment of marked TILs was performed following the guidelines described by the International TILS Working Group 2014 (19). For the evaluation of the T-cell compartment within the high-grade serous ovarian cancer (HGSOC) tumors, two subpopulations were described: the stromal (s) sCD8⁺ and the intraepithelial (ie) ieCD8⁺ lymphocytes (in direct contact with TCs).

Pre- and post-NACT EOC samples were included in the patient clustering analyses. A Ward hierarchical clustering method was performed on CD4⁺, sCD8⁺, ieCD8⁺, NK (CD3⁺CD8⁻), Foxp3⁺, CD20⁺, GrzB⁺, CD68⁺, CD163⁺, and DC-lamp⁺ populations. We obtained 105 pre-NACT and 90 post-NACT assessable samples. A tumor sample was considered positive for IDO1, TIM3, LAG3, or PDL1 markers if ≥1% of cells (ie⁺ and/or s⁺) stained positive. Tumors were considered HLA I–high if the median H score >110 in pre-NACT and >125 in post-NACT cohorts. Using the values of the density of cells per mm² for CD4⁺, sCD8⁺, ieCD8⁺, NK, Foxp3⁺, and CD20⁺ cells and the percentage of positive-stained surface for GrzB⁺_, CD68⁺, CD163⁺, and DC-lamp⁺ cells, we calculated Euclidean distance after log₂(x + 1) transformation and Z score and median normalization for pre- and post-NACT cohorts, respectively. The normalized values were further scaled to [−1, 1] for each cell population and represented in blue to red colors. The uniform manifold approximation and projection (UMAP) dimension reduction algorithm was applied to visualize the different clusters obtained from the hierarchical clustering in a 2D space (20).

Statistical analysis was performed on the whole cohort and paired samples. A Fisher exact test was used to compare categorical variables. Descriptive statistics were used to summarize patient demographics and clinical characteristics. Continuous variables were compared with Mann–Whitney and Wilcoxon signed-rank tests. A P value of <0.05 was considered significant. OS and disease-free survival were estimated using Kaplan–Meier curves and presented with Rothman 95% confidence intervals (CI) at 5 and 10 years. HRs with 95% CIs were calculated using Cox models with no adjustment. Correlations between immune coregulators were tested using Spearman correlation analysis. All statistical analyses were performed using GraphPad Prism 9.

The data generated in this study are not publicly available to guarantee the confidentiality of clinical data from patients, but they are available upon reasonable request from the corresponding author.

One hundred and eighty-eight patients were recruited in the CHIVA trial. The median age was 64 years (range: 31–79), 78% of patients had FIGO IIIC diseases, and 89% of patients had a performance status of 0 or 1. Within the trial, formalin-fixed, paraffin-embedded samples were prospectively collected from 155 patients to construct the TMAs at baseline and after interval surgery. Overall, 124 tumors were interpretable at diagnosis and 107 tumors after interval surgery. To characterize the iTME at baseline and its evolution after NACT exposure, we performed spatial multiplexed immune profiling and cell clustering analyses in ovarian cancer samples from patients enrolled in the clinical trial before and after NACT exposure (Fig. 1).

All tumors were centrally re-reviewed by a pathologist (C. Genestie): 97% were high-grade tumors, and 90% were serous/papillary tumors. Tumor BRCA mutational status was known for 105 patients, and 20 harbored a deleterious BRCA1/2 mutation. The median follow-up of the patients was 42 months, and their median PFS and OS were 14.2 (95% CI, 13–15.7) and 38.9 months (95% CI, 34.2–47.2), respectively. Overall, 43% were considered OR⁺ (with partial or complete response to NACT), and 47% benefited from complete cytoreductive surgery after three cycles (CC0; Supplementary Table S3).

In our study, ovarian cancer samples were collected from 124 patients at baseline and 107 after NACT, including 86 pre/post-NACT patients. Median CD4⁺ and CD8⁺ cells increased significantly post-NACT (118 vs. 172 cells/mm²; P = 0.03; and 117 vs. 199 cells/mm²; P = 0.002), respectively. NACT exposure was associated with a significant decrease in Foxp3⁺ Treg cells (32 vs. 17 cells/mm²; P = 0.008), resulting in an increased ratio of effector to suppressor T cells (CD8/Foxp3 median: 4.4–9.6; P = 0.002), in favor of increased surveillance (Fig. 2A–C). GrzB expression, a serine-protease released by CD8⁺ citotoxic (T_citox) and NK cells, did not change after NACT (median: 0.7%–0.8%; P = 0.09). With regard to the localization of sCD8⁺ and ieCD8⁺ T cells, only the sCD8⁺ subset increased significantly with NACT (P = 0.04), whereas the ieCD8⁺ population remained unchanged (P = 0.7; Supplementary Fig. S2). When considering changes in CD8⁺ cell density (s + ie), 64% of patients exhibited increased CD8⁺ cells post-NACT (Fig. 2D). Overall, survival was significantly improved among patients with increasing CD8⁺ cells compared with the CD8⁺ T–stable/decreasing group (median survival: 47.3 vs. 33.1 months; P = 0.04 log-rank test; Fig. 2E). In contrast, infiltration by Foxp3⁺ Tregs did not correlate with PFS or OS, whether at baseline or post-NACT. Among the 105 tumors with known BRCA status, 20 were BRCA1/2m. However, there were no significant differences in immune features between BRCAwt and BRCAm.

We next assessed CD20⁺ B-cell recruitment upon NACT. B-cell analysis in 89 pre/post-NACT patients showed a significant increase in CD20⁺ cells post-NACT (median: 3.7–9.1 cells/mm²; P = 0.02; Fig. 2A). Using a median cutoff of 4 cells/mm² to detect B cell–infiltrated tumors (21), we showed a trend toward improved OS in CD20⁺-infiltrated tumors pre-NACT (median survival: 37.7 vs. 47.1 months; low vs. high B-cell infiltrated; P = 0.2 log-rank test). CD20⁺ B cell clusters were only rarely detected in 14% (18/126) of pre-NACT samples and 23% (25/109) of post-NACT samples.

We next evaluated the infiltration density of tumor-associated macrophages (TAM) and dendritic cells (DC) by using CD68 (M1), CD163 (M2), and DC-lamp markers. M1, M2, and DC populations were analyzed in 82 pre/post-NACT patients. There was no significant change in CD68⁺ M1 densities (median: 2%–2%; P = 0.4). Both CD163⁺ M2 macrophages (P = 0.01) and DC-lamp⁺ cells (P = 0.004) infiltration increased significantly post-NACT, with no significant change in the M1/M2 ratio post-NACT (Fig. 3; Supplementary Table S4). However, there was no correlation between macrophage or DC cell infiltration and prognosis.

Using an H score, HLA I expression was detected in 116 and 100 pre- and post-NACT samples, respectively. Few tumors exhibited complete HLA I loss, with only 9.5% (11/116) of tumors showing an H score = 0. Using the pre-NACT median as a cutoff, we classified tumors as low–HLA I when they presented an H score ≤ 110 (Fig. 4A). At diagnosis, CD8⁺ cell recruitment was significantly higher in HLA I–high tumors (median: 189 vs. 93 cells/mm²; P = 0.007). Similar results were obtained in post-NACT samples, and HLA I expression did not change under NACT. With regard to prognosis, HLA I–high tumors were associated with better OS at both diagnosis (44 vs. 33 months; P = 0.04; log-rank test) and post-NACT (47 vs. 33 months; P = 0.04; log-rank test; Fig. 4B).

In an effort to classify ovarian cancer tumors by integrating all immune features, we performed clustering analysis on pre- and post-NACT cohorts. Based on 10 IC parameters, pre-NACT samples were segregated into four spatially distinct clusters: high B_inf, high T_inf, low T_inf, and desert clusters. The largest cluster of patients was low T_inf (40%), followed by high B_inf (31%), desert (16%), and high T_inf (13%). The high B_inf group was highly enriched in B clusters, CD20⁺, sCD8⁺, CD4⁺, and Foxp3⁺ cells. The high T_inf group was mainly driven by ieCD8⁺ cells, with moderate levels of sCD8⁺, CD4⁺, and Foxp3⁺ cells and low densities of CD20⁺. The low T_inf group presented low densities of all ICs, and the desert group was almost depleted of ICs (Fig. 5A; Supplementary Fig. S3).

Upon NACT exposure, our cohort was segregated into three clusters: high B_infT_inf, low T_inf, and desert. The predominant subgroup, the high B_infT_inf (49%), was markedly segregated from the low T_inf (33%) and desert groups (18%). The high B_infT_inf subset presented an inflamed iTME, highly enriched in sCD8⁺ cells but also with high densities of B clusters, CD20⁺, NK, and DC-lamp⁺ cells. The low T_inf group presented low densities of ieCD8⁺, sCD8⁺, CD20⁺, NK, and DC-lamp⁺ cells, whereas the desert group was very poorly immune-infiltrated (Fig. 5B and C; Supplementary Fig. S3). Interestingly, PFS was significantly improved in the high B_infT_inf cluster compared with low T_inf and desert groups (16.0 vs. 13.3 months; P = 0.04; log-rank test), as was OS (47.2 vs. 33.4 months; P = 0.02, log-rank test; Fig. 5D and E). Our results suggest that although ieCD8⁺ T cells alone might be prognostic pre-NACT, the inflamed high B_infT_inf cluster post-NACT predicted a better outcome.

Given the association between patients with BRCAm tumors and IC infiltration (22, 23), we evaluated the BRCA status within the clusters. Pre-NACT, BRCAm tumors were enriched in the high T_inf pre-NACT cluster (31%), compared with high B_inf (9.4%), low T_inf (9.5%) or desert (11%) subgroups. Post-NACT, the high B_infT_inf cluster was enriched for BRCAm tumors (30%) compared with the low T_inf (10%) or desert (6.3%) group. Thus, our results also revealed an enrichment of both HLA I and several immune coregulators in high B_inf and high T_inf pre-NACT and in high B_infT_inf post-NACT clusters, compared with low T_inf and desert pre/post-NACT clusters.

Additionally, cluster analysis was conducted on a limited number of paired pre/post-NACT samples (n = 61 × 2) using the 10 IC features. When analyzing unpaired and paired pre- and post-NACT cohorts together, no differences between the pre-NACT samples were detected. The paired post-NACT high B_infT_inf subset was still the largest cluster, although the proportion was lower (38% vs. 49% paired vs. unpaired).

Given the limited benefit of targeting the immune coregulator PDL1 in ovarian cancer (5, 24), we aimed to study the relevance of other coinhibitory molecules, including IDO1, TIM3, and LAG3, and their dynamics under NACT. As we have previously observed (12) at diagnosis, TIM3 was the most abundant coregulator (91%), followed by LAG3 (56%) and IDO1 (50%). In contrast, only 36% of tumors were PDL1⁺ (Supplementary Table S5). Among samples with available data for all biomarkers in the pre-NACT cohort (n = 93), 72% of the patients expressed two or more immune coregulators (Fig. 6A). At diagnosis, coregulators were only weakly positively correlated to each other, suggesting heterogeneity in immune-tolerance regulatory pathways among ovarian cancer tumors (Supplementary Table S6). Our results also revealed an enrichment of IDO1, TIM3, LAG3, and PDL1 immune coregulators in high B_inf and high T_inf pre-NACT clusters compared with low T_inf and desert pre-NACT clusters (Supplementary Table S7).

Upon NACT exposure, LAG3 expression increased significantly on both stromal ICs (s⁺; P < 0.0001) and ie compartments (ie⁺; P < 0.0001), whereas TIM3 and PDL1 expression decreased significantly on s⁺ post-NACT (P < 0.0001 and P = 0.03, respectively; Fig. 6B and C). Similarly, an enrichment of IDO1, TIM3, LAG3, and PDL1 immune coregulators was found in the high B_infT_inf post-NACT cluster compared with low T_inf and desert post-NACT clusters (Supplementary Table S7).

Targeting PDL1/PD1 alone in ovarian cancer has failed to demonstrate any benefit in patients with ovarian cancer. Instead, harnessing the antitumor immune response in ovarian cancer may require targeting other immune regulators, possibly in combination, and being adapted to the unique features of the iTME in individual patients. In the present study, we evaluated the iTME at baseline and its evolution under NACT. Previous reports have shown that high TIL infiltration inhibits tumor growth and is associated with improved prognosis in melanoma, colorectal cancer, and ovarian cancer (7, 19, 25–29). We contribute to the body of evidence that high ieCD8⁺ T cells are associated with a favorable prognosis at diagnosis (28). In addition, we demonstrated that NACT significantly increased the balance of immune effector to suppressor T cells in favor of an antitumor response. NACT was associated with a significant increase in CD4⁺ T_helper and CD8⁺ T_citox infiltration, together with a significant decrease in Foxp3⁺ Tregs (P = 0.008). Favorable CD8⁺/Foxp3⁺ (T_citox/Tregs) balance is associated with favorable prognosis in cancer outcome (28–30). As we already reported (16), NACT exposure also increased the proportion of tumors showing favorable ratios of T_citox/T_regs post-NACT, in favor of enhanced immune surveillance and antitumor immunity. However, somewhat surprisingly, post-NACT, neither absolute ieCD8⁺ TILs nor a favorable CD8⁺/Foxp3⁺ (T_citox/T_regs) ratio was associated with improved survival.

CD8⁺ cell dynamics during NACT was heterogeneous among tumors but correlated with survival. Patients with an increased influx of CD8⁺ cells in paired samples had significantly better OS compared with the CD8⁺ T–decreasing/stable groups, suggesting that improved outcomes in certain ovarian cancer may be linked to the immune-stimulatory properties of chemotherapy. The fact that the CD8⁺ influx post-NACT is not uniform may simply reflect the intertumor heterogeneity in HGSOC.

In the current study, we also aimed to characterize the iTME beyond T cells, integrating information on B cells, TAMs, and DCs. The recruitment of B cells and the presence of B cell–enriched tertiary lymphoid structures have been associated with antitumor immunity and improved survival in sarcoma, melanoma, and other cancers (31–33), whereas the presence of B cells in breast and ovarian cancers predicts a beneficial response to therapy (34). Despite the low density of CD20⁺ B cells detected pre-NACT, their levels increased significantly in paired samples during NACT exposure. B-cell clusters were rare both at diagnosis and post-NACT and did not correlate with better clinical outcomes.

Similarly, M1 and M2 TAM and DC populations are frequently observed in ovarian tumors. Although macrophage infiltration has been associated with poor prognosis and a high tumor grade, distinct populations of macrophages with opposite pro- and antitumorigenic roles coexist within the same tumor (35). M2 macrophage infiltration is related to cancer development, tumor tolerance, growth, highly suppressive iTME, and poor clinical outcomes. Conversely, M1 macrophages, a high ratio of M1/M2, and increased DCs are related to an antitumoral response (36). Targeting TAMs arises as a good anticancer therapy through either their ablation or their re-differentiation from the M2-protumoral toward the M1-antitumoral state (37). In the current study, no changes in the CD68⁺ M1 population or M1/M2 ratio were observed. Instead, a significant increase in CD163⁺ M2 recruitment post-NAC is detected, which could in theory contribute to immune tolerance in a subset of tumors. However, no association between macrophage infiltration and survival was found. Finally, NACT significantly increased DC-lamp⁺ cells in favor of an antitumor response. However, these represent the smallest subset of IC in the TME and were detected in <0.05 of the tumor area (100-fold less than TAMs); therefore, although differences in the DC-lamp⁺ population were statistically significant post-NACT, it is unclear whether they are biologically relevant.

The clustering analysis of 10 IC populations collected from 105 pre-NACT and 90 post-NACT patients helped us provide an insight into iTME in EOC at diagnosis and upon NACT exposure. Pre-NACT samples were segregated into four IC patterns: high B_inf, high T_inf, low T_inf, and desert clusters. The largest cluster was the low T_inf group, composed of 41% of the analyzed samples, followed by the high B_inf (31%), desert (16%), and high T_inf groups (12%). The high T_inf, a more homogeneous cluster, was mainly infiltrated with ieCD8⁺ cells. However, these clusters provided little prognostic information.

In contrast, upon NACT, we identified three subsets: the high B_infT_inf, low T_inf, and desert clusters. The largest cluster was the high B_infT_inf, representing ∼50% of the samples. This subset presented a more inflamed iTME than the high T_inf pre-NACT cluster, and it was highly infiltrated not only by sCD8⁺ cells but also with high densities of CD20⁺ B, NK, or DC-lamp⁺ cells. Of particular interest was the enrichment in NK cells, as this population did not seem to be relevant pre-NACT. Although prior studies showed a correlation between CD8⁺ cell infiltration and patient survival post-NACT (15, 29), few studies have reported the association between B cells and prognosis post-NACT in ovarian cancer. We found that following NACT, a significant improvement in PFS and OS was found in the high B_infT_inf cluster. Our results suggest that although ieCD8⁺ infiltration alone may be relevant to OS at diagnosis, an evaluation of sCD8⁺ cells in combination with B, NK, and DC cells may provide better prognostic information post-NACT. Interestingly, the myeloid compartment did not seem to contribute to tumor clustering at diagnosis or after NACT.

In addition, we found an enrichment in BRCA1m tumors in both high T_inf pre-NACT and high B_infT_inf post-NACT clusters. Our study and others have shown that highly CD8⁺-infiltrated tumors, harboring BRCAm, or epigenetic loss tumors are associated with a favorable prognosis (22, 23). In this report, BRCAm-enriched clusters were also related to a better outcome.

The presence of HLA class I molecules is essential for immunosurveillance and ICI efficacy, and the loss of HLA I expression in tumors is an escape mechanism from immunotherapy (38). In our cohort, less than 10% of patients showed a complete loss of HLA I expression, in line with previous studies (39, 40). Our study showed that highly HLA I–expressing tumors (H score > 110) had significantly higher CD8⁺ T-cell infiltration and OS, and they were also enriched in high T_inf and high B_infT_inf pre-NACT and high B_infT_inf post-NACT clusters.

ICIs provide a wide opportunity to identify new targets to improve the modulation of ovarian cancer iTME. Targeting different immune coregulators could more effectively harness antitumor immunity in this anti-PDL1 poorly sensitive malignancy. Here, 93 patients were scored for PDL1, TIM3, LAG3, and IDO1. As we have previously reported (12), TIM3 was by far the most prevalent immune coregulator with 91% of tumors being TIM3⁺, whereas PDL1 was the rarest expressed in only 36% (12). Both TIM3 and IDO1 expressions were prevalent in more than 50% of tumors. Importantly, the vast majority of tumors coexpressed two or more coregulators. Our data also showed that NACT exposure significantly increased the expression of LAG3 in both intraepithelial and stromal IC but induced a decrease in both TIM3 and PDL1 expressions, further supporting the crucial relevance of immune profiling ovarian cancer post-NACT and highlighting the potential for targeting other immune checkpoints beyond PD1/PDL1. Our study also revealed that the highest expression of immune coregulators was detected in high T_inf pre-NACT and high B_infT_inf post-NACT clusters. Desert clusters pre- and post-NACT also presented the lowest immune coregulator expression. The identification of these patients would help search for immunotherapeutic approaches.

In summary, the current study has provided insights into the iTME and its evolution during NACT treatment. The impact of NACT on the iTME is heterogeneous, and not surprisingly, those tumors exhibiting an influx of CD8⁺ cytotoxic cells in paired samples during NACT have improved outcomes. CD20⁺ and NK cells increased under NACT, and immune clustering post-NACT identified a spatially distinct high B_infT_inf subset with a more diverse antitumor immune composition and improved outcomes. In ovarian cancer, PDL1 may not be the most relevant target. Other actionable checkpoints, such as TIM3, LAG3, and IDO1, are much more prevalent yet remain poorly correlated with each other. This suggests heterogeneity in immune-tolerance pathways among HGSOC and the crucial relevance of personalizing IO strategies in ovarian cancer.

Q. Zeng reports other support from the China Scholarship Council outside the submitted work. E. Pujade-Lauraine reports personal fees and nonfinancial support from AstraZeneca and Roche, personal fees from Agenus and Incyte, and other support from ARCAGY outside the submitted work. F. Joly reports personal fees and other support from GSK, Eisai, MSD, and AstraZeneca outside the submitted work. N. Dohollou reports grants from AstraZeneca, Bristol Myers Squibb, Boehringer Ingelheim, Genomic Health, MSD, Novartis, and Pfizer; personal fees and nonfinancial support from Daiichi; grants, personal fees, and nonfinancial support from Eli Lilly and Company, Roche, and Seagen; and nonfinancial support from Gilead outside the submitted work. M. Fabbro reports personal fees from GSK outside the submitted work. D. Bello Roufai reports personal fees from AstraZeneca and MSD outside the submitted work. J. Gantzer reports personal fees from Deciphera, grants from INTERSARC, Alsace contre le cancer, ARD Fondation, and Canceropole Est, and nonfinancial support from PharmaMar outside the submitted work. E. Rouleau reports grants and personal fees from AstraZeneca and GSK and personal fees from Clovis and MSD during the conduct of the study. A. Leary reports grants from ARCAGY Research during the conduct of the study, as well as grants and personal fees from AstraZeneca, Clovis, and GSK and grants from MSD, Ability, Zentalis, Agenus, Iovance, Sanofi, Roche, Ose Immunotherapy, and Bristol Myers Squibb outside the submitted work. No disclosures were reported by the other authors.

E. Yaniz-Galende: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Q. Zeng: Software, formal analysis, investigation, methodology. J.F. Bejar-Grau: Software, formal analysis, investigation, methodology. C. Klein: Software, validation, methodology, writing–review and editing. F. Blanc-Durand: Formal analysis, validation, investigation, writing–review and editing. A. Le Formal: Validation, methodology. E. Pujade-Lauraine: Investigation, writing–review and editing. L. Chardin: Validation. E. Edmond: Methodology. V. Marty: Methodology. I. Ray-Coquard: Investigation. F. Joly: Investigation. G. Ferron: Investigation. P. Pautier: Investigation. D. Berton-Rigaud: Investigation. A. Lortholary: Investigation. N. Dohollou: Investigation. C. Desauw: Investigation. M. Fabbro: Investigation. E. Malaurie: Investigation. N. Bonichon-Lamichhane: Investigation. D. Bello Roufai: Investigation. J. Gantzer: Investigation. E. Rouleau: Formal analysis, validation. C. Genestie: Data curation, formal analysis, supervision, validation, investigation, methodology. A. Leary: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, project administration, writing–review and editing.

This work was supported by donations from patients and families through the “Parrainage Cancers Gynécologiques” Program and by the Association de Recherche Cancers Gynécologiques Research.