Profiling tumor immune microenvironment of epithelial ovarian carcinoma

Epithelial ovarian carcinoma (EOC) is the most lethal gynecological malignancy worldwide. It comprises five major histological subtypes, high-grade serous (HGSOC), low-grade serous (LGSOC), endometrioid (ENOC), clear cell (CCOC), and mucinous (MOC), each characterized by distinct origins, molecular alterations, and clinical behavior. 1-3 HGSOC, representing 70% of EOCs, is likely to originate from the secretory epithelium of the fimbrial end of the Fallopian tube. 4 It is characterized by nearly universal presence of TP53 mutations, extensive genomic instability, homologous recombination deficiency in approximately 50% of tumors and high initial sensitivity to platinum-based chemotherapy. 5 By contrast, LGSOC evolves from serous borderline tumors, is characterized by KRAS or BRAF mutations and relative genomic stability, displays a more indolent course, and shows limited responsiveness to conventional platinum-based regimens. 2 ENOC and CCOC are strongly associated with endometriosis and frequently harbor mutations in ARID1A and PIK3CA. 6 These subtypes are typically diagnosed at an earlier stage than HGSOC but, particularly for CCOC, show reduced sensitivity to platinum therapy and poorer outcomes in advanced disease. 7 MOC represents the rarest major epithelial subtype. It is generally unilateral and large, usually diagnosed at early stage, and exhibits activation of the RAS/MAPK pathway (through KRAS or BRAF mutations) and ERBB2 amplification. 2 , 8 Like LGSOC, they mostly progress from borderline mucinous tumors. MOC is notably less responsive to standard platinum-based chemotherapy. 7 Collectively, these observations illustrate the biological heterogeneity of EOC and highlight the necessity of subtype-specific molecular and therapeutic approaches. Numerous studies have extensively characterized the phenotype, functional alterations, and prognostic significance of immune infiltrates in EOC. For instance, the presence of intraepithelial CD3 + T cells, 9 particularly CD8 + T cells, 10 or a high CD8 + /FoxP3 + regulatory T-cell ratio has been associated with prolonged survival. 11 Conversely, the accumulation of myeloid cells such as dendritic cells and tumor-associated macrophages is generally considered tumor-promoting and immunosuppressive, 12 and correlates with poor outcomes. 13-15 However, most of these studies have assessed the prognostic value of immune infiltrates in EOC. Few studies compared immune infiltration patterns across the 5 major histotypes. 10 , 16 This represents an important gap, as each histotype constitutes a distinct disease case with specific pathogenesis, genomic alterations, clinical presentation, and therapeutic responses. 7 Recently, several studies have employed multiplexed immunohistochemistry (IHC) to provide a more detailed characterization of the tumor immune microenvironment. However, they typically focused on a single histotype. 12 , 13 , 17-19 We hypothesize that EOC subtypes harbor distinct, spatially organized immune microenvironments. Herein, we performed sequential multiplexed imaging of seven markers, CD45, CD8, CD68, CD163, FoxP3, CD20 and cytokeratin on a tissue-microarray that included 139 clinically annotated EOCs encompassing the major 5 histotypes (HGSOC, LGSOC, CCOC, ENOC, and MOC). In addition, we analyzed pre-invasive tumors, i.e. borderline tumors (serous and mucinous). We found spatially divergent immune landscapes among the 5 EOC histotypes, with distinct abundances and spatial distributions of immune cell subpopulations within tumor compartments. Our study provides an extensive characterization of the tumor immune microenvironment across EOC subtypes, and insights into immune alterations associated with the transition from borderline to invasive carcinoma.

We performed 7-plex IHC (CD45, CD8, CD68, CD163, FoxP3, CD20 and cytokeratin) on a tissue microarray (TMA) constructed from 195 tumor samples collected from EOC and borderline tumors, all derived from patients who underwent upfront surgical treatment at the University Hospital Zurich ( Figures 1A and B , S1A and 1B). From the original cohort, 139 carcinoma cases and 26 borderline tumors were retained for further analysis. The final number reflects the exclusion of TMA cores with inadequate staining, imaging artifacts, core drop-out or other types of EOC outside the scope of the study. The subtype distribution comprised 62 HGSOC, 8 LGSOC, 26 CCOC, 14 MOC, 29 ENOC, as well as 15 serous borderline tumors and 11 mucinous borderline tumors. Clinico-pathological characteristics of the final cohort are summarized in Table 1 . Clinical and pathology characteristics of the study cohort. The cohort included 62 high-grade serous ovarian carcinomas (HGSOC), 8 low-grade serous ovarian carcinomas (LGSOC), 26 clear cell ovarian carcinomas (CCOC), 14 mucinous ovarian carcinomas (MOC), and 29 endometrioid ovarian carcinomas (ENOC), as well as 15 serous and 11 mucinous borderline tumors. The table summarizes the number of cases per subtype ( n ), median age at diagnosis (range), FIGO stage (I, II, III, IV), WHO 2020 tumor grade (1, 2, 3), Residual disease (R0, R1/2) and Platinum status (PR: platinum resistant, PS: platinum sensitive) where applicable. Histotype-specific immune infiltration in epithelial ovarian cancer. (A) Workflow of the multiplex immunohistochemistry (IHC) and image analysis pipeline. Tissue microarrays (TMAs) were constructed from representative tumor areas and stained using a seven-plex multiplex IHC panel. Slides were digitally scanned, and marker-specific images were individually analyzed in QuPath using artificial intelligence–based algorithms. Following marker-based analysis, the multiplex stainings were computationally aligned, merged in MATLAB, and visualized as pseudo-colored composite images. (B) Representative seven-plex multiplex IHC images of TMAs from 165 epithelial ovarian and borderline tumors stained for CD45, CD8, FoxP3, CD20, CD68, CD163, and cytokeratin. A key advantage of this multiplexed approach is its ability to generate a detailed and quantitative assessment of immune cell composition and spatial distribution across tumor compartments (intraepithelial vs stromal) and histotypes within the same core/section. To enable such analyses in our cohort, we implemented a pipeline for positive cell detection, integrated spatial mapping and statistical analysis ( Figure 1A ). This analytical workflow helped us to generate a multiparametric spatial profile of immune cell populations across EOC histotypes. To profile the immune cell composition among the 5 histotypes EOC (HGSOC, LGSOC, CCOC, MOC, and ENOC) on a common scale, we first assessed the abundance of each immune cell subset (CD45, CD8, CD68, CD163, FoxP3, and CD20) estimated as the percentage of positive cells over total cells ( Figure 2A ). Across all subtypes, macrophages constituted the dominant immune population, while adaptive immune cells such CD8 + T cells were moderate, FoxP3 + regulatory T cells (Tregs) and CD20 + B cells remained sparse. Immune infiltration differed markedly between histotypes. HGSOC and CCOC exhibited the highest levels of immune infiltration, driven predominantly by macrophages (CD68 + and CD163 + ), with moderate CD8 + T-cell presence. Marker-specific immune profiling and immune cell ratios across ovarian cancer histotypes. (A) Quantification of immune cell subsets across ovarian cancer histotypes showing marked inter-histotype differences. (B) Comparison of individual immune markers across HGSOC, LGSOC, CCOC, MOC, and ENOC, using HGSOC as reference. (C) Comparison of immune cell ratios revealed a significantly higher CD8 + /FoxP3 + ratio and lower CD163 + /CD68 + ratio in MOC, while LGSOC showed a reduced CD8 + /FoxP3 + ratio. Group comparisons were performed using two-sided Wilcoxon rank-sum tests and reported p -values were adjusted for multiple comparisons using the false discovery rate (FDR) method; Statistical significance is indicated as follows: p  < 0.05 (*), p  < 0.01 (**), p  < 0.001 (***), p  < 0.0001 (****). Box: Q1–Q3, line: median, whiskers: ±1.5 × IQR; outliers not shown. We next performed a marker-specific analysis of immune infiltration across EOC histotypes (HGSOC, LGSOC, CCOC, MOC, and ENOC), using HGSOC as the reference group ( Figure 2B ). While Panel A provides a global overview of immune cell composition across histotypes, this marker-stratified approach enables a more focused, cell-type comparison and facilitates statistical testing of differences for each immune subset. This allows clearer identification of histotype-specific immune deviations. CCOC had the highest infiltration by CD45 + leukocytes ( p  < 0.05), indicating elevated immune cell enrichment, whereas leukocytes levels in LGSOC, MOC, and ENOC were comparable to HGSOC. The enrichment of CCOC in leukocytes was mainly driven by CD68 + macrophages ( p  < 0.01) and M2-like CD163 + macrophages ( p  < 0.05). CD8 + cytotoxic T cells remained relatively low-to-moderate across all histotypes with high variability across samples, except for LGSOC which consistently demonstrated lower infiltration. Another specificity of LGSOC was significant enrichment in FoxP3 + regulatory T cells ( p  < 0.05), and significant reduced abundance of CD20 + B cells ( p  < 0.05), which were rare across all subtypes. Overall, we found that CCOC is macrophage- and M2-skewed, LGSOC shows Treg enrichment with fewer B cells, ENOC closely resembles HGSOC, while MOC displays a tendency toward elevated B-cell levels, suggesting a distinct microenvironment. Analysis of immune cell ratios further highlighted subtype-specificities ( Figure 2C ). In MOC, the CD8 + /FoxP3 + ratio was significantly higher and the CD163 + /CD68 + ratio lower than in HGSOC, suggesting a more immune-active and less immunosuppressive microenvironment. Conversely, LGSOC exhibited a lower CD8 + /FoxP3 + ratio with unchanged macrophage polarization. No significant differences were observed for CCOC or ENOC. These results demonstrate that immune cell composition and proportions vary markedly among EOC histotypes and underscore the complexity of the tumor immune microenvironment. The prognostic value of immune infiltrates in EOC varies according to their location in tumor compartments (tumor core versus stroma). 9 , 10 , 17 , 18 We investigated the abundance of immune cells per compartment. Across the 5 EOC histotypes, immune cell infiltration was consistently higher in the stromal compartment than in the tumor core, consistent with previous findings. 13 , 17 For nearly all markers assessed, the stroma showed significantly higher abundance compared with the tumor nest, indicating immune excluded pattern ( Figure 3A ). The stromal enrichment in leukocytes was mainly driven by CD68 + macrophages ( p  < 10 −3 ) and M2-like CD163 + macrophages ( p  < 10 −3 ). Although the absolute proportion varied among histotypes, the overall pattern of stromal enrichment was uniform, highlighting a compartment-specific distribution in which immune cells preferentially accumulate in the surrounding stroma rather than infiltrating tumor nests. Compartment-specific immune architecture across epithelial ovarian cancer histotypes. (A) Comparison of immune cell densities in stromal compartment versus tumor epithelium as defined by cytokeratin staining. (B) Compartment-specific analysis of CD8 + /FoxP3 + and CD163 + /CD68 + ratios per histotype. Group comparisons were performed using two-sided Wilcoxon rank-sum tests and reported p -values were adjusted for multiple comparisons using the false discovery rate (FDR) method; Statistical significance is indicated as follows: p  < 0.05 (*), p  < 0.01 (**), p  < 0.001 (***), p  < 0.0001 (****). Box: Q1–Q3, line: median, whiskers: ±1.5 × IQR; outliers not shown. Given this consistent compartmental enrichment, we then investigated whether the relative proportions of individual immune cell populations differed among histotypes within each compartment (Figure S2). The stratified analysis revealed that stromal immune composition showed substantial variability, with certain markers exhibiting pronounced differences among tumor histotypes. CD45 + cells were more abundant in CCOC than in other histotypes, likely driven by the elevated macrophage infiltration (CD68 + and CD163 + cells), whereas CD8 + cell levels remained relatively comparable across groups. Although FoxP3 + regulatory T cells were generally low, we noted a notable lower infiltration in the stroma of MOC. We did not observe any significant differences in the abundance of CD20 + B cells, which remained low in the stromal compartment across all histotypes. In contrast, tumor nests exhibited a more heterogeneous immunologic landscape, with overall lower immune infiltration but clear histotype-specific deviations. Consistent with the stromal observations, CD68 + macrophages were again elevated in CCOC tumors accompanied by an upward, though not significant, trend in M2-like macrophages (CD163 + ), which in turn were notably reduced in MOC. Interestingly, LGOSC exhibited a relative increase in FoxP3 + cells compared to other histotypes. Lastly, CD20 + B-cell infiltration remained again minimal across all tumor nests, with significantly lower levels detected in LGSOC and CCOC. To further characterize the balance of infiltrated immune cells, we calculated the ratios of cytotoxic to regulatory T cells (CD8 + /FoxP3 + ) and M2-like to total macrophages (CD163 + /CD68 + ) across stromal and tumor compartments of each EOC histotype ( Figure 3B ). Marked compartment-specific differences were particularly evident in HGSOC, where tumor nests exhibited a significantly higher CD8 + /FoxP3 + ratio than the stroma, suggesting a relative enrichment of cytotoxic T cells within the epithelial compartment, whereas the stroma showed a significantly higher CD163 + /CD68 + ratio, indicative of a more pronounced immunosuppressive macrophage polarization in the surrounding microenvironment. In contrast, LGSOC, CCOC and ENOC demonstrated no significant compartmental differences for either ratio, although substantial inter-sample variability was observed, consistent with heterogeneous immune landscapes. Finally, in MOC, the CD163 + /CD68 + ratio was significantly higher in the stroma than matched tumor nests, supporting a preferential accumulation of M2-like macrophages within the stromal compartment. Collectively, these spatial distributions suggest that while immune exclusion from tumor nests is a shared feature, the composition and magnitude of stromal and intratumoral immune populations differ among EOC histotypes. Borderline tumors are non-invasive neoplastic ovarian lesions and the putative precursor lesions of low grade invasive carcinoma. 31 To investigate immune changes associated with malignant progression, we compared borderline tumors and invasive carcinoma within the serous and mucinous subtypes, analyzing stromal and tumor compartments separately. In serous tumors, comparison of borderline lesions and LGSOC was marked by a selective reshaping of the immune milieu. While overall leukocyte content (CD45 + ) remained largely comparable, LGSOC exhibited lower abundance in CD8 + T-cell ( p  < 0.05) and CD68 + macrophage infiltration ( p  < 0.05), accompanied by enrichment in FoxP3 + regulatory T cells, particularly within tumor nests ( Figure 4A ). This shift reflected in a profound decrease in the CD8 + /FoxP3 + ratio and increase of CD163 + /CD68 + ratio in carcinomas, most prominently in the epithelial compartment, suggesting the establishment of a immunosuppressive microenvironment during malignant transformation ( Figure 4B ). Immune remodeling during progression from borderline tumors to invasive ovarian carcinomas. (A and B) Compartment-specific comparison of immune cell composition and CD8 + /FoxP3 + and CD163 + /CD68 + ratios between borderline serous tumors and low grade serous ovarian carcinomas (LGSOC). (C and D) Corresponding compartment-specific analyses in mucinous tumors, comparing borderline lesions with invasive mucinous ovarian carcinomas (MOC). Group comparisons were performed using two-sided Wilcoxon rank-sum tests and reported p -values were adjusted for multiple comparisons using the false discovery rate (FDR) method; Statistical significance is indicated as follows: p  < 0.05 (*), p  < 0.01 (**), p  < 0.001 (***), p  < 0.0001 (****). Box: Q1–Q3, line: median, whiskers: ±1.5 × IQR; outliers not shown. In mucinous tumors, immune changes characterizing the borderline to carcinoma transition were subtler but followed similar trend to serous tumors. While total leukocyte showed no consistent differences, invasive carcinomas displayed reduced T-cell infiltration within tumor nests compared with borderline tumors ( p  < 0.01; Figure 4C ), and higher CD163 + /CD68 + ratio in the stroma ( p  < 0.05; Figure 4D ). Together, these data suggest that progression from borderline to invasive disease is associated with a histotype-specific “reprogramming” of the immune microenvironment, with serous tumors showing a pronounced shift toward T-cell suppression and macrophage polarization, while mucinous tumors displaying a more moderate but detectable trend towards immune suppression. To validate the clinical relevance of our cohort, we estimated overall survival (OS) according to established clinicopathologic prognostic factors: FIGO stage, tumor grade, residual disease and platinum-free interval (Figure S3A–D). We found that advanced FIGO stage ( p  < 10 −3 ; Figures S3A and S4C), residual disease ( p  < 10 −3 ; Figures S3C and S4C) and platinum-resistance were associated with shorter OS ( p  < 10 −3 ; Figures S3D and S4C) in multivariable analyses, as expected. Additionally, we found that OS differed substantially across the major histotypes of epithelial ovarian cancer. Patients with endometrioid carcinoma experienced the most favorable outcomes (HR = 0.34, adjusted p  < 0.01), whereas those with HGSOC had the poorest survival (Figures S4A and B). Next, we investigated within the HGSOC subgroup whether immune cell infiltration patterns correlated with survival. We found a trend toward prolonged survival in tumors with high intraepithelial CD8 + T-cell infiltration in univariable analysis ( p  = 0.16; Figure 5A left panel). Notably, elevated intraepithelial FoxP3 + Treg infiltration was significantly associated with improved survival (HR = 0.41; p  < 0.01; Figure 5A right panel). Stromal enrichment FoxP3 + Treg (HR = 0.50; p  = 0.02; Figure 5B ) and CD20 + B cells (HR = 0.44; p  = 0.06; Figure 5B ) also tended to associate with better outcome. Together, these findings suggest that intraepithelial CD8 + and FoxP3 + T-cells may reflect a potentially beneficial immune milieu, although multiplex IHC-based immune quantification provided limited prognostic discrimination within this HGSOC cohort. Prognostic impact of immune infiltration in HGSOC. (A) Kaplan–Meier analysis of overall survival according to intratumoral CD8 + T-cell and FoxP3 + Tregs densities in HGSOC patients. Patients stratified in above and below groups using the median percentage of CD8 + cells as cutoff. (B) Multivariable Cox regression model analysis for the different immune cell types in HGSOC patients. For every immune cell marker, patients were dichotomized using median percentages of positively stained cells. Separate multivariable Cox proportional hazards models were generated for each immune marker, with adjustment for FIGO stage (III vs. IV), residual disease status (R0 vs. R1/2), and the other immune cell markers included in the analysis. Hazard ratios (HRs), p -values, and adjusted p -values are shown.

Patients were surgically treated at the University Hospital of Zurich (Switzerland) between 1995 and 2005. Clinical and pathological characteristics were extracted from medical files and pathology records. Tumor staging was assessed according to the 1988 Federation Internationale de Gynecologie et Obstetrique (FIGO) OC staging. Tumor grade and histological subtype were defined according to the 2020 WHO classification. All tumors were reviewed by a gynecologic pathologist, as previously described. 20 The selected cohort consisted of 165 cases, including high-grade serous ovarian carcinoma (HGSOC; n  = 62), low-grade serous ovarian carcinoma (LGSOC; n  = 8), endometrioid ovarian carcinoma (ENOC; n  = 29), clear cell ovarian carcinoma (CCOC; n  = 26), mucinous ovarian carcinoma (MOC; n  = 14), as well as serous ( n  = 15) and mucinous ( n  = 11) borderline tumors. Formalin-fixed paraffin-embedded (FFPE) tissue blocks were retrieved from the tissue archive at the University Hospital of Zurich, Switzerland. Areas involving OC were marked on hematoxylin & eosin (H&E)-stained sections. A tissue microarray (TMA) was constructed using a semi-automatic tissue arrayer (Beecher Instruments, Woodland, USA). Two cylindrical cores of 0.6 mm diameter were punched out from each specimen and inserted into a recipient block, as previously described. 20 FFPE TMA sections were cut at 4 μm thickness and dried overnight. Sections were deparaffinized and rehydrated, followed by heat-induced epitope retrieval (HIER) in citrate buffer (pH 6.0) using a TintoRetriever Pressure Cooker (BSB-7087, Bio SB) at 95 °C under low pressure for 10 min. All subsequent steps were performed at room temperature unless otherwise stated. Endogenous peroxidase activity, non-specific protein binding, endogenous biotin, and avidin were blocked using corresponding blocking reagents from Dako® (see Supplementary Tables 1 and 2 for product details). The first primary antibody was applied to the tissue sections, followed by incubation with a biotinylated secondary antibody and a streptavidin-horseradish peroxidase (HRP) complex. Chromogenic detection was achieved using 3-amino-9-ethylcarbazole (AEC) substrate and chromogen according to the manufacturer's instructions (Innovative Diagnostik-Systeme Dr. Christian Sartori GmbH & Co. KG). Tissue sections were counterstained with Harris hematoxylin for 1 min and mounted with an aqueous mounting medium under glass coverslip. Whole-slide brightfield images were acquired at 20× magnification using the Pannoramic 250 Flash III scanner (3DHISTECH, Budapest, Hungary). Glass coverslips were gently detached by immersion in distilled water at 37 °C. AEC chromogen was removed by dehydration through a graded ethanol series up to 95% ethanol until no visible reaction product remained. Antibodies were stripped by boiling sections in HIER buffer for 10 to 20 min, and putative residual antibodies were blocked by Fab fragments directed against the host of the previously applied primary antibody. Multiplexed IHC was performed through sequential cycles of staining, scanning, destaining, antibody stripping and staining of the subsequent antibodies. A total of seven sequential staining rounds were conducted in the following order: FoxP3, CD45, CD8, CD68, CD163, CD20, and cytokeratin. Detailed information on primary and secondary antibodies, reagents, dilutions, and specificities is provided in Supplementary Tables 1 and 2. Only tumor cores containing invasive adenocarcinoma in at least 5% of the total tissue area and free of artifacts were selected for further analysis. Digitized whole-slide images were analyzed using the built-in tissue microarray (TMA) functionality in QuPath (v0.4.4). 21 Tumor and stromal compartments were manually annotated based on cytokeratin staining. Cell segmentation was performed using the Cellpose algorithm 22 implemented via the QuPath extension, with nuclei identified from the hematoxylin counterstain. Marker-positive cells were detected based on 3-amino-9-ethylcarbazole (AEC) chromogen intensity, with positivity thresholds defined by visual inspection. Quantification of each marker was performed independently on individual brightfield images rather than composite images to avoid exclusion of TMA cores in cases of marker-specific staining failure or tissue deterioration following multiple staining rounds. This approach enabled extraction of quantitative data for individual markers even when high-quality images were not available for all seven markers. Cell-type proportion was estimated as the number of positive cells over total tissue cells detected with nuclei identified from the hematoxylin counterstain. In addition, pseudo-multichannel images were generated for visualization purposes by image alignment in MATLAB. For each core, tumor cells islets (epithelial compartment) were drawn manually as well as stroma areas in order to analyze the two compartments separately. To analyze strictly the same area on all images when image merge is not possible, an annotation transfer script was created, to transfer the TMA detection as well as the previously traced area. 23 A machine learning script was written to efficiently detect cells and their nuclei, and previously described. 24 Statistical analyses were performed in R (version 4.5.2) 25 using RStudio. 26 Data cleaning, organization and preparation were carried out using the tidyverse suite of packages (v2.0.0). 27 Group comparisons were assessed using the Wilcoxon rank-sum test for non-parametric distributions. p -values were adjusted for multiple testing using the false discovery rate (FDR) method, and adjusted p -values < 0.05 were considered statistically significant. All data visualizations were generated with ggplot2 (v4.0.0). 28 Overall survival was calculated using the survival package (v3.8-3) 29 and visualized by the survminer package (v0.5.1) 30 in RStudio. Overall survival (OS) was defined as the time from the date of diagnosis to the date of death by any cause. Patients alive at the time of last follow-up were censored on the date of last clinical contact. Survival time was calculated in months as the interval between diagnosis and death or last follow-up. Patients were stratified into above and below median groups based on the median percentage of positive cells for each marker as the cutoff. Survival functions for the different subgroups were estimated using the Kaplan–Meier method and compared with the log-rank test. Hazard ratios for each group were estimated using Cox proportional hazards regression. The multivariable model of overall survival in HGSOC was adjusted for advanced FIGO stage (III–IV) and the presence of macroscopic residual disease.

Although EOC has long been considered an immunogenic tumor type, 10 compiling data have demonstrated that this immunogenicity does not translate into clinical benefit from immune checkpoint blockade. Several randomized phase III trials evaluating PD-1/PD-L1 inhibitors in combination with chemotherapy, with or without anti-angiogenic agents or PARP inhibitors, failed to improve survival in both first-line and recurrent disease settings. 32-35 These disappointing results underscore the need for more refined studies of immune infiltrate of EOC, to enable precision immunotherapy. 36 While numerous studies have extensively characterized the phenotype, functional alterations, and prognostic significance of immune infiltrates in EOC, 37 few investigated immune infiltrate across the 5 histotypes with multiplexed imaging and digital pathology. 16 , 38 Further, most quantitative estimations of the total abundance of immune cells in EOC were mainly performed using transcriptomics data. 39 , 40 In this study, we applied multiplexed IHC and digital pathology to a clinically annotated TMA that included the 5 major histotypes of EOC according to the 2020 WHO classification: high-grade serous (HGSOC), low-grade serous (LGSOC), endometrioid (ENOC), clear cell (CCOC), and mucinous (MOC). Our study revealed common patterns with macrophages being the most abundant immune cell subsets. Another consistent feature across all histotypes was the preferential localization of immune cells within the stroma, with relative exclusion from tumor epithelium. This immune-excluded architecture represents a major barrier to effective anti-tumor immunity and likely contributes to primary resistance to immune checkpoint blockade, which requires spatial proximity between effector T cells and tumor cells. 41 Our work also revealed substantial heterogeneity in both the composition and spatial organization of immune infiltrates across EOC histotypes, highlighting that EOC is not represented by a single immunological repertoire but rather a spectrum of distinct immune microenvironments. CCOC is of particular interest since this histotype showed so far, the highest benefit from immune checkpoint blockade. 42 , 43 We found that CCOC had the highest infiltration by CD45 + leukocytes, indicating elevated overall immune cell density. Consistently with previous reports, we found that macrophages were the predominant immune population in all subtypes, with particular enrichment of CD68 + and CD163 + cells in CCOC, indicating a skewing toward an M2-like phenotype. 39 Interestingly, MOC exhibited a higher CD8 + /FoxP3 + ratio and reduced M2 macrophage polarization despite low immune cell abundance. As elevated CD8 + /FoxP3 + ratios have been associated with prolonged survival 11 , 44 and enhanced responsiveness to PD-1/PD-L1 blockade, 45-47 this immune balance in MOC may represent a biologically relevant, but currently unexplored, therapeutic opportunity. Finally, ENOC closely resembled HGSOC, supporting partial convergence of immune landscapes across these histotypes. Analysis of LGSOC and its putative precursor lesions, serous borderline tumors revealed greater abundance of Treg, less CD8 + T cells and a shift toward M2 macrophages. These observations are consistent with recent findings in borderline tumors 48 as well as serous tubal intraepithelial carcinoma, 49 the precursor lesion of HGSOC, suggesting that immune escape is a dynamically acquired feature during tumorigenesis and progression of the disease. Mucinous tumors followed a similar, although less pronounced trajectory. The limited number of patients per histotype limited the investigation of the prognostic value of individual immune populations. Thus, we focused on HGSOC. Consistent with the literature, we found that intraepithelial CD8 + , a well-established prognostic factor, 10 and stromal B cells, 12 , 50 were associated with a trend toward improved survival. Interestingly, intraepithelial FoxP3 + Treg infiltration was significantly associated with prolonged survival, while stromal FoxP3 + Treg enrichment also showed a favorable association. Although this observation contrasts with the conventional perspective that Treg accumulation promotes immunosuppression and poor prognosis, 51 increasing evidence suggests that the prognostic role of FoxP3 + cells in HGSOC is context-dependent. Recent studies using multiplex IHC and spatial immune profiling have similarly reported that subsets of FoxP3 + lymphocytes may associate with improved clinical outcome or reflect coordinated adaptive immune infiltration rather than a purely suppressive immune compartment. 50 , 52 , 53 In particular, Marchenko and colleagues demonstrated that increased CD25 + /FoxP3 + Treg densities in both intratumoral and stromal compartments were associated with prolonged overall survival and underlined the importance of spatial compartmentalization when assessing immune infiltrates. 53 Their study further points that methodological differences in Treg identification, histologic composition of cohorts, and compartment-specific analyses may explain conflicting findings reported in earlier ovarian cancer studies. In concordance with this, a well-refined exploratory biomarker analysis from the KEYNOTE-100 study highlighted higher densities of FoxP3 + immune infiltrate in responders to anti-PD1 pembrolizumab together with activated CD8-related immune signatures (CD8 + /GB + /Ki67 + activated, proliferating cytotoxic T lymphocytes), supporting the idea that FoxP3 + enrichment may reflect broader immune activation in selected EOC microenvironments. 54 A strength of this study is the large, well-annotated cohort and robust histotype classification according to WHO 2020 classification, which included both invasive carcinoma and borderline tumors. Further, the clinical outcome was well-annotated and the major prognostic factors (FIGO stage, residual disease and histotype) were associated with outcome, as expected. However, several limitations should be acknowledged, including the use of a restricted multiplex IHC marker panel (six immune markers plus cytokeratin) and the inherent technical constraints of sequential staining, which may affect signal fidelity and subsequent accurate immunological characterization. Other limitations are: (1) the analysis of TMA core that does not reflect intratumor heterogeneity; (2) the limited numbers per histotype, preventing survival analyses except for HGSOC; (3) the absence of genomic characterization such as BRCA/HRD and microsatellite instability; and (4) the fact that the cohort included patients treated between 1995 and 2005, before the approval of maintenance therapies such as PAPRPi and bevacizumab. Future studies integrating spatial transcriptomics, genomic alterations and multiomics approaches are needed to better elucidate the immune landscape of EOC.

In conclusion, EOC displays a heterogeneous, histotype-specific immune microenvironment characterized by macrophage predominance and immune exclusion from the tumor epithelium, a spatial organization that may contribute to resistance to immune checkpoint blockade. Distinct immune landscapes were identified across histotypes, with CCOC showing the highest immune infiltration and enrichment of M2-like macrophages, while MOC exhibited a relatively favorable CD8 + /FoxP3 + balance despite low immune cell abundance. LGSOC tumors showed features of progressive immune escape, including increased Treg infiltration, reduced CD8 + T cells, and immunosuppressive macrophage polarization. In HGSOC, intraepithelial CD8 + T cells and stromal B cells were associated with improved survival, whereas FoxP3 + Treg infiltration surprisingly correlated with favorable outcome, supporting a context-dependent role of FoxP3 + cells in ovarian cancer. Together, these findings support a model of histotype-driven immune architecture in EOC and highlight the necessity for histotype-specific immunotherapies.

supp_fig4.tif supp_fig1.tif supp_fig3.tif supp_fig2.tif KONI_A_2722470_SM6866.docx KONI_A_2722470_SM6868.docx