Serum CD95L Level Correlates with Tumor Immune Infiltration and Is a Positive Prognostic Marker for Advanced High-Grade Serous Ovarian Cancer Free

Ovarian carcinoma is the seventh most common cancer in women and the eight most common cause of cancer-related death worldwide (1). At the time of diagnosis, the majority of patients with epithelial ovarian cancer (EOC) present with advanced disease, which is characterized by a high and widespread tumor load in the peritoneal cavity, often accompanied by malignant ascites. Thus, the prognosis of women with ovarian cancer remains poor, with a 5-year overall survival (OS) rate estimated at <45% for all cancer stages, and only 20% to 30% for patients with Federation of Gynecologists and Obstetricians (FIGO) stage III or IV disease (1, 2).

Patients with EOC and other cancers who exhibit a robust immune response show increased survival rates (3). Recent studies show that number of tumor-infiltrating lymphocytes (TIL), which are lymphocytes that extravasate from blood vessels to access the tumor, may be a positive predictive factor for melanoma (4), colorectal cancer (5), esophageal carcinomas (6), breast cancer (7, 8), or endometrial cancer (9). The presence of TILs affects the outcome of ovarian cancer; high numbers of CD8⁺ T cells in the immune infiltrate are associated with improved OS (10–15), particularly in patients with high grade serous ovarian carcinoma (HGSOC; refs. 16–18). In contrast, clinical outcome of ovarian cancers infiltrated by regulatory FoxP3⁺ T cells (Tregs) remains unclear; studies suggest either decreased OS (19–21) or improved clinical outcomes (22–24). Furthermore, some works establish that ovarian cancer is often accompanied by systemic immunosuppression (25, 26), which correlates with a poor prognosis (27, 28). New treatment strategies based on neutralizing antibodies that target checkpoint inhibitors represent a revolution in the fight against cancer; indeed, such treatments have shown survival benefits in patients with melanoma (29) or lung cancer (30). Therapeutic antibodies that block the PD1/PD-L1 checkpoint (such as nivolumab or pembrolizumab) or the CTLA4 checkpoint (such as ipilimumab) show therapeutic responses linked to tumor immune infiltration (31, 32). Phase II studies of anti-PD1/PD-L1 therapy in ovarian cancer suggest that it triggers antitumor responses (33–35); and, several phases III studies are underway. In such cases, a serum theranostic biomarker would be useful for selecting patients that are eligible for immunotherapy trials.

CD95L (also called FasL), which belongs to the TNF family, binds to the receptor CD95 (also known as Fas). Whereas CD95 is ubiquitously expressed, CD95L shows a more restricted expression pattern; it is expressed mainly on the membrane of lymphocytes, where it plays a pivotal role in eliminating infected and transformed cells (36). Binding of membrane bound CD95L to CD95 recruits the adaptor protein Fas Associated Death Domain (FADD; ref. 37), which in turn aggregates caspase-8 and caspase-10 to induce apoptosis (38). CD95L is also expressed by endothelial cells lining the blood vessels of patients with tumors and chronic inflammatory disorders (39–41). These CD95L-expressing endothelial cells seem to behave as a selective immune barrier, killing CD8 T cells while being permissive for Treg cells (41). CD95L can be cleaved by metalloproteases, thereby releasing soluble CD95L (s-CD95L) into the bloodstream. Binding of s-CD95L to CD95 fails to trigger cell death but rather induces a nonapoptotic signaling pathway that promotes the migration of T cells (42). Here, we wondered whether s-CD95L plays a role in ovarian cancer and examined its impact on the immune landscape in patients with HGSOC.

IGROV-1, OVCAR-3, OVCAR-8, and SKOV-3 cell lines were obtained from ATCC (LGC Standards) and were authenticated by short tandem repeat. Each month, routine testing was conducted on all cultured cells using the sensor-cell approach PlasmoTest - Mycoplasma Detection Kit (InvivoGen). O170, O370, O386, and O829 cell lines were established in-house from primary solid tumor or carcinomatosis samples cut into small pieces (<1 mm³) with a scalpel and then subjected to enzymatic digestion with collagenase (0.23 Wünsch units/mL; Liberase Research Grade; Roche). These in-house cell lines had more than 20 passages. All cell lines were cultured in DMEM complete medium supplemented with l-glutamine and 8% FBS. Cells were used within 6 months after resuscitation of frozen aliquots.

Cells (1,000/well) were seeded in 96-well ultra-low attachment plates (Corning) and cultured for 7 days in serum-free culture medium (MammoCult Human Medium Kit, Stemcell Technologies) supplemented with heparin (Stemcell Technologies) and hydrocortisone (Stemcell Technologies). At day 7, the number of spheres per well was counted manually by taking five large field pictures of each well using a 4 × objective lens.

HEK/293T cells cultured in 1% FCS containing medium were transfected with 3 μg of empty plasmid or a wild-type CD95L-containing vector using the calcium/phosphate precipitation method. Medium containing s-CD95L and exosome-bound full-length CD95L was harvested 5 days after transfection. Dead cells and debris were removed by centrifugation (2 × 4,500 rpm/15 minutes). Exosomes were pelleted by ultracentrifugation at 100,000 × g for 2 hours. Finally, debris- and exosome-free supernatants were concentrated (Centricon; 10 kDa cutoff) and dialyzed against PBS.

Cells were lyzed for 30 minutes at 4°C in lysis buffer (25 mmol/L HEPES pH 7.4, 1% v/v Triton X-100, 150 mmol/L NaCl, 2 mmol/L EGTA supplemented with a mix of protease inhibitors; Sigma-Aldrich). Protein concentration was determined using the Bicinchonic Acid Method (Pierce). Proteins were separated on an 8% SDS-PAGE gel and transferred to a nitrocellulose membrane (GE Healthcare). The membrane was blocked for 30 minutes with TBST (50 mmol/L Tris, 160 mmol/L NaCl, 0.05% v/v Tween 20, pH 7.8) containing 5% w/v dried skimmed milk or BSA and incubated overnight at 4°C with primary antibodies (mouse anti-E-Cadherin, clone 36, BD Pharmingen; mouse anti-Vimentin, clone V9, DAKO; and mouse anti- α-actin, clone AC-74, Sigma) diluted in the same buffer used for saturation (milk for anti-E-cadherin and anti-α-actin antibodies and BSA for anti-vimentin antibodies). The membrane was washed with TBST, followed by incubation for 1 hour with a peroxidase-conjugated anti-mouse antibody (SouthernBiotech). Proteins were visualized using the enhanced Chemiluminescence Substrate Kit (GE Healthcare).

Cells were stained for 30 minutes at 4°C with anti-CD24 (clone ML5, BD Pharmingen), CD44 (clone G44-26, BD Pharmingen), and anti-CD95 (clone DX2, BD Pharmingen) mAbs. Isotype-matched murine fluorochrome-conjugated immunoglobulins (PE mouse IgG2a, APC mouse IgG2b, and PE mouse IgG1, respectively; BD Pharmingen) were used as negative controls.

Total RNA was extracted from the cells using the NucleoSpin RNA Isolation Kit (Macherey-Nagel) and cDNA was generated from 1 μg of total RNA using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Quality and quantity of total RNA and cDNA was measured using a DS-11 Spectrophotometer (DeNovix). Expression of mRNA was measured using the TaqMan Gene Expression Assay Kit (Applied Biosystems), which contains primers and TaqMan MGB probes specific for the following genes: FAP1 (assay no: Hs00196632_m1), ZEB1 (Hs00611018_m1), ZEB2 (Hs00207691_m1), CDH1 (Hs00170423_m1), SNAIL (Hs00195591_m1), TWIST1 (Hs01675818_s1), and human GAPDH (Hs99999905_m1; endogenous control). The probes targeting SLUG (ENST00000020945.3|ENSG00000019549.9) were designed using the Universal Probe Library's Assay Design Center Tool (Roche), and the forward (TGGTTGCTTCAAGGACACAT) and reverse (GCAAATGCTCTGTTGCAGTG) primers were purchased from Eurogentec. Expression of SLUG mRNA was measured using PowerUp SYBR Green Master Mix (Applied Biosystems), with human GAPDH used as an endogenous control. cDNA (50 ng) was used for each qPCR reaction. All qPCR experiments were performed in a QuantStudio5 Machine (Applied Biosystems) and the data were analyzed using Thermo Fisher Connect Software (Thermo Fisher Scientific). Relative expression of each gene was calculated using the 2−ΔΔC_t method and expressed as a fold change.

Cell viability was measured in an MTT assay. Briefly, 4 × 10⁴ cells were cultured for 24 hours in flat-bottom, 96-well plates along with the indicated concentrations of apoptosis inducer (final volume, 100 μL). Next, 15 μL MTT (5 mg/mL in PBS) solution was added and incubated at 37°C for 4 hours. Absorbance was measured using the Infinite F200 Pro (Tecan) at a wavelength of 570 nm.

To determine whether s-CD95L contributed to T-cell migration, T cells were exposed to s-CD95L or control medium in Boyden chambers. This migration assay was described previously (43). Briefly, to measure trans-endothelial migration of activated T cells, a 3-μm sized porous membrane of a Boyden chamber was used. Human umbilical vein endothelial cells (HUVEC) were plated to form a monolayer mimicking endothelial barrier. Membranes were first hydrated in sterile PBS, then, CD3/CD28-activated peripheral blood lymphocytes (PBL) isolated from healthy donors (3 × 10⁵ cells/300 μL) were added to the top chamber covered with a monolayer of HUVEC in a low serum (1%)-containing RPMI. Bottom chamber contained 500 μL of RPMI 1% FBS in presence or absence of s-CD95L (100 ng/mL). Cells were cultured in a CO₂ incubator at the same conditions as adherent cells for 24 hours. Transmigrated cells were then counted in the lower reservoir.

All clinical investigations were conducted in accordance with the principles outlined in the Declaration of Helsinki. Blood samples were collected from patients diagnosed with ovarian cancer after written informed consent was obtained. The study was approved by the local institutional review board (CEROG 2016-GYN-1003). Samples collected prospectively from patients diagnosed with ovarian cancer between January 2010 and December 2013 were reviewed retrospectively. All samples were obtained at the time of diagnosis and before chemotherapy treatment. A total of 51 patients with advanced stage ovarian cancer were analyzed: 37 with HGSOC, 6 with endometrioid subtype, 4 with mucinous subtype, 2 with clear cell subtype, and 2 with low grade serous ovarian cancer. Thirty-six patients received neoadjuvant chemotherapy after tumor sample plus interval debulking surgery, and 15 underwent primary debulking surgery followed by adjuvant chemotherapy. The surgical specimens were evaluated histologically at the Department of Pathology. All patients were staged according to the FIGO staging system (44). After surgery, all patients received standard chemotherapy comprising carboplatin plus paclitaxel. For this study, the main inclusion criterion was ovarian cancer with FIGO stage IIIC and IV (i.e., carcinomatosis stage or higher) and a serum sample in which the s-CD95L level before chemotherapy could be measured. Progression-free survival (PFS) was defined as the time from diagnosis of ovarian cancer to the time of recurrence or death. OS was defined as the time from diagnosis of ovarian cancer to the time of death. Observation time was defined as the interval between diagnosis and time of last contact (death or last follow-up). Data were censored at death or at the last follow-up for patients without recurrence.

S-CD95L concentrations in the serum of patients with HGSOC and healthy donors were measured by ELISA (Diaclone). All blood samples were harvested at the time of diagnosis.

Surgical specimens were fixed in 4% formalin, embedded in paraffin, and stained with hematoxylin–eosin–saffron (HES). Sections from each histologic specimen were reviewed by two experienced pathologists (University Hospital, Rennes, France) to confirm the diagnosis and grade according to the method of Silverberg or Malpica (35). For each patient, a representative HES slide and the corresponding paraffin block were selected. The selected slide had to contain both tumor and adjacent stroma. The formalin-fixed, paraffin-embedded blocks were cut into 5-μm slices and mounted on SuperFrost Plus Microscope Slides (Menzel-Gläser). Expression of CD3 (clone SP7; Thermo Fisher Scientific), CD4 (clone SP35; Cell Marque), CD8 (clone C8/144B; Dako) CD20 (clone L26; Dako), CD163 (clone 10D6; Leica), CD31 (clone JC70A; Dako), Podoplanin (clone D2-40; Invitrogen), FoxP3 (clone SP97; Eurobio Scientific), Granzyme B (clone GrB-7; Millipore), IL17 (clone bs-2140R; Bioss antibodies), and PD-L1 (clone E1L3N; Cell Signaling Technology) was assessed by IHC (Ventana Discovery XT automaton, Ventana Roche). CD95L (clone G247-4, BD Pharmingen) was immunostained manually. After deparaffinization with toluene and rehydration with ethanol, sections were incubated at 95°C and bathed in Tris-EDTA at pH 8 prior to staining for CD4, CD8, FoxP3, CD20, CD3, CD31, CD163, CCR6, and Granzyme B. For CD95L, sections were deparaffinized and rehydrated, incubated at 95°C, and immersed in EDTA (pH 9). For all preparations, endogenous peroxidase activity was blocked by incubation in 0.3% hydrogen peroxide. The reactivity of all antibodies, except CD95L, was revealed with a horseradish peroxidase (HRP)-labeled polymer-conjugated secondary antibody followed by diaminobenzidine (DAB; OmniMap DAB Roche). CD95L was revealed with an anti-mouse HRP-labeled polymer-conjugated secondary antibody followed by DAB (DAB Dako).

According to the breast cancer guidelines (45) TILs can be subdivided according to their location within the tumor: (i) stromal TILs when located in the peritumoral space and (ii) intraepithelial TILs, when they have penetrated the tumor islets. These recommendations are for breast cancer, and yet, there is no standardized approach to evaluate TILs in EOC. Thus, to avoid a biased evaluation, we decided to perform a representative HES slide, which had to contain both tumor and adjacent stroma, to evaluate overall immune cells whatever their localization in tumor. Each immunostained slide was scanned with a NanoZoomer (Hamamatsu). For each patient, five large field pictures were taken using a 5 × objective lens and Hamamatsu's Software (NDPview). All fields were analyzed using NIS-Elements Software (Nikon) and positively stained cells were counted; for analysis, a specific threshold was applied for each antibody.

Statistical analysis was performed using SAS, v.9.4 (SAS Institute) and R logical Version 3.4.1 software programs. Quantitative results were expressed as the statistical mean ± SD and qualitative results as percentages (%). The Mann–Whitney–Wilcoxon test was used to compare the distribution of quantitative variables between two groups (abnormal statistical distributions). The χ² or Fisher exact test was used to compare the distribution of qualitative variables between two groups with theoretical headcounts <5. The method of Contal and O'Quigley was used to determine cut-off values for the continuous variables used to examine prognosis (46). The correlation between TILs and levels of cleaved CD95L was assessed using Spearman correlation. The Kaplan–Meier method was used to compare survival curves between groups. All tests were two-sided and a P < 0.05 was deemed significant.

This study was agreed to by local institutional review board and French laws. All patients consented to participate.

The clinical data and outcomes of our current HGSOC cohort match those of previously described cohorts (47, 48), indicating that, although the number of patients was relatively small the cohort is likely to be representative of larger HGSOC cohorts (Supplementary Table S1). In agreement with known clinical prognostic factors (2, 48, 49), univariate analysis identified involved lymph nodes (HR, 4.87; 95% CI, 1.67–14.16) and residual disease after surgery (HR, 5.74; 95% CI, 2.40–13.74) as clinical parameters significantly associated with disease-free survival (DFS) and OS (Table 3). Multivariate analysis revealed that residual disease after surgery (HR, 6.16; 95% CI, 2.27–16.67) correlated with recurrence rate (Table 4). Because we showed that serum CD95L is a poor prognosis marker in triple-negative breast cancer (TNBC) women (26), we next investigated whether the concentration of serum CD95L in women with advanced ovarian cancer was associated with clinical outcome. Serum CD95L concentrations in patients with ovarian cancer showed a more heterogeneous distribution than in healthy donors (Fig. 1A), and there was no statistically significant difference between serum CD95L concentrations in patients with ovarian cancer and healthy subjects (234.6 ± 28.1 pg/mL vs. 359 ± 73.8 pg/mL, respectively).

The Contal and O'Quigley method (46) revealed that a serum CD95L concentration of 516 pg/mL showed the strongest prognostic value for patients with HGSOC (Fig. 1C and E). We found that this cutoff failed to discriminate patients with long-term DFS from the whole population of patients with ovarian cancer (Fig. 1B). Nonetheless, for patients with HGSOC, this concentration discriminated poor responders from long-term survival patients (Fig. 1C). In agreement with this, the median concentration of s-CD95L in patients with DFS < 12 months was significantly lower than that in patients with a DFS >12 months [236 (0–480) vs. 317 (0–738), respectively; P = 0.039; Fig. 1D], and the OS of patients with serum CD95L concentrations > 516 pg/mL was significantly longer than that of patients with a lower concentration (Fig. 1E). Furthermore, univariate analysis pointed out that s-CD95L concentrations below 516 pg/mL were significantly correlated with node involvement (28% vs. 46%, respectively; P = 0.0409; Table 1) and node involvement is strongly associated with tumor relapse (Table 3).

Because high concentrations of s-CD95L in patients with HGSOC were associated with a better clinical outcome, we asked whether this ligand had a direct effect on death, proliferation, and/or differentiation of ovarian cancer cells. CD95 can induce cell death upon binding to membrane-bound CD95L (m-CD95L; ref. 38). Therefore, we examined whether CD95L could kill ovarian cancer cells. All tested ovarian cancer cell lines (OVCAR-3, OVCAR-8, SKOV-3, and IGROV-1) and cells established from patients with HGSOC (O170, O370, O386, and O829) expressed CD95 (Supplementary Fig. S1A). These cells were classified as epithelial (OVCAR3, O370, O170, and 0386) or mesenchymal (O829, IGROV1, SKOV-3, and OVCAR-8) cells based on the vimentin/cadherin ratio (Supplementary Fig. S1B). Using a high s-CD95L concentration (i.e., 100 ng/mL), we did not observe any cytotoxic effect of s-CD95L alone or in combination with doxorubicin and carboplatin chemotherapy drugs (Supplementary Fig. S2A and S2B).

Tumor heterogeneity is explained mainly by the hierarchical organization of tumor tissues, in which several subpopulations of self-renewing cancer stem cells (CSC) sustain the long-term oligoclonality of the neoplasm (50). Moreover, CSCs are thought to be the seeds required to establish distant metastasis, which is responsible for poor clinical outcome (51). Addition of s-CD95L, with or without chemotherapy agents, did not affect the number of ovarian CSCs, defined as CD24^LowCD44^HighALDH1⁺ (Fig. 2A and C; Supplementary Fig. S2C). Moreover, neither the epithelial profile of OVCAR-3 nor the mesenchymal profile of OVCAR-8 varied in the presence of s-CD95L (Fig. 2B). Finally, in all tested ovarian cancer cells, the rate of proliferation did not change in the presence of s-CD95L (Supplementary Fig. S1D–S1F). Overall, these findings indicated that soluble CD95L has no effect on cell death, proliferation, or differentiation of ovarian cancer cells and does not enhance the antitumor effects of chemotherapeutic drugs. This leads to the hypothesis that s-CD95L may affect the tumor microenvironment, specifically the number/composition of immune tumor-infiltrating cells, in patients with HGSOC.

CD95L is expressed by endothelial cells lining the vessels of some ovarian cancers (41). Therefore, we wondered whether this was also the case for HGSOCs. IHC analysis of tumor sections identified mCD95L in 29 (79.2%) patients, but not in 8 (20.8%). Using serial tumor slices, we validated that mCD95L was expressed by CD31-expressing endothelial cells (Fig. 3A), but not lymphatic endothelium (D2-40+). Interestingly, the intensity of m-CD95L expression in endothelial cells was correlated with the quantity of s-CD95L dosed in serum of these patients (Supplementary Fig. S3A and S3B). Of note (and unlike s-CD95L), we found no significant correlation between the intensity of staining for m-CD95L in tumor tissue and PFS or OS (Table 3), suggesting that metalloprotease-mediated release of s-CD95L in ovarian cancers could exert an important biological function that impacts prognosis.

Next, because tumor-infiltrating immune cells (10, 11, 18) are associated with a good clinical outcome in HGSOC, we examined tissues for the presence of tumor-infiltrating T cells (CD3/CD4/CD8/Th17), B cells (CD20), and macrophages (CD163). The cytolytic activity of CD8⁺ T cells was monitored by staining for Granzyme B. Checkpoint inhibitors are promising therapeutic regimens for patients with HGSOC; therefore, we also stained tissues for PD-L1. Although the intensity of CD3, CD8, CD4, CD20, and CD163 markers did not change regarding the staining of membrane CD95L in tumor tissues, transmembrane CD95L expression was significantly associated with the number of tumor-infiltrating FoxP3 T cells (Table 2,Table 3). This finding is in agreement with results published by Coukos and colleagues (41), who suggested that m-CD95L is a selective immune barrier; it killed CD8⁺ cells but spared FoxP3-T cells. Nonetheless, we found no significant correlation between the intensity of m-CD95L staining and PFS or OS, indicating that unlike metalloprotease-cleaved s-CD95L, its membrane bound counterpart is not a prognostic marker for patients with HGSOC and that a yet unidentified metalloprotease plays a pivotal role in the disease progression. Because s-CD95L promotes T-cell trafficking in patients with lupus (ref. 52; Supplementary Fig. S4C) and expression of matrix-metalloproteinase-7 (MMP7) is higher in these patients (53), MMP7 might be a good candidate for the metalloprotease involved in CD95L cleavage in HGSOC. Although the number of tumor-infiltrating CD8⁺ T cells did not correlate with the expression of m-CD95L, we found that the number of activated (GZB⁺) CD8⁺ T cells was higher in tumor tissues with greater expression of m-CD95L (Supplementary Fig. S3C). Because m-CD95L levels were not associated with a good prognosis, but activated (GZB⁺) CD8⁺ T cells were, this observation suggests immunosuppressive activity even in tumors showing high endothelial expression of m-CD95L.

Because high levels of serum s-CD95L are associated with a good prognosis in patients with HGSOC, we next investigated whether the concentration of s-CD95L was associated with the number of TILs. We observed that the concentration of s-CD95L was correlated with the number of tumor-infiltrating CD3⁻ and CD4-expressing T cells (CD3: r = 0.4373, P = 0.0068; CD4: r = 0.3284, P = 0.0472; Supplementary Fig. S4A). Counterintuitively, the number of tumor-infiltrating Treg cells (FoxP3⁺) in patients with HGSOC also correlated with s-CD95L expression (r = 0.4584, P = 0.0043); in contrast, the number of infiltrating IL17-producing Th17 cells did not (Fig. 3B; Supplementary Fig. S4A). The number of B cells was associated significantly with the concentration of s-CD95L (Fig. 3B; Supplementary Fig. S4A). Finally, we found no correlation between s-CD95L levels and the number of tumor-infiltrating macrophages (CD163⁺ cells; Fig. 3B; Supplementary Fig. S4A). Also, there was no correlation between s-CD95L and Granzyme B (a marker of CD8⁺ T-cell activation) or PD-L1 (Fig. 3B; Supplementary Fig. S4A), suggesting that s-CD95L is not involved in tumor recruitment/activation of cytolytic CD8⁺ T cells (GZB) or expression of PD-L1, which contributes to exhaustion of CD3⁺ T cells.

Overall, these findings suggested that the chemoattractant s-CD95L increases recruitment of CD8⁺ T cells and Treg cells, among the CD4⁺ T cells, to HGSOC tumors, leading to an improved clinical outcome. Of note, we did not find any correlation between serum s-CD95L level and CD3⁺ T-cell count in blood sample of patients with HGSOC (Supplementary Fig. S4B). To confirm that s-CD95L was able to promote cell motility of activated PBLs, we incubated CD3/CD28-activated PBLs in the presence or absence of s-CD95L and evaluated cell migration using Boyden chambers. As shown in Supplementary Fig. S4C, s-CD95L enhanced the migration of activated T cells.

High levels of serum CD95L (s-CD95L) is correlated with the number of tumor-infiltrating immune cells in high-grade serous ovarian cancer (HGSOC). These findings suggest that s-CD95L might be used as a noninvasive marker of tumor immune infiltration. This would avoid tumor biopsy, which is difficult when a patient has relapsed. Immune checkpoint inhibitors such as PD-1/PD-L1 or CTLA4 antibodies are more effective in tumors with immune infiltration. In future a personalized approach could be envisaged: after initial treatment, patients with HGSOC are tailored according to immune factors. Thus, s-CD95L as a surrogate of tumor immune infiltration could be used to select patients with HGSOC referred for immunotherapy trials. Patients with low s-CD95L levels may be more appropriately directed to clinical trials using molecular therapies.

Here, we show that patients with HGSOC with high levels of s-CD95L have a better prognosis than those with low levels. In addition, high s-CD95L levels correlate with increased numbers of tumor-infiltrating immune cells, including T cells (CD8⁺ lymphocytes and Tregs) and B lymphocytes. These findings suggest that s-CD95L plays a role in regulating tumor immune responses in women with HGSOC and could therefore be a noninvasive marker of tumor immune infiltration. Such a marker would avoid the need for tumor biopsy, which is difficult when a patient has relapsed. Immune checkpoint modulators such as PD-1/PD-L1 or CTLA4 antibodies do not recruit lymphocytes, but break immunosuppression, which lead to restore cytotoxic T-cell activity. Indeed, PD-1/PD-L1 blockade is more effective in tumors with immune infiltration. PD-1/PD-L1 antibodies trigger objective tumor responses in only 20%–30% of patients with recurrent ovarian cancer (34, 35). So far, no biomarkers are associated strongly with high response rates, although PD-L1 expression on both tumor and immune cells can be used to select patients that are more likely to respond to PD1/PD-L1 treatment (34). However, a pathology sample is needed to assess these biomarkers. Moreover, due to the heterogeneous nature of cancer, such a sample may not reflect the disease. In future, a personalized approach could be envisaged: after initial treatment, patients with HGSOC are tailored according to expression of immune factors. Thus, s-CD95L, as a surrogate marker of tumor immune infiltration, could be proposed to select patients with HGSOC for immunotherapy trials: indeed, patients with high s-CD95L expression could be offered check point inhibitor treatment as a maintenance therapy, and patients with low s-CD95L expression may be more appropriately referred to clinical trials of chemotherapies. Nevertheless, a stronger efficiency of PD1/PD-L1 checkpoint inhibitor has still to be demonstrated for patients with ovarian cancer with high infiltrate of immune cells as compared with women showing a low immune response. By now, the results of immunotherapy trials in ovarian cancer remain disappointing. Mesnage and colleagues showed that neoadjuvant chemotherapy increases tumor immune infiltration in some patients with HGSOC (49); thus s-CD95L as a marker of tumor immune infiltration may allow us to monitor immune responses during neoadjuvant chemotherapy, and to select patients for immunotherapy when a strong tumor immune infiltrate is observed after neoadjuvant chemotherapy.

Although our data required an external validation with an independent cohort, one strength of this study is that we selected only patients with HGSOC and FIGO III and IV stage (i.e., carcinomatosis stage or tumor peritoneal spread), which avoids confusion with other subtypes of ovarian cancer harboring different molecular mutations (e.g., mucinous, endometrioid, or clear cell ovarian cancer) and showing different immune responses (54, 55). Indeed, we observed no correlation between survival and s-CD95L levels in patients with HGSOC, mucinous, or endometrioid cancer. Furthermore, IHC experiments identified several types of immune cell in tumor tissue from these highly selected patients: Tregs (FoxP3 T cells), CD8 T cells, B cells (CD20 cells), and macrophages (CD163 cells). Zhang and colleagues showed a correlation between TIL numbers and OS and PFS (11), as did Tomsova and colleagues (13), Sato and colleagues (10), and others (56, 57). Nevertheless, these studies suffer from lack of immune cell markers and heterogeneity. A counter-intuitive result is the better prognosis in patients with higher rate of tumor-infiltrating Tregs (FoxP3 T cells). The magnitude of the immune reaction including tumor-infiltrating CD4⁺ and CD8⁺ T cells and FosP3⁺ T cells is associated with a better prognosis. Of note, in multivariate analysis, CD3⁺ and CD8⁺ T cells remain independent parameters associated with a better survival, while FoxP3 T cells are not (Tables 4 and 5). It is also noteworthy that our IHC analysis fails to discriminate the different subsets of Tregs. Indeed, recent data showed that Tregs are heterogeneous, with five major structurally and genetically distinct cell subsets, each representing a stage of maturation with distinct functional capacities, which could be proinflammation or tolerance (58).

In other tumor models such as breast cancer, TILs can be subdivided according to their location within the tumor: stromal TILs when located in the peritumoral space and intraepithelial TILs, when they have penetrated the tumor islets. This classification was widely used in breast cancer and is now recommended in clinical routine (45). Although these recommendations exist in breast cancer, there is no standardized approach to evaluate TILs in ovarian cancer. In this study, we decided to perform a representative HES slide, which had to contain both tumor and adjacent stroma, to evaluate overall immune cells whatever their localization in tumor. Nonetheless, it would be interesting to perform, in the future, additional studies to address whether a correlation exists between TIL distribution and m-CD95L expression or s-CD95L serum level. Because evaluation of tumor immune infiltration is difficult, the International Immuno-Oncology Biomarkers Working Group advocates evaluating intratumor lymphocytes in ovarian carcinoma using standard staining methods such as HES. They suggest a semiquantitative evaluation of the area occupied by inflammatory mononuclear cells (lymphocytes and plasma cells, excluding polymorphonuclear cells; ref. 45). These recent recommendations suggest the need for a more reliable method of evaluating tumor immune infiltration and accordingly, we examined TILs in the recommended way using IHC. The serum concentration of s-CD95L may be a reproducible and simple tool for evaluating tumor immune responses in patients with HGSOC. Furthermore, we found that s-CD95L is an independent prognostic factor of PFS in HGSOC (P = 0.0063). Patients with HGSOC with high level of s-CD95L show a good prognosis, which is opposite to that found for those with TNBC (59). This discrepancy may be due to differences in tumor load, disease history (metastasis for TNBC and carcinomatosis for HGSOC), and different roles played by s-CD95L. In TNBC, s-CD95L triggers a promotile signal in tumor cells, whereas in HGSOC it seems to contribute to the immune landscape through molecular mechanisms that remain to be elucidated.

Although we observed a correlation between expression of m-CD95L (assessed by IHC) in the tumor and serum s-CD95L levels in patients with HGSOC, the 8 patients lacking m-CD95L in the tumor still express s-CD95L in serum. This discrepancy could be due to intratumor heterogeneity; further studies are required to assess m-CD95L expression in a cohort from whom multiple biopsies are obtained from anatomically distinct sites. Another hypothesis could be that the membrane-bound ligand would already have been stripped from the endothelial cell surface.

Motz and colleagues showed that m-CD95L on endothelial cells in ovarian cancer selectively killed cytotoxic CD8 T lymphocytes, without affecting FoxP3⁺ regulatory T lymphocytes; they postulated that CD95L expression on tumor endothelial cells is a poor prognostic factor due to this mechanism, which generates a tolerogenic microenvironment (41). Our data suggest that the mechanism proposed by Motz and colleagues is not complete and that the CD95/CD95L system plays a more complex role in cancer. Indeed, we showed that s-CD95L is a good prognostic factor because it could contribute to drive antitumor immune responses; however, data obtained with m-CD95L confirm the results of Motz and colleagues, in that a higher number of FoxP3 T cells are accumulated in tumors of patients with endothelial expression of m-CD95L as compared with those in which CD95L is not detected. The metalloprotease that cleaves m-CD95L to generate s-CD95L is the missing link that prevents us from presenting a more complete overview of HGSOC prognosis based on the CD95/C95L system. Nevertheless, our findings support the utility of s-CD95L as a biomarker for selecting patients with HGSOC for checkpoint inhibitor treatment trial.

Serum concentration of s-CD95L is associated with an increased number of tumor-infiltrating FoxP3 T cells and a better prognosis in women with HGSOC. s-CD95L levels correlate with prognosis. An s-CD95L level > 516 pg/mL is a cut-off value for assessing tumor immune infiltration in patients with HGSOC and could be used to select patients for inclusion in clinical trials that evaluate checkpoint inhibitor treatment responses, such as inhibitors that block PD-1/PD-L1.

T. de La Motte Rouge is an advisor/consultant at Roche, AstraZeneca, Clovis Oncology, Tesaro, and has received speakers bureau honoraria from AstraZeneca and Tesaro. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. de La Motte Rouge, J. Corné, P. Legembre, V. Lavoué

Development of methodology: T. de La Motte Rouge, J. Corné, A. Cauchois, M. Le Boulch, B. Laviolle, A. Fautrel, P. Legembre

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. de La Motte Rouge, J. Corné, A. Cauchois, M. Le Boulch, C. Poupon, N. Rioux-Leclercq, V. Catros, J. Levêque, A. Fautrel, M. Le Gallo, V. Lavoué

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. de La Motte Rouge, A. Cauchois, C. Poupon, S. Henno, E. Le Pabic, B. Laviolle, M. Le Gallo, P. Legembre, V. Lavoué

Writing, review, and/or revision of the manuscript: T. de La Motte Rouge, M. Le Boulch, P. Legembre, V. Lavoué

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. de La Motte Rouge, J. Corné, M. Le Boulch, B. Laviolle, V. Catros, J. Levêque, P. Legembre, V. Lavoué

Study supervision: T. de La Motte Rouge, P. Legembre, V. Lavoué

The authors would like to thank Centre de ressources biologiques du CHU de Rennes, France and bioedits for editing the article. The project was supported by La Vannetaise, Canceropole grand ouest (PLASTICO), INCa PLBIO, Ligue Contre le Cancer, Fondation ARC, and ANR PRCE.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.