Lower expression of NINJ1 (Ninjurin 1), a mediator of plasma membrane rupture, is associated with advanced disease and worse prognosis in serous ovarian cancer

Gasdermin proteins (GSDMs) form pores in cell membranes upon various stimuli, leading to the release of certain proinflammatory molecules such as IL-1β and IL-18, and this ultimately results in pyroptotic cell death. NINJ1 (Ninjurin 1) has recently been identified as a cell membrane protein responsible for the final complete plasma membrane rupture following lytic cell death mechanisms including pyroptosis, causing the release of relatively larger molecules such as HMGB1 and LDH. In this study, we reported the presence of higher GSDMD and lower GSDME protein levels in ovarian tumors compared to surrounding non-malignant stroma in the tumor microenvironment. GSDME protein levels are also lower in the tumors of the omentum compared to adjacent stromal cells. We found that NINJ1 expression decreases from early to late stage in serous ovarian cancer, and the percentage of NINJ1 copy number loss events is the highest in ovarian cancer among other cancers. Moreover, we showed that low expression of NINJ1 is associated with shorter overall survival of patients with ovarian cancer. In support of the findings showing that low NINJ1 expression contributes to worse prognosis in this most lethal gynecological malignancy, NINJ1 expression was found to be lower in cisplatin-resistant ovarian cancer cells compared to cisplatin-sensitive counterparts in vitro. We suggest that the members of gasdermin family might have distinct functions in serous ovarian cancer, and low levels of NINJ1 might contribute, at least in part, to the progression and poorer prognosis of ovarian cancer. A complete picture of how pyroptosis and subsequent plasma membrane rupture are involved in ovarian cancer will be of high importance in order to identify actionable therapeutic vulnerabilities within this newly identified group of proteins.


Introduction
Ovarian cancer (OC) is the deadliest gynecological malignancy, responsible for more than 140,000 deaths in women worldwide each year [1]. The majority of ovarian cancer patients are diagnosed at an advanced stage with the presence of malignant ascites fluid and highly spread metastases in the peritoneal cavity and other organs [2]. The 90% of primary ovarian tumors are epithelial ovarian cancer (EOC) which comprise several histologic subtypes (histotypes): serous, mucinous, endometrioid, clear cell, transitional cell (Brenner tumors), mixed, and undifferentiated type. High-grade serous ovarian cancer (HGSOC) is the most common and lethal form of ovarian cancer [3,4]. The current standard of care for ovarian cancer patients includes cytoreductive surgery (i.e., the surgical resection of ovarian tumors) and primary chemotherapy with platinum-and taxane-based drugs such as cisplatin [5]. Although most patients initially respond well to this treatment regimen, more than 70% of the patients develop resistance to platinum drugs within 5 years leading to short life expectancy [5]. Ovarian cancer patients are also poorly responsive to new immunotherapy applications such as those with immune checkpoint inhibitors (ICI); therefore, these treatment options still remain insufficient to confer long-term survival benefit to patients with these cold tumors. Thus, a complete understanding of ovarian cancer immunobiology at the molecular level is urgently needed to develop more effective treatment strategies for those living with this malignancy.
Pyroptosis is a lytic form of programmed cell death with a highly pro-inflammatory nature [6][7][8][9]. This form of regulated cell death are executed by certain proteins, most of which are identified in the recent years [6][7][8][9]. Gasdermins (GSDMs) are members of a family of pore-forming effector proteins, leading to pyroptotic cell death upon activation by certain stimuli [6][7][8][9]. These proteins, excluding PJVK (Pejvakin; DFNB59), lead to membrane permeabilization [6][7][8][9]. The formation of pores in the cell membrane by gasdermins allow the release of multiple proinflammatory molecules below a certain threshold size, including mature IL-1β and IL-18 as well as small DAMPs (damage-associated molecular patterns) [9]. Gasdermin proteins contain a cytotoxic N-terminal domain which has an intrinsic poreforming activity in the plasma membrane, and a C-terminal domain which represses pore-forming activity of N-terminal domain, when an activating signal such as an infection is not present [6][7][8][9]. These N-and C-terminal domains are connected by a central flexible linker (except for PJVK). These linker regions can be cleaved by certain caspases upon induction by pathogen-derived or host-derived danger signals (such as pathogen-associated molecular patterns (PAMPs) and DAMPs, respectively). Following the release of intramolecular inhibition on N-terminal domain due to the proteolytic cleavage of linker region by these caspases, this fragment of gasdermin localizes into cell membranes to form large oligomeric pores causing the physical rupture of the cell membrane, which ultimately results in the disruption of ion homeostasis, in the induction of proinflammatory cell death and the release of certain proteins such the activated IL-1β and IL-18 [10][11][12][13]. The ultimate release of these cytoplasmic contents from pyroptotic cells provides signals enough to initiate an inflammatory cascade.
Different caspases are responsible for the specific cleavage of certain gasdermins within their central flexible linker regions, and thus for the activation of these proteins. These inflammatory caspases are activated by various inflammasomes upon induction by different stimuli: canonical inflammasomes including the NLRP3 inflammasome can activate caspase-1, whereas oxidized lipids and lipopolysaccharide (LPS) can activate mouse caspase-11 or human caspase-4 and -5 by forming noncanonical inflammasomes [10,[14][15][16][17][18]. Likewise, these caspases can only cleave specific gasdermin proteins at certain regions to activate them. For example, GSDMD should be cleaved at a certain amino acid position by inflammatory caspases including caspase-1, -4, -5, and -11 or by caspase-8, to be able to form pores in the cell membrane to promote plasma membrane rupture [10,[14][15][16][17][18]. Similarly, caspase-3 and -8 have been identified as proteases responsible for the cleavage of GSDME and GSDMC, respectively [19][20][21]. It should also be noted that the expression of the N-terminal pore-forming domain of GSDMD or of some other gasdermin family members is sufficient to induce pyroptotic cell death without the requirement of caspase activation [6,7].
So far, several member proteins belonging to the gasdermin family and some gasdermin-like proteins were identified based on sequence homology [9]. Gasdermin protein family in humans currently has six paralogous members: GSDMA, GSDMB, GSDMC, GSDMD, GSDME (DFNA5), and PJVK (Pejvakin, DFNB59) [9]. Among these proteins, GSDME and PJVK cluster together more closely relative to other members [9]. Many studies previously reported the involvement of these proteins in the initiation and progression of certain cancer types. The expression of GSDMA was shown to be negatively regulated in primary gastric cancers and gastric cancer cell lines [22]. This study named the first gasdermin gene and identified its expression in certain tissues in mice. Indeed, the name gasdermin was coined based on the specific expression of GSDMA proteins in the gastrointestinal tract and skin epithelium in mice (gas + dermin) [9,22]. GSDMB was found to be involved in tumor progression in multiple cancer types, namely, gastric cancer, hepatocarcinoma, cervix, and breast cancers [23,24]. Another study reported that the expression of GSDMC is upregulated in metastatic melanoma cells [26]. Using data for cancers in the upper gastrointestinal epithelium, others suggested that GSDMC, GSDMD, and GSDMA may be potential tumor suppressors and GSDMB, which was amplified and upregulated in some gastric cancers, could function as an oncogene, highlighting that each member of the gasdermin family can have distinct functions in these tissues [27]. Contrarily, GSDMC was reported to contribute to tumorigenesis in colorectal cancer, since its knockdown led to decreased proliferation in colorectal cancer cell lines [28]. Zhang et al. found that GSDME limits tumor growth by promoting pyroptosis [29]. It has been reported that pyroptosis activates anti-tumor immunity through the augmentation of the tumor cell phagocytosis by tumor-associated macrophages and also by boosting the number and activity of tumor-infiltrating natural-killer (NK) and CD8 + T lymphocytes [29]. Similar to GSDME, GSDMB was also found to increase anti-tumor immunity [30]. It has been shown that NK-or cytotoxic T lymphocyte-derived granzyme A (GZMA) cleaves GSDMB in GSDMB-positive cells to promote its pore-forming activity, leading to pyroptotic killing of target cancer cells [30]. They also demonstrated that the introduction of GZMA-cleavable GSDMB proteins into mouse cancer cells promotes the clearance of these tumors cells. Therefore, it can be stated that gasdermin-mediated pyroptosis might also be a mechanism by which cytotoxic lymphocytes kill tumor cells and that it may increase antitumor immunity in certain cancers.
Recently, we found that GSDMD and GSDMC show increased expression, and in contrast, GSDME and PJVK show decreased expression in serous ovarian cancer compared to normal ovaries, at the mRNA level [31]. We also showed that the percentage of copy number gain events in genes encoding GSDMC and GSDMD is around 50% in ovarian cancer patients, in parallel to their increased expression in ovarian cancer [31]. Moreover, the percentage of copy number variation (CNV, both gains and losses) events in genes encoding GSDMC, GSDMD, GSDME, and PJVK is the highest in ovarian cancer among all cancer types. In contrast to other gasdermins, GSDME is mostly characterized by copy number losses in ovarian cancer, again in line with its decreased expression in this cancer type. We suggested that since one in two patients with ovarian cancer have GSDMD and GSDMD copy number gains and the expression of these gasdermins is upregulated in serous ovarian cancer and that GSDME have both decreased expression and more copy number losses than gains, these gasdermins might be of high importance in the initiation of serous ovarian cancer.
Another recent report identified cell-surface protein NINJ1 (Ninjurin 1) as an important mediator of plasma membrane rupture in lytic cell death pathways including pyroptosis, and showed that NINJ1 is an essential protein for pyroptosis-related plasma membrane rupture [32,33]. It has been shown that NINJ1 −/− cells are able to release small molecules such as IL-1β and IL-18 through gasdermin pores; however, NINJ1 is required for the release of larger molecules including LDH (lactate dehydrogenase) and HMGB1 (high-mobility group box 1 protein), which are large DAMP molecules that propagate inflammation [32,33]. Ninj1 −/− cells release similar levels of IL-1β in response to either nigericin (a microbial toxin) or cytoplasmic LPS, and that NINJ1 is also dispensable for the release of IL-18, another pyroptosis-associated IL-1 family member, showing that the release of IL-1β and IL-18 from cells is independent of plasma membrane rupture by NINJ1 and probably occurs via GSDMD pores of approximately 18 nm in size [32]. Since plasma membrane rupture is a subsequent event following the initial formation of small pores in plasma membrane by certain gasdermins, and NINJ1, by using an evolutionarily conserved extracellular domain for oligomerization and acting downstream of these gasdermin pores, is required for subsequent plasma membrane rupture, NINJ1 also needs to be studied in the context of cell death in cancer. This is also important because the newly identified function of NINJ1 as a mediator of plasma membrane rupture overturns the idea that cell death-related plasma membrane rupture is a passive event. Cellular processes mediated by NINJ1 might represent later events in a cascade initiated by the function of gasdermins; therefore, it can be stated that functions of both NINJ1and gasdermin proteins are highly overlapping.
In the present study, we showed that GSDMD protein levels are increased in ovarian tumor compared to adjacent non-malignant stromal cells. In contrast, GSDME protein levels are decreased in ovarian tumor relative to surrounding normal stromal cells. These observations are in line with the findings we previously reported using transcriptome and CNV data. We also found that NINJ1 expression is decreased in late stage serous ovarian cancer compared to early stage, and that the percentage of NINJ1 copy number loss events is the highest in ovarian cancer among other cancer types. Moreover, we reported that low NINJ1 expression is associated with worse overall survival in ovarian cancer. In addition, we observed that cisplatin-resistant ovarian cancer cell lines have lower NINJ1 levels compared to cisplatinsensitive cells, again suggesting that decreased NINJ1 levels might contribute poorer disease outcomes in ovarian cancer. This study points that NINJ1 might be an important player in ovarian cancer initiation, progression, and therapy response, by possibly influencing anti-tumor immunity.

Gene expression datasets and transcriptome analysis
Following gene expression datasets from Gene Expression Omnibus (GEO; https:// www. ncbi. nlm. nih. gov/ geo/) were used in the present study: GSE24789 (n = 36) [34] and GSE15709 (n = 10) [35]. In GSE24789, expression profiling by array was performed by using RNA which was isolated from MOSE (mouse ovarian surface epithelial) cells, for which early cells represent a pre-neoplastic, non-malignant stage; intermediate cells represent a neoplastic, pre-invasive state; and late cells represent a malignant, invasive stage. Three biological replicates were used to take into account variations within the heterogeneous cultures, in this dataset [34]. The Cancer Genome Atlas (TCGA) data for serous ovarian cancer (n = 578) was loaded into R statistical computing environment using curatedOvarianData Bioconductor package [36,37]. This package contains multiple clinically annotated ovarian cancer transcriptome datasets besides TCGA-OV [36]. TCGA-OV dataset contains gene expression data only for ovarian cancer patients with serous histological type of epithelial ovarian cancer (EOC) [37]. The distribution of patients in terms of tumor stage and grade is as follows: early stage (n = 43), late stage (n = 520), low grade (n = 75), and high grade (n = 480). Expression data and clinical metadata were retrieved from large Expression Set objects using functions from Biobase R package [38].

Proteome data
Proteomics data for high-grade serous ovarian carcinoma (HGSOC) patients, acquired using liquid chromatography (LC)-mass spectrometry (MS) from formalin-fixed, paraffin-embedded (FFPE) specimens, were accessed from MaxQB -the MaxQuant DataBase of Max-Planck-Institut für Biochemie [39] (http:// maxqb. bioch em. mpg. de/ mxdb/ proje ct/ show/ 93730 12627 500). This dataset (n = 11 patients with HGSOC) contains protein levels for tumor cells and surrounding non-malignant stromal cells at four different anatomical locations in HGSOC patients (namely, omental metastasis, serous tubal in situ carcinoma (STIC), invasive fallopian tube (FT) lesions, and invasive ovarian lesions) [39]. Sample sizes (n) for each protein are as follows in this dataset: GSDMD (n = 94), GSDME (n = 73), and GSDMA (n = 46). In creating this dataset, authors collected tissues prospectively during the initial debulking surgery and they reported that all patients were chemotherapy-naive [39]. For each patient and all anatomic sites, both tumor and stromal compartments were microdissected, and proteins were then extracted using an optimized high-sensitivity, label-free proteomic workflow for low-input samples as stated in the original paper [39].

Copy number variation analysis
Copy number variation (CNV) data for ovarian cancer patients was accessed from National Cancer Institute Genomic Data Commons Data Portal (https:// portal. gdc. cancer. gov/) [40][41][42]. Cancer types were ordered based on the percentage of copy number gains or losses in NINJ1 gene from the highest to the lowest. These copy number alteration data for cancer patients were originally obtained in The Cancer Genome Atlas (TCGA) project.

Survival analysis
Kaplan-Meier plots were drawn to show the overall survival (OS) or recurrence-free survival (RFS) of ovarian or breast cancer patients (for all histotypes and only for serous histotype) with low and high expressions of NINJ1 using Kaplan-Meier Plotter tool (n = 1656) [43,44] (http:// kmplot. com/ analy sis/ index. php?p= servi ce& can-cer= ovar). We selected JetSet best probe set indicated in green as recommended, and we otherwise used the default values. We did not restrict the analysis to subtypes and treatment groups, and used the default parameters in the tool. Logrank p and false discovery rate (FDR) values were taken into account in the comparison of survival rates between low-and high-expression cohorts. Hazard ratio (HR) and logrank p values were given in the top right corner of each Kaplan-Meier plot.
We also calculated HR values for NINJ1 expression in ovarian cancer for different datasets independently and combined (overall) using curatedOvarianData Bioconductor package, after adjusting for the success of debulking surgery (debulking status defined as residual tumor smaller than 1 cm following cytoreduction surgery) and The International Federation of Gynecology and Obstetrics (FIGO) stage [36]. A forest plot (blobbogram) was drawn using these data [36].

Higher GSDMD and lower GSDME protein levels in ovarian tumors compared to surrounding non-malignant stroma
We previously showed that GSDMD is upregulated and GSDME is downregulated in ovarian cancer compared to normal ovaries, at the mRNA level [31]. In the current study, we found that, similar to the findings at the mRNA level, GSDMD expression increases and GSDME expression decreases in ovarian tumors compared to surrounding normal stroma at the protein level (Fig. 1). We did not observe any significant change in the protein levels of GSDMA between ovarian tumors and adjacent normal cells. This is in parallel to our previous observation that GSDMA mRNA levels do not change between ovarian tumors obtained from patients and healthy ovaries [31]. Also note that, in addition to that was observed in ovary, GSDME protein levels are lower in tumor cells relative to adjacent non-malignant stromal cells in the omentum (Fig. 1, middle plot).
Data for the other three members of the gasdermin family (namely, GSDMB, GSDMC, and PJVK) is not available in this proteomics dataset; therefore, they could not be analyzed in this study.

NINJ1 expression decreases from early to late stage in serous ovarian cancer
A recent study showed that gasdermin pores formed in the plasma membrane are not sufficient for the complete membrane rupture in pyroptosis, and NINJ1 mediates the final cataclysmic event in lytic cell death mechanisms including pyroptosis [32,33]. It has been reported that cells lacking NINJ1 are unable to release various intracellular proteins with larger sizes such as LDH (a standard measure of plasma membrane rupture) and HMGB1 (a known damageassociated molecular pattern [DAMP]); however, they are able to release smaller molecules such as IL-18 through gasdermin pores [32]. They concluded that NINJ1is essential for pyroptosis-related plasma membrane rupture [32].
We previously found that NINJ1 mRNA levels are higher in serous ovarian cancer compared to normal ovaries [31] (Supplementary Fig. 2; p ≤ 0.05). This time, we analyzed changes in NINJ1 expression during cancer progression from early to late stage in serous ovarian cancer, using patient transcriptome data from TCGA-OV project. We observed that NINJ1 expression is lower in late stage compared to early stage in serous ovarian cancer at the mRNA level (p = 0.027, n = 578) ( Fig. 2A). Since cancer stage describes the size of a tumor and how far it has spread from its primary site, decreased NINJ1 levels might be contributing, at least to a certain extent, to increased tumor cell proliferation and higher metastatic potential in serous ovarian cancer.
We also compared NINJ1 levels between mouse ovarian surface epithelial (MOSE) cells at different stages of malignancy, as these cells transition from a pre-neoplastic to a malignant state (early cells: pre-neoplastic, non-malignant stage; intermediate cells: neoplastic, pre-invasive state; late cells: a malignant, invasive stage). We found that NINJ1 expression first increases from pre-neoplastic to neoplastic state, but then it decreases from neoplastic state to malignant and invasive state ( Fig. 2B; each subplot shows data obtained using a different probe for NINJ1 whose IDs were indicated at the top of each subplot). This in vitro data is in line with our previous observation showing that NINJ1 levels are increased in ovarian cancer compared to normal ovaries [31] and also with data presented in Fig. 2A showing that NINJ1 levels are decreased from early to late stage in ovarian cancer. In other words, NINJ1 mRNA levels seem to first increase during cancer development and then decrease during cancer progression (for instance, from early to late stage), pointing to possibility of highly dynamic regulation of its expression at the mRNA level during cancer initiation and progression.  Furthermore, we showed that the correlation between GSDMD and NINJ1 expression at the mRNA level in ovarian tumors (R = 0.3; first subplot) is higher than that in healthy ovaries (R = 0.11; second subplot) (Fig. 3). We also reported that these two proteins have not been found to interact based on currently available data in two independent protein-protein interaction databases, namely, BioGRID and STRING databases (Supplementary Fig. 1).

The percentage of NINJ1 copy number loss events is the highest in ovarian cancer among other cancers
We found that the percentage of NINJ1 copy number loss events (8.717%) is the highest in patients with ovarian cancer compared to those with other cancer types (Fig. 4A). Surprisingly, the percentage of copy number gain events in NINJ1   [56] gene in ovarian cancer (5.128%; third from top) is also among the highest within patients with cancer (Fig. 4B). The fact that the percentage of both copy number losses and gains is high in ovarian cancer compared to other cancers can possibly be explained by our two previous observations. We suggest that NINJ1 copy number increases might be responsible for the development of ovarian cancer in some patients; however, NINJ1 copy number losses might be responsible for the progression of ovarian cancer in others. If this is the case, how NINJ1 might contribute to tumor initiation and how it might negatively regulate tumor progression in ovarian cancer is currently unknown and further research is needed.

High expression of NINJ1 is associated with better overall survival in patients with ovarian cancer
Next, we compared overall survival (OS) of ovarian cancer patients with low or high expression of NINJ1. We found that high expression of NINJ1 is associated with better overall survival in ovarian cancer patients (all histotypes combined; Fig. 5, first panel; logrank p = 4e − 06, hazard ratio (HR) = 0.71) and in serous ovarian cancer patients (Fig. 5, second panel; logrank p = 0.0024, HR = 0.78). In ovarian cancer patients (all histotypes combined), median overall survival of NINJ1 low expression cohort was 33.77 months, compared to 48 months in NINJ1 high-expression cohort (difference of 11.23 months) (Fig. 5, last panel). Specifically in patients with serous histological type of ovarian cancer, median overall survival of NINJ1 low-expression cohort was 38.4 months, compared to 45.77 months in NINJ1 high-expression cohort (difference of 7.37 months) (Fig. 5, last panel). Furthermore, we analyzed the association of NINJ1 expression with overall survival in patients with ovarian cancer, after adjusting for the success of debulking surgery (debulking status defined as residual tumor smaller than 1 cm following cytoreduction surgery) and The International Federation of Gynecology and Obstetrics (FIGO) stage, using 15 independent datasets with applicable expression and survival information. The forest plot in Fig. 4 shows that overall hazard ratio (HR) for NINJ1 is significantly lower than 1 (0.87, p = 4.885028e − 07). This indicates that patients with high NINJ1 levels have better outcome. Therefore, this analysis indicates that NINJ1 can be considered a prognostic of overall survival in patients with ovarian cancer (Fig. 5). Since we observed that NINJ1 expression is lower in late stage compared early stage serous ovarian cancer (Fig. 2C) and that low NINJ1 expression is associated with worse survival in this patient group, we can speculate that low NINJ1 expression might contribute, at least to a certain extent, to shorter survival of patients with late stage ovarian cancer. Mechanistic studies are required to confirm these observations.  5 High expression of NINJ1 is associated with better overall survival in patients with ovarian cancer. Overall survival (OS) of ovarian cancer patients with low (black lines) or high (red lines) expression of NINJ1. First plot shows data for patients with all histological types (histotypes) of ovarian cancer, and second plots shows data only for patients with serous histotype of ovarian cancer. Third plot shows median overall survival of the indicated cohorts in months. First two plots were drawn using Kaplan-Meier Plotter tool (n = 1656) [43,44] (http:// kmplot. com/ analy sis/ index. php?p= servi ce& cancer= ovar). We did not restrict the analysis to disease subtypes and treatment groups, and used the default parameters in the tool. Last panel: A forest plot showing the association of NINJ1 expression with overall survival in patients with ovarian cancer after adjusting for the success of debulking surgery (debulking status defined as residual tumor smaller than 1 cm following cytoreduction surgery) and Federation of Gynecology and Obstetrics (FIGO) stage, using datasets with applicable expression and survival information (the names of the datasets were given in the left) [57]. HR: hazard ratio Similar to that observed in ovarian cancer, we found that lower expression of NINJ1 is also associated with poor prognosis in breast cancer, another cold tumor in women that is unlikely to trigger a strong immune response (Fig. 6).

NINJ1 expression is lower in cisplatin-resistant cells compared to cisplatin-sensitive ovarian cancer cells in vitro
Since we observed that NINJ1 mRNA levels are lower in late stage ovarian cancer, and that low NINJ1 expression is associated with shorter overall survival of ovarian cancer patients, we wanted to see if lower NINJ1 expression is also associated with other parameters which contribute to worse prognosis. We found that cisplatin-resistant A2780 ovarian cancer cells have decreased NINJ1 expression compared to cisplatin-sensitive A2780 cells (Fig. 7; p = 0.03, n = 10). Thus, it can be inferred from this data that decreased NINJ1 expression might also contribute to resistance to chemotherapeutics such as cisplatin; however, further in vivo studies are required. Lower survival rates observed in patients with low NINJ1 expression (thus increased resistance) might also be explained, at least in part, by their decreased response to drugs such cisplatin which is used in the standard treatment of ovarian cancer patients. However, further in vivo studies in animal models and clinical samples will be of high importance to confirm this in vitro observation.

NINJ1 expression is positively correlated with immune infiltration by macrophages and monocytes in ovarian cancer
Lastly, we showed that NINJ1 mRNA levels are mostly positively correlated with the infiltration of ovarian tumors by macrophages and monocytes, using TIMER2.0 tool (Spearman's p > 0, p < 0.05) (Fig. 8). This data points that lower expression of NINJ1 in advanced ovarian cancer might lead to decreased immune infiltration and thus might result in worse outcome due to lower anti-tumor immunity. In other words, NINJ1 might affect the immunogenicity of ovarian cancer, influencing survival rates in patients with this cold tumor. Also note that NINJ1 expression is positively correlated with the infiltration of tumors by these two immune cell types in most other cancer types, potentially showing that NINJ1 might have broader implications in cancer.

Discussion
In the current study, we showed that both GSDMD and GSDME expressions at the protein level differ between ovarian tumors and adjacent normal stromal cells in the tumor microenvironment (Fig. 1). GSDMD protein levels are higher in ovarian tumors, whereas GSDME protein levels are lower, compared to adjacent non-malignant cells (Fig. 1). This is in line with our previous observations at the mRNA level [31]. We also formerly showed that GSDMD copy number gains are present in almost half of the patients with ovarian cancer, and that more than 50% of total CNV events in GSDME gene are copy number losses [31]. Therefore, we propose that GSDMD and Partial_Cor GSDME might have opposite functions in ovarian cancer. Increased expression of GSDMD might contribute, at least in part, to the development of ovarian cancer; however, higher expression of GSDME might negatively regulate ovarian cancer initiation. Also, in parallel to the findings at the mRNA level [31], GSMDA protein levels seem to not change between normal and malignant ovaries (Fig. 1). This indicates that not all gasdermin family members are necessarily of equal importance in the context of ovarian cancer. Since NINJ1 has been recently identified as a protein responsible for complete plasma membrane rupture in pyroptosis in addition to other cell death pathways [32,33], we studied its expression during cancer progression in serous ovarian cancer. We found that its expression is decreased in late stage compared to early stage in serous ovarian cancer (Fig. 2). In the late stage, tumor size is larger and it has already spread from where it originated [61]. In parallel, we showed that its low expression is associated with worse overall survival of patients with ovarian cancer (in addition to patients with breast cancer, another cold tumor in women), and that NINJ1 can be considered a prognostic marker for better outcome in patients with ovarian cancer (Fig. 5). In support, we observed that cisplatin-resistant ovarian cancer cells have lower NINJ1 expression in vitro, compared to cisplatin-sensitive cells, and NINJ1 expression is positively correlated with immune infiltration of ovarian tumors by macrophages and monocytes, possibly explaining why its lower expression is associated with worse prognosis (Fig. 7  and 8). All these data points that ovarian cancer cells might acquire the ability to enhance proliferation or inhibit cell death, to promote metastasis, and to develop resistance to cisplatin by decreasing NINJ1 levels, thus possibly blocking complete plasma membrane rupture and ultimately limiting anti-tumor immunity which otherwise might be induced by the released cytoplasmic contents. How NINJ1 mechanistically contributes to worse prognosis is currently not known and further research is needed.
Besides, we found that the correlation between gasdermin D (GSDMD) and NINJ1 mRNA expression is higher in ovarian tumors compared to that in healthy ovaries (Fig. 3), possibly indicating their coordinated involvement in ovarian cancer initiation or progression. Since NINJ1 functions downstream of gasdermin pore formation by using its evolutionarily conserved extracellular domain for oligomerization to promote subsequent plasma membrane rupture [32,33], cooperative mechanism of these two proteins in cell death might require the coordinated regulation of their expression. Further research is needed to understand whether expression of GSDMD and NINJ1 is regulated by the same upstream proteins in the context of ovarian cancer.
Furthermore, we recently found that NINJ1 expression is higher in ovarian tumors obtained from patients compared to normal ovaries [31]. Therefore, it seems that NINJ1 contribute differently to tumor initiation and progression. In support of this, we showed in the present study that NINJ1 expression first increases from pre-neoplastic to neoplastic state, but ultimately it decreases from neoplastic state to malignant and invasive state, in mouse ovarian surface epithelial (MOSE) cells at different stages of malignancy (Fig. 2) [34]. We suggest that during cancer progression from early to late stage, cancer cells might attempt to suppress plasma membrane rupture by negatively regulating NINJ1 expression to a certain extent to be able better survive in blood and new sites in the body during the process of metastasis. In late stage, they can also acquire the ability to partially inhibit plasma membrane rupture upon treatment with chemotherapeutics. In other words, during ovarian cancer initiation and progression, the expression of NINJ1 might be dynamic and highly regulated. A better understanding of NINJ1-related mechanisms in tumor progression and drug resistance in ovarian cancer is needed.
Finally, we observed that the percentage of NINJ1 copy number loss events is the highest in ovarian cancer patients among those with other cancers (Fig. 4). This group of patients may represent ovarian cancer patients with advanced disease. The percentage of NINJ1 copy number gain events is also high in ovarian cancer patients (top third among all cancer), and this group of patients may represent patients with early stage disease. A complete study with a larger sample size on NINJ1 CNV events based on tumor stage and grade is required to confirm these observations.
In summary, certain members of gasdermin family and NINJ1 might be important players in tumor initiation and progression in serous ovarian cancer. These proteins participating in different steps of pyroptotic cell death might have complementary functions in this highly lethal malignancy. A complete picture of how pyroptosis and ultimate plasma membrane rupture are involved in ovarian cancer will be of high importance in order to identify potential therapeutic targets within this recently identified group of pyroptotic proteins. We suggest that certain proteins in this proinflammatory cell death mechanism may represent actionable therapeutic vulnerabilities.