Deciphering tumour microenvironment and elucidating the origin of cancer cells in ovarian clear cell carcinoma

This study’s approval was obtained from the SingHealth Centralized Institutional Review Board, encompassing the National Cancer Centre Singapore, Singapore General Hospital, and Kandang Kerbau Women’s and Children’s Hospital (CIRB 2015/2595). Tissue samples from ovarian cancer patients with informed consent were obtained ex vivo and dissected freshly by the pathologist. Samples were collected in RPMI 1640 medium and delivered on ice to the laboratory within 1 hour of tissue collection. Tissue samples of ovarian tumours were cut into ~1–2 mm 3 pieces using sterile scalpels and digested with collagenase type IV (STEMCELL Technologies) for 2 hours at 37°C on a rotator. Cell suspensions were filtered through a 70 mm cell strainer and centrifuged, and red blood cells (RBC) in the cell pellets were removed through RBC lysis. Cells were washed with PBS and snap frozen with liquid nitrogen and stored at −80°C until ready for single-cell analysis experimental use. The sample was dissociated, and the single-cell suspension was prepared by trypsin treatment for 2–3 hours at 37 C. The single-cell suspension was processed for single-cell analysis using 10× genomics single-cell RNAseq protocol. The single-cell library was sequenced using the Hiseq 4000 Illumina sequencer. Cellranger v7.1 was used to map the 10X scRNA-seq data to GRCh38 v32 human reference genomes. The resulting count data was analysed using Seurat v4.3 73 . The cells were filtered out if the cells contained less than 500 genes or greater than 25% mitochondria reads. The cells with UMI counts less than 0.01% quantile and greater than 95% quantile across all cells were removed. The genes were filtered out if expressed in less than 10 cells. The genes with log10 of average UMI count across all cells greater than −2.5, genes with at least 2 UMIs and genes detected in at least 10 cells were used for downstream analysis. DoubletFinder v2.0.3 74 was used to identify and remove doublets. Next, to annotate the cell types, we used scATOMIC 26 a machine-learning cell annotation method trained using public pan-cancer scRNA-seq data. We additionally manually curated the annotation using well-known marker genes for each cell type. The gene count for each tumour sample was normalised using SCTransform 75 and dimension reduction was performed using PCA and UMAP. The seven samples were integrated using the Pearson residuals integration method from Seurat SCTransform integration, resulting in 13,543 cells. We integrated publicly available endometriosis datasets 24 , 25 with 93,766 and 65,903 single cells to compare with our CCC data using the Pearson residuals integration method from Seurat SCTransform integration. Pairwise differential expression analysis was performed using MAST v1.12.0 76 on genes expressed in at least 0.25 fractions of the cells. Genes are differentially expressed if a minimum log fold-change threshold is greater than 0.25 and a p-value is less than 0.05. The gene set enrichment analysis was performed in clusterProfiler 77 on the identified differentially expressed genes. Using Spearman correlation, similar epithelial cell types across all endometrial conditions and CCC were identified. Then, pseudo-time trajectory analysis was carried out using Slingshot 78 , 79 on the PCA projection and CellRank 78 on the UMAP projection. Diffusion pseudotime was calculated to determine the genes that change across stages of cancer. The changes in overall signalling pathway communication between different cell types present within the CCC TME and other endometrial conditions were computed and visualised as dot plots and chord plots using CellChat 80 . The SNPs were detected in the scRNA-seq data using scAllele 81 . The identified SNPs were analysed and visualised using Maftools 82 . Drugs targeting the TME in CCC at different stages of cancer and different cell types were computed using drug2cell 63 based on the enrichment of drugs for the transcriptional signatures.

The samples of surgically resected ovarian CCC tissue from seven patients of different disease stages, including FIGO stages 1A, 2A, 2B, 3A and 4 were collected ( Figure 1a , Table S1). The H&E staining of the CCC samples showed that these samples have abnormal morphologies with multilobed nuclei (Figure S1a). Of these, three patients were in stage 1 with cancer in a single ovary, two patients were in stage 2 where cancer has spread to the bladder or bowel, one in stage 3 where cancer spread into the ovary and one in stage 4 where cancer has metastasised and spread to other organs. Additionally, three of these seven patients were diagnosed with endometriosis along with CCC and are referred to as CCC-MD ( Figure 1a ). To understand the cellular composition within the tumour sample, we performed high-resolution scRNA-seq on the seven patient tissues and integrated the cells across the patients ( Figure 1b ). Following quality control, we obtained 13,543 cells across these CCC samples to dissect the tumour microenvironment (TME) (Figure S1b-e). To determine the cell types in the tumour samples, we used scATOMIC 26 , a model trained with the transcriptional profiles of the cell types annotated from publicly available scRNA-seq datasets on TMEs across 19 common cancers, including ovarian cancer. Using scATOMIC and manual curation, we annotated the different cell types in the CCC TME ( Figure 1c ). We observed 19 cell types in the CCC, including epithelial cells, endothelial cells, fibroblasts, neurons, and immune cells. The cells annotated as cancer cells have higher expressions of malignancy markers obtained from the CancerSCEM 27 database ( Figure 1d ). Consistent with the previous studies, cells annotated as B cells express CD79A , CD19 , and MS4A1 28 , exhausted T cells express PDCD1 29 , Treg cells express FOXP3 30 . Macrophages in CCC samples were separated into two groups, M2 macrophages and macrophages based on the high expression of CD86 and low expression of CCL5 31 . Further, cancer-associated fibroblasts were annotated based on the expression of ACTA2 and TAGLN 32 (Figure S1f). The epithelial and cancer cells showed higher KRT8 , KRT18 , KRT19 and ANXA4 expression 27 . Subsequently, differential gene ontology biological processes indicated that the cancer cells had enrichment for cell division and proliferation-related terms compared to the epithelial cells, and the cancer-associated fibroblasts showed enrichment of cell-cell assembly-related gene ontology biological processes ( Figure 1e , S1g). ARID1A and PIK3CA are well-known oncogenes associated with CCC 33 , and here, we also found a higher mutation rate at these genes across the samples (Figure S2a-b). Based on cell population, a significantly higher proportion of immune cells were present in all the CCC tumours compared to epithelial or fibroblasts, indicating an immune infiltration at the tumour site ( Figure 1f ). Henceforth, we sought to decipher the cell-cell communications within the TME of the collected patient tumour samples. The top signalling pathways active in the TME across all samples were pathways such as MIF, TGFb, PARs, SPP1 and CXCL (Figure S2c). The cancer cells in CCC predominately secreted ligands for pathways such as MIF and SPP1 ( Figure 2a ). The cancer cells and cancer-associated fibroblasts produced the MIF ligands, signalling to the immune cells such as macrophages and T cells ( Figure 2b ). Interestingly, the surrounding normal epithelial cells in CCC are not involved in MIF - CD74 signalling and are cancer-specific. MIF cytokine is known to have a pro-tumour role with functions covering several hallmarks of cancer, including resisting cell death, inducing angiogenesis, promoting genome instability, tumour proliferation and acting as an immunosuppressor 34 . Similarly, the SPP1 pathway showed significant signalling interaction between cancer and macrophages, and SPP1 is a poor prognostic marker for clear cell carcinoma 35 , 36 . Cancer-associated fibroblasts signal to the cancer cells for tumour growth and angiogenesis via pathways such as HGF, ANGPTL4, HBEGF and FGF ( Figure 2d ). Dysregulation of the HGF - MET axis is well-known in tumorigenesis and invasion 37 , and here in CCC, cancer-associated fibroblasts and macrophages signal the cancer cells ( Figure 2e ). Similarly, PPIA and BSG co-expression is significantly associated with poor prognosis in tumours 38 , and in CCC the signalling involves all the immune and cancer cells ( Figure 2f ). The other top chemokine signalling pathways in TME were CXCL and CCL signalling and macrophages show significant secretion of these chemokines (Figure S2d). For instance, CXCL16 is a chemoattractant for the Treg cells to the tumour site and its pro-tumourigenic functions 39 , and here we show the macrophages and dendritic cells recruits Treg, Th 1 helper CD4+ T and other CD8+ T cells ( Figure 2g ). Recently, the CCL20 - CCR6 axis has been known to promote cancer progression by enabling cancer cell migration and proliferation via CCL20 signalling also known as macrophage inflammatory protein 40 and in CCC macrophages recruit Treg, CD8+ T and effector CD4+ T cells ( Figure 2h ). Interleukins play a critical role in promoting cancer progression while essential for tumour-directed immune response 41 , the interleukins prominent in the CCC TME are IL1 , IL2 , IL6 , IL16 and IL10 (Figure S2d). IL1 signalling promotes angiogenesis 42 , and macrophages direct signalling to endothelial cells to promote angiogenesis. Furthermore, IL1 is shown in the conversion of macrophages towards the M2 phenotype 42 , and there is evident IL1A signalling between the two macrophage populations in TME ( Figure 2i ). Another pathway is the CALCR pathway with the ADM - CALCRL communication, the macrophages and cancer-associated fibroblasts signal the endothelial cells via ADM secretion, suggesting the promotion of angiogenesis ( Figure 2j ). ADM has been indicated in the upregulation of the VEGF pathway promoting angiogenesis in ovarian carcinoma 43 . This shows that cellular communication in the CCC TME promotes angiogenesis and has a pro-tumourigenic microenvironment in the patients. The key regulatory transcription factors (TFs) for each cell population in the CCC TME were identified using SCENIC 44 . The critical master regulators of CCC cancer cells were HNF1B and SOX4 ( Figure 2k , S2e). HNF1B is a hepatocyte-specific TF expressed in several cancers, and the reduction of HNF1B levels in CCC cell lines has been shown to cause apoptosis 45 . Similarly, SOX4 is upregulated in subtypes of cancerous cells of clear cell renal carcinoma 46 . SOX4 plays a key role in maintaining the stemness of cancer cells 47 and is a known master regulator of EMT pathways in breast cancer 48 . M2 macrophages in CCC showed high NFE2L3 regulon activity and NFE2L3 expression is a poor prognosis marker in colon, pancreatic and renal cancers 49 – 51 . A study has also shown that loss of NFE2L3 protects against inflammation caused by colorectal cancer by modulating TME 52 . Interestingly, exhausted T cells had higher regulon activity of IRF4 compared to CD4 and CD8 T cells ( Figure 2k ). IRF4 has been shown to drive the exhaustion of CD8+T cells in chronic infection 53 . Overall thesignalling pathways and transcriptional regulators in CCC are pro-tumourgenic. Next, we aimed to study the nature of CCC cancer cell progression across the four cancer stages. Interestingly, the more advanced stages of cancer stages 3 and 4 have reduced responses to chemokines and humoral immune responses based on differential gene ontology (Figure S3a). The chemokines are usually responsible for the recruitment of immune cells to the tumour site and humoral immunity comprising of B cells is important for tumour suppression. Here, the advanced stages of CCC seem to be immune-compromised. Further, cell trajectory analysis was performed to map the progression of cancer cells in CCC ( Figure 3a , S3b-c). We observed that the stage 1 cancer cells are on one end of the spectrum, and metastasised stage 4 cancer cells are on the other, demonstrating the cancer cells’ overall progression. Next, the cell fate probabilities were determined to identify the gene expression changes during cancer stage progression (Figure S3c). The sequential genes upregulated during the progression from cancer stage 1 to 4 are CALB1 , TPD52L1 and ITGB8 ( Figure 3b ). CALB1 is an oncogene known in ovarian cancer and acts by inhibiting p53 pathway 54 . Similarly, TPD52L1 55 and ITGB8 56 are known to be in other cancer transitions. Additionally, in metastasised stage 4 cancer, ligands such as GDNF and DKK1 are also responsible for tumour progression 57 , especially DKK1 is known to promote tumour metastasis 58 . Since CCC is associated with endometriosis, we compared CCC with healthy endometrium and endometriosis-related conditions using two publicly available scRNA-seq datasets 24 , 25 . Firstly, we combined the single cells in the two datasets and grouped them as “control” representing healthy endometrium, “eutopic endometrium” representing the healthy endometrium of donors with endometriosis, “endometriosis” representing the tissue with endometriosis and “endometrioma” which are cystic lesions that stem from endometriosis. The integrated scRNA-seq data from this study and public datasets comprising six conditions such as control, eutopic endometrium, endometriosis, endometrioma, CCC and CCC-MD (CCC with a mixed diagnosis of endometriosis) are shown via UMAP dimension reduction projection ( Figure 3c ). To understand the composition of the tissue in different endometrial conditions, the proportion of cell types present in each condition was determined ( Figure 3d , Figure S3d-g). The endometrial region has an abundance of immune cells, with the healthy control containing more than 20% immune cells and 50% epithelial cells. During the endometriosis condition, we can already observe an increase in immune and endothelial cells by more than 10% and a reduction in the proportion of epithelial cells. The population of immune cells further increases to 65% in cancerous conditions like the CCC and CCC-MD, suggesting an immune infiltration. Meanwhile, the endometrioma cyst is predominantly made of fibroblasts and fewer immune cells. Endometriosis is defined by the presence of endometrial tissue in regions other than the endometrium. As women with endometriosis are more likely to get CCC, we aimed to decipher whether endometriosis is a precursor or facilitator of CCC by comparing the cancer cells with the epithelial cells found in endometrium and other endometriosis conditions. The transcriptional similarity was detected by Spearman correlation between CCC cancer cells and epithelial cells found in other conditions ( Figure 3e , Table S2). The transcriptionally closest epithelial cells were glandular secretory, SOX9+ LGR5+ epithelial cells and ciliated in the eutopic endometrium of donors with endometriosis. The glandular secretory and ciliated epithelial cells are found in the luteal phase, and SOX9+ LGR5+ epithelial cells are found in the follicular phase of the menstrual cycle 25 . These cell types are transcriptional-related closely based on UMAP projection and pseudotime trajectory ( Figure 3f – e ). Further, gene ontology analysis also reveals the similarity between the cell types (Figure S3h). The one key signalling pathway active in cancer cells and not in the other compared to epithelial cells was the CyP signalling, known in poor tumour prognosis 38 (Figure S4a, Figure 2f ). Interestingly, it was shown that glandular secretory epithelial cells have high expression of ARID1A and KRAS 25 and this is correlated with ovarian clear cell carcinoma samples 59 . These findings suggest that the CCC cancer cells might have originated from the epithelial cells related to the endometriosis condition such as glandular/secretory, ciliated and SOX9+ LGR5+ epithelial cells. It is necessary to decipher the differences in cell-cell communication to understand the dynamics of different endometrial conditions and CCC. Here, we performed cell-cell communication analysis based on the transcriptional profiles of the ligands and receptors in cell types found in the microenvironment. The major signalling pathways specific to the cancer condition are VEGI, APRIL, IL10, and CD70 ( Figure 4a ), and these pathways mostly involve the signalling from immune cells. The macrophages recruit the T cells, including the Treg and Th1 helper cells via VEGI (TNFSF15) signalling ( Figure 4b ), and it has been shown that TNFSF15 promotes macrophage polarisation toward M1 for tumour suppression 60 . On the other hand, APRIL ( TNFSF13 ) is known to stimulate B cell growth ( Figure 4c ) and cancer cell growth 61 . Further, IL10 an anti-inflammatory ligand, is secreted by M2 macrophages to Th1 helper T cells and dendritic cells ( Figure 4d ). Finally, dysregulation of the CD70-CD27 axis within TME has been associated with tumour progression and immunosuppression 62 , and we observed significant signalling between B cells, Tregs, and exhausted T cells ( Figure 4e ). The OX40 and CD137 show differential presence in CCC and CCC with mixed diagnosis of endometriosis. Interestingly, the CALCR and BTLA pathways are present in both endometriosis conditions and CCC but not in healthy tissue, indicating a dysregulation in the calcium pathway and immune checkpoint, respectively. To predict the potential repurposing drugs for CCC, we aimed to predict the drugs based on the transcriptional profile of the TME. We stratified the patients based on the stages of cancer and used drug2cell 63 to predict the drugs that could potentially target the TME ( Figure 4f , S4b-c). The most significant drug predicted for Stage 1 TME is Tipiracil hydrochloride, which is a thymidine phosphorylase inhibitor used for the treatment of colorectal cancer 64 . The remaining top predicted drugs for Stage 1 were Abatacept and Belatacept targeting the T cells, Efalizumab targeting lymphocytes and Elotuzumab, which has an anti-tumour function by enhancing activation of natural killer cells based on the DrugBank database 65 . Some of the top predicted drugs for Stage 2 were Dacomitinib and Neratinib, and these drugs are used to treat non-small cell lung cancer and breast cancer, respectively. The drugs predicted for higher stages of CCC were drugs for preventing cardiovascular or asthma, suggesting the effects of medication the patients are already taking. Overall, the scRNA-seq profiling of CCC TME enables a detailed understanding of cell-cell communication and the possibility of identifying drugs for repurposing and working towards helping in personalised medicine.

This is the first study to profile single cells from clear cell carcinoma TME during different stages of cancer progression. We decipher the cellular communication in the TME and compare it with other endometrial-related conditions. We further show the potential of scRNA-seq to predict drugs for treatment. However, this study has some limitations, such as a smaller sample size and here we aim to use this as a discovery cohort to understand CCC progression. Another limitation is profiling was done in a single modality, therefore, as an extension, multimodal sequencing can be applied to comprehend CCC TME dynamics such as chromatin accessibility 66 , histone profiling 67 and RNA modification 68 . Across all donors, MIF and SPP1 signalling was prominent in cancer cells while SPP1 signalling was also active in M2 macrophages. The potential recipients of SPP1 signalling included immune cells infiltrating tumours, cancer-associated fibroblasts, and regulatory T cells. Interestingly, SPP1 signalling was high in CCC compared to endometrial conditions. A recent pan-cancer analysis using public data from TCGA showed that SPP1 is overexpressed in most cancers 69 . SPP1 has been shown to recruit macrophages to turn into tumour-associated macrophages in gliomas 70 . Anti- SPP1 antibodies have been shown to suppress colon tumour growth in vivo 71 . SPP1 inhibitors like parecoxib, brefelamide and simvastatin could be used as adjuvants in chemotherapy for the treatment of CCC. Examination of the CCC TME revealed activation of several signalling pathways (HGF, ANGPTL4, HBEGF, and FGF) that promote tumour growth and angiogenesis. M2 macrophages displayed prominent PPIA - BSG signalling, previously linked to poor prognosis in various cancers. Infiltrating immune cells showed active CXCL16 and CCL20 signalling, which aid cancer progression. Furthermore, TME recruitment of interleukins ( IL1 , IL2 , IL6 , IL16 , and IL10 ) further promotes angiogenesis. Finally, ADM signalling, which upregulates VEGF signalling to promote blood vessel growth, was active in macrophages, CAFs, and endothelial cells. Gene regulatory network analysis revealed distinct transcription factor activity in different cell types within the CCC tumour microenvironment. Exhausted T cells displayed the strongest regulon activity of IRF4 , a transcription factor known to drive T-cell exhaustion during chronic inflammation. Blocking IRF4 could be a potential therapeutic strategy to reinvigorate exhausted T cells and enhance anti-tumor immunity. Cancer cells, on the other hand, showed high expression of HNF1B , contributing to maintaining stemness in cancer cells, a characteristic essential for tumour growth. This finding suggests that HNF1B is another potential target for therapeutic intervention in CCC. To understand the unique molecular signature of CCC, we compared it with endometriosis and healthy endometrium. As prior research indicates, women with endometriosis have a threefold increased risk of developing CCC. This comparison aimed to identify potential causative links by analysing cell-cell communication patterns within both conditions. We determined pathways specific to CCC, such as OX40 and CD137, by excluding signalling pathways common to endometriosis and healthy endometrium. Interestingly, CCC cancer cells displayed transcriptional similarity to glandular secretory, ciliated, and SOX9+ LGR5+ epithelial found in endometriosis conditions. Furthermore, ARID1A, a well-established oncogene in CCC, is highly expressed in these same endometriosis cells 25 . This is particularly noteworthy as mutations in ARID1A are found in 57% of ovarian CCC cases 72 . Based on this shared expression of ARID1A and the observed transcriptional similarities, we postulate that ovarian CCC cells may originate from these specific epithelial cells present in the endometriosis condition. Finally, we aim to find druggable targets during different stages of cancer progression. Tipiracil hydrochloride, which has been used for the treatment of colorectal cancer, was shown to be drug-specific for stage 1 of CCC. For Stage 2, Dacomitinib and Neratinib were predicted to counter CCC. These drugs are used to treat non-small cell lung cancer and breast cancer, respectively. Established based on this cohort, we show how we could determine potential candidates using scRNA-seq for patient-specific drug interventions in ovarian CCC.

Ovarian cancer (OC) is a lethal gynaecological cancer affecting women globally and is the 7 th most common cause of female cancer-associated death 1 , 2 . The World Health Organisation’s (WHO) International Agency for Research on Cancer (IARC) estimates ovarian cancer to increase in incidence and mortality in the next two decades 3 . The high mortality rates associated with OC are primarily due to the advanced disease stage at presentation and the lack of effective screening tools. It is categorised into four subtypes: serous, endometroid, clear cell, and mucinous 4 . Ovarian clear cell carcinoma (CCC) exhibits different prevalence rates across various regions, representing 5–10% of cases in North America but ranging from 13–25% in East Asia 5 , 6 . The greater East Asian preponderance is mirrored by the higher incidence of endometriosis in East Asia 7 – 10 , CCC being the subtype more highly correlated with endometriosis 11 – 16 . Endometriosis is an inflammatory, non-malignant lesion affecting 10% of women of reproductive age worldwide 17 , causing pelvic pain and infertility. Sampson’s theory of retrograde menstruation 18 postulated that endometrial cells moved from the inner lining of the uterus, through the Fallopian tubes, and into the abdominopelvic cavity. Women with endometriosis are three times more likely to develop CCC 16 . The pattern of distribution of endometriotic lesions in the abdominal or pelvic space is similar to ovarian cancer metastasis. Oncogenic mutations in ARID1A , PIK3CA , KRAS and PPP2R1A 19 have been described in endometriotic tissue. PIK3CA mutation has been described as an early event in CCC 20 . Altogether, these data suggest that endometriosis is less-than-benign, prompting an enduring clinical question as to whether endometriosis is a precursor or facilitator of CCC. Our study compares scRNA-seq data on CCC with endometriosis to identify the precursor cells for CCC. A recent study identified two expression subtypes in CCC: epithelial-like (associated with earlier stages) and mesenchymal-like (associated with later stages) 21 . Recently, single-cell RNA sequencing (scRNA-seq) technology 22 , 23 has been applied to decipher normal and tumour conditions. Two recent cohort studies have used scRNA-seq to describe endometriosis 24 , 25 . However, there is currently a lack of high-resolution scRNA-seq data on CCC. To fill this gap, we present scRNA-seq profiling of CCC patients in our cohort, which includes various clinical stages of CCC. We used previously published studies on endometriosis to understand the differences and associations between CCC and endometriosis and, as well as to elucidate the origin of ovarian CCC cells. Analysing the differential genes expressed in heterogenous cell types present in CCC compared across the controls, we sought to acquire molecular insights that might clarify the role of endometriosis in CCC. Further, we provide insights into the critical TME cellular communications, thereby identifying molecules and pathways potentially targetable by existing drugs or further therapeutic development.