To investigate the potential influence of MTs on the immune response in EOC, 24 tumor samples were taken from a previously well-established patient cohort (32, 33). An overview of the working procedures of the entire study is given in Figure 1 . Furthermore, the study was conducted retrospectively.

All tumor samples were collected at the Institute of Pathology (University Hospital Essen, Germany), where they were routinely deposited in the form of formalin-fixed, paraffin-embedded (FFPE) tumor blocks. An experienced pathologist (AB) reviewed and re-evaluated the initial diagnoses. Re-evaluation was performed on tissue sections stained with hematoxylin and eosin. All tumor samples were confirmed as HGSOC.

Treatment of patients was conducted at the University Hospital of Essen within a span of 5 years (2005–2010). Primary treatment consisted of cytoreductive surgery, followed by six cycles of adjuvant platinum-based chemotherapy. Table 1 displays the general patient and tumor characteristics in the study cohort. Patients’ age at diagnosis ranged from 23 to 84 years, resulting in a mean age of 62 years across the cohort [standard deviation (SD) = 13.08]. Disease-free survival (DFS) varied between 4 months and 80 months, whereas overall survival (OS) fluctuated between 11 months and 126 months. Mean DFS and mean OS were 20.26 months (SD = 19.99 months) and 46.17 months (SD = 32.68), respectively. The observation period ended in October 2016. By then, 19 of the 24 patients had died (79.1%). The tumor recurrence rate among patients was rather high (83.3%).

The procedure of RNA extraction, integrity analysis, and digital quantification of mRNA expression has already been described in a previous study (33).

FFPE samples were utilized for RNA isolation (two sections of 10-µm thickness each). Isolation procedures were performed semi-automatically on the Maxwell purification system (Maxwell RSC RNA FFPE Kit, AS1440, Promega) as specified by the manufacturer. RNA was eluted in 50 µl of ribonuclease (RNAse)-free water and quantified using RNA broad-range assay on a Qubit 2.0 fluorometer (Life Technology). Finally, a fragment analyzer (Advanced Analytical Inc., Ames, IA, USA) was used to evaluate RNA integrity. For this purpose, the DNF-489 standard sensitivity RNA analysis kit was utilized.

Digital gene expression analysis was performed on the NanoString nCounter platform, using the dedicated PanCancer Immune Profiling panel. The panel consisted of 30 reference and 770 target genes associated with various immune processes and pathways. Minimum sample input was 100 ng of RNA for each reaction. Probes were hybridized on the nCounter Prep-Station (MAX/FLEX System) using the high-sensitivity protocol. Afterward, the cartridge was scanned on the Digital Analyzer (NanoString) at 555 fields of view.

NanoString data processing, the subsequent data analysis, and statistical analyses were performed in R (v4.0.2). Data were processed utilizing the NAPPA and NanoStringNorm package (34).

In the beginning, a technical normalization was performed in a two-tier process. First, the mean counts of the panel inherent negative controls plus two times the standard deviation were subtracted from all counts as technical background correction. Inherent positive controls, presented as dilutional series of artificial oligonucleotides, were used for determination of the dynamic range. Subsequently, a biological normalization using a normalization factor calculated from the geometric means of the included reference genes was performed. Actual expression signals were differentiated from basal noise by applying a one-sided t-test between the expression of the inherent negative controls and each gene. A p-value below 0.05 was considered as an actual signal.

Systematic immunohistochemical analysis was performed on tissue micro-arrays. Blocks were cut into slices with a thickness of 4 µm. To take intratumoral heterogeneity into consideration, three distinct tissue cores (2 mm in diameter) from different tumor areas were used. Tissue staining was done automatically on the Ventana BenchMark ULTRA system (Ventana Medical Systems, Arizona, USA). Staining was performed according to the standard protocols of the manufacturer. Regarding the validation process, pleura samples served as negative control, whereas myoepithelial cells deriving from healthy breast tissue were used as reference tissue. MT-I and MT-II were stained using the clone E9 (Dako/Agilent, Santa Clara, CA, USA) as the primary antibody, because the targeted epitope is shared between the two MT isoforms (35). Samples were pretreated with a sodium citrate buffer for the purpose of antigen retrieval (0.01 M, pH 6). During pretreatment, samples were placed in hot water (90°C) with antigen retrieval solution for 30 min. The primary antibody was incubated at room temperature with a dilution factor of 100 and incubated for 30 min. For secondary (horseradish peroxidase (HRP)-conjugated) antibody application, the DAKO EnVision TM+ Visualization System, Peroxidase kit (Dako Cytomation, Cat. No. K4004) was utilized according to the manufacturer’s instructions.

A conventional light microscope (Nikon Eclipse 80i, Nikon Ltd. Düsseldorf, Germany) was used for evaluation of MT-I/MT-II expression by a board-certified pathologist. MT immune expression was quantified by a semi-quantitative scoring system. It consisted of four tiers based on the percentage of positively stained tumor cells, either in the nucleus or in the cytoplasm. The staining intensity was not taken into consideration. The cutoffs for the four tiers were expression in 0%, 1%–5%, 6%–50%, and, finally, more than 50% of tumor cells. The tiers were named as Score 0 (no signal), Score 1 (weak expression), Score 2 (moderate expression), and Score 3 (strong expression), respectively (35, 36).

In addition, slides were stained using antibodies against programmed death ligand 1 (PD-L1), CD3, CD4, and CD8 to analyze immune cell infiltration. Tissue samples were deparaffinized using xylene. For antigen retrieval, slides were incubated in cell conditioning solution 1 (CC1)/ethylenediaminetetraacetic acid buffer for 32 min, 40 min, 32 min, and 40 min, respectively. The samples were then incubated with the primary antibody for 32 min, 24 min, 32 min, and 24 min at room temperature. The dilution factors were 100, 200, 100, and 150 for the respective targets. Staining was performed according to the standard procedures as described above. The OptiView 3,3′-diaminobenzidine detection kit (760-700, Roche) was used for visualization. Slides were scanned using an Aperio ScanScope AT2 platform (Leica biosystems, Wetzlar, Germany) and analyzed with QuPath (v.0.2.0-m2, qupath.github.io/) as cells/mm². A list of all used antibodies and conditions is given in Table 2 .

As described in Section 2.3, graphical and statistical analyses were performed in R (v.4.0.2).

Whether data resembled normal distribution patterns was examined by applying Shapiro–Wilks test. Rejection or acceptance of the null hypothesis resulted in applying a non-parametric (Wilcoxon rank sum test and Kruskal–Wallis test) or parametric statistic test (Students t-test and ANOVA) to evaluate variable dependencies, respectively.

Ranked parameters with exactly two groups were subjected to Fisher’s exact test, which utilized double-dichotomous contingency (2 × 2) tables. In case at least one of the ranked parameters exceeded the group size of 2, Pearson’s Chi-squared test was applied. Correlations between two metric variables were analyzed by Pearson product–moment correlation coefficient and Spearman’s rank correlation test, respectively.

Various candidate patterns were identified by clustering methods (supervised and unsupervised) or principal component analysis. Gene set enrichment analysis (GSEA) was performed using the WEB-based Gene SeT AnaLysis Toolkit (WebGestalt) website (37). Correlations between a specified variable and the expression of genes in certain signaling pathways were visualized by network maps obtained from the KEGG (Kyoto Encyclopedia of Genes and Genomes) database. Group-based differences were visualized by −log2 fold changes of median or mean counts, depending on their distribution.

OS was calculated using Cox proportional hazard models. Hazard ratios with appropriate 95% confidence intervals, as well as p-values calculated by the Wald test, the score (log-rank) test, and the likelihood ratio test, were used as statistical measures.

To overcome multiple testing variances, p-values were corrected through application of false discovery rate. The cutoff for statistical significance was set at p ≤ 0.05 after adjustment.

Overall, MT immunoexpression could be detected in about 43% (10/23) of all HGSOC samples ( Figure 2A ). Moreover, those positive samples showed positivity only in a minor proportion of tumor cells, ranging between 5% and 20% ( Figure 2B ).

In addition to overall positivity, we also analyzed the subcellular distribution of MTs. Of the 10 positive tumor samples, only one sample showed an exclusive nuclear staining pattern, whereas the majority of cases (n = 7) showed cytoplasmic MT expression. Only two samples showed up with both nuclear and cytoplasmic staining. Interestingly, the subcellular localization seems to be of importance for action and activity of MTs. The sample with nuclear-only staining showed only low immunoexpression, whereas immunoexpression rose with the proportion of MT staining from cytoplasmatic to both nuclear and cytoplasmatic ( Figures 2C , 3 ).

Overall, the number of positively MT-stained cells shows a significant influence on OS in the recent cohort [likelihood ratio test, p = 0.0262; score (log-rank) test, p = 0.0569; Wald test, p = 0.0478]. In particular, when binarized in expressing/non-expressing cases, the calculated significance further rises [likelihood ratio test, p = 0.0144; score (log-rank) test, p = 0.0129; Wald test, p = 0.0084], showing prolonged survival in positively stained tumors. Not only the five-year survival rate (50.0% vs. 26.3%; odds ratio, 2.66) but also the 3-year survival rate (75.0% vs. 47.4%; odds ratio, 3.17) was notably improved between both groups ( Figure 2D ).

Moreover, progression-free survival (PFS) shows the same association to MT protein expression levels—a shortened PFS in MT-negative samples [likelihood ratio test, p = 0.0222; score (log-rank) test, p = 0.0503; Wald test, p = 0.0345] ( Figure 2E ).

Importantly, subcellular spatial distribution/localization seems to be an important prognostic factor in this context. The present data show that the survival-dependent effects are only based on those cases showing cytoplasmic or combined cytoplasmatic/nuclear immunoexpression of MTs [likelihood ratio test, p = 0.0101; score (log-rank) test, p = 0.0118; Wald test, p = 0.0076].

In the present cohort, we found high gene expression levels of certain immune response pathways associated with MTs. GSEA revealed that all gene sets significantly enriched in MT-I/I– positive samples comprise immune-associated pathways, including T-cell receptor signaling pathway, T_H17 cell differentiation, interleukin-17 (IL-17) signaling pathway, or cell-adhesion molecules. An overview of the GSEA is illustrated in Figure 4A ; a list with the complete statistical metrics is given in Figure 4B . Specific enrichment plots for each of the aforementioned pathways are displayed in Figure 4C and Supplementary Figure 1 .

In addition to the pathways refered by the GSEA, we have a specific look on those pathways that are important for an anti-tumor immune response, such as T-cell receptor signaling ( Figures 4C, D ), B-cell receptor signaling ( Supplementary Figure 2A ), NK cell–mediated toxicity ( Supplementary Figure 2B ), leucocyte transendothelial migration pathways ( Supplementary Figure 3A ), and IL-17 signaling ( Supplementary Figures 1B , 3B ). Furthermore, MT immunoexpression seems to strongly affect phosphoinositide 3-kinases (PI3K) signaling ( Supplementary Figure 4A ).

In particular, T-cell receptor signaling ( Figures 4C, D ) shows overexpression of most surface receptor molecules, including co-inhibitory PD1, cytotoxic T-lymphocyte-associated protein 4, and co-stimulatory signals, e.g., inducible T-cell costimulator, CD28, and CD40L. In addition, an upregulation of the different CD3 isoforms as well as CD4 and CD8 can be observed. This could be also shown immunohistochemically, with a high correlation between the respective markers and the infiltration density (cell/mm²) of the respective cell type (CD4, p = 0.0004, rho = 0.46; CD8, p < 0.0001, rho = 0.66; Figure 5A ). On a group level (high vs. low infiltration density defined by mean cells/mm²), Fisher’s exact test reveals a significantly higher infiltration with overall T cells (CD3, p < 0.0001), especially not only CTC (CD8, p = 0.0005) but also helper cells (CD4: p = 0.0237), all without exactly computable odds ratios due to the lack of high infiltration samples in the MT-negative section. An overview of the exact distribution metrics can be found in Figures 5B–E .

Furthermore, downstream signaling cascades including lymphocyte-specific protein tyrosine kinase, roto-oncogene tyrosine-protein kinase Fyn, zeta-chain-associated protein kinase 70, IL-2–inducible T-cell kinase, nuclear factor of activated T cells and Jun proto-oncogene, AP-1 transcription factor subunit (AP1), as well as resulting IL-10 and interferon-gamma (IFNγ) expression levels generated significantly higher counts in the MT-positive compared with the negative group.

Despite of the high transforming growth factor beta (TGFβ)/IFNγ levels, Suppressor of Mothers Against Decapentaplegic (SMAD2/3) seem to be downregulated in MT expressing samples, indicating no overall activation of canonical TGFβ signaling ( Supplementary Figure 4B ). Contrarily, PI3K signaling pathway as an alternative downstream cascade (a part of the non-canonical TGFβ signaling) showed a strong induction in MT-positive cases, including not only different types of activating surface molecules but also PI3K and Rho family-alpha serine/threonine-protein kinase (AKT1) itself as well as additional different AKT1 downstream targets ( Supplementary Figure 4A ).