Single-cell analyses implicate ascites in remodeling the ecosystems of primary and metastatic tumors in ovarian cancer

1As a heterogeneous disease, ovarian cancer (OC) is the most lethal gynecological malignancy, which accounts for 5% of cancer deaths in females. OC is a heterogeneous disease consisting of malignancies with different histological subtypes, molecular biology and microenvironment features, which affect its treatment response and clinical outcomes. Among all OC types, high-grade serous OC (HGSOC) is the most common histological subtype accounting for more than 70% of patients with OC. Once diagnosed, over 75% of patients with HGSOC present an advanced disease with widespread metastasis and ascites,. As reported, a predilection of metastasis to omentum in OC is consistently identified owing to the fatty structure of omentum and peritoneal circulation. Although treatments with chemotherapy plus bevacizumab prolong the 5-year survival, the overall benefits are still limited. Additionally, immunotherapies such as immune-checkpoint inhibitors only showed an objective response rate of 10% in clinical trials and OC subtypes often exhibited diverse responses to immunotherapy owing to the different proportion and quality of tumor-infiltrating lymphocytes (TILs),. Therefore, it is essential to characterize the tumor microenvironment (TME) of OC, which harbors diverse cellular components playing important roles in disease progression and therapy response.

2Single-cell mRNA sequencing (scRNA-seq) is a powerful tool to characterize the cellular features and dynamic relationships of different cell populations in multiple malignancies–. For instance, a previous single-cell atlas of primary ovarian tumor has revealed a GZMK+ CD8+ effector memory (TEM) T cell subset as pre-dysfunctional effector memory cells. Moreover, another OC study defined a population of stem cell-like tissue-resident memory T cells with a maximal expression level of GZMK, which would develop into exhausted T (TEX) cells; however, where these memory T cells originate from is still unknown due to the limited sampling tissues in previous studies. Besides primary tumors, omentum metastases and malignant ascites are equally important in OC studies. For example, interleukin (IL)-6 secreted from cancer-associated fibroblasts in the ascites ecosystem could stimulate JAK–STAT signaling in malignant cells, leading to a poor prognosis and resistance to chemotherapies. But previous single-cell analysis of OC ascites focused largely on malignant cells and other CD45− cells and little is known about the immune milieu in the OC ascites and how malignant ascites influence the immune status of OC. Thus, a high-resolution cellular landscape involving multiple-site tissues is needed to characterize the comprehensive TME of different OC sites, especially omentum metastasis and ascites.

3Here, we delineated a comprehensive landscape of OC TME via scRNA-seq by comparing the unique cellular compositions of five tumor-related sites, including primary ovarian tumor (Pri.OT), omentum metastasis (Met.Ome), ascites, pelvic lymph node (PLN) and peripheral blood (PB). Through T cell receptor (TCR)-based lineage tracing and trajectory inference, we unveiled potential dynamic characteristics of T cells from ascites to tumor tissues. We characterized the functional states and ontogeny of macrophages in ascites and tumor tissues and also highlighted DES+ mesothelial cells as important immunoregulators reprogramming OC ascites. Additionally, we revealed the associations between distinct cellular compositions and the clinical responses to platinum-based chemotherapy, which might serve as indicators of treatment effectiveness. Taken together, our findings provide insights into the functions of malignant ascites and would provide an important resource to guide the development of additional therapeutic strategies.

4To elucidate the complexity of cellular compositions in ovarian cancer, we utilized scRNA-seq to analyze unsorted cells from PB, PLN, Pri.OT, matched Met.Ome and malignant ascites of 14 patients with advanced OC (Fig. 1a and Supplementary Table 1). These patients exhibited five distinct histological subtypes and varying responses to platinum-based chemotherapy. In total, we cataloged 223,363 high-quality single cells into five major cell lineages annotated by canonical marker expression (Fig. 1b,c, Extended Data Fig. 1a–c and Supplementary Table 2).

5We first quantified relative tissue enrichment of major cell clusters by calculating the ratio of observed to expected cell numbers (Ro/e) using data of patients with HGSOC (Fig. 1d,e, Extended Data Fig. 1d and Supplementary Table 3). As expected, B cells and CD4+ T cells dominated the PLNs, whereas lymphocytes and monocytes constituted the main cellular components of PB samples. Of note, we identified all five major cell lineages in both Pri.OT and Met.Ome and the enrichment pattern of most cell types showed no significant differences between these two sites, suggesting a similar complex TME necessary to the development of both primary and metastatic tumor cells (Fig. 1e and Extended Data Fig. 1e). Ascites, frequently found in patients with advanced OC and associated with chemotherapy response, harbored a large number of immune cells and stromal cells. Among them, CD8+ T cells, macrophages and dendritic cells (DCs) were major constituents of ascites with high cell proportions, indicating an inflammatory microenvironment. Mesothelial cells, recently reported to be tightly associated with metastasis of OC, were also preferentially found in malignant ascites (Fig. 1d,e).

6Unlike nonmalignant cells, tumor cells as defined by inferred copy number variations (inferCNV), exhibited a strong interpatient heterogeneity (Extended Data Fig. 2a–c). Notably, tumor cells were identified in all ascites samples, with an averaged proportion of 2.7% (1,444 of 53,499) (Extended Data Fig. 2d). Our observation was consistent with the notion that OC tumor cells prefer to ‘seed’ to the peritoneal cavity rather than spreading via vasculature, which highlights the tight association between ascites and intraperitoneal spread of OC. Further, inferCNV analyses showed that the subclones of tumor cells found within Met.Ome were also detectable in that of Pri.OT (Extended Data Fig. 2e), indicating these subclones as tumorigenic populations of peritoneal metastasis.

7Given that HGSOC is the most common OC subtype, we focused on HGSOC in the subsequent analyses of specific cellular compartments in the TME. We first focused on the intrinsic properties and potential functions of T cell populations in OC. By unsupervised clustering, we identified five CD4+ clusters, five CD8+ clusters and two unconventional clusters (Fig. 2a, Extended Data Fig. 3a,b and Supplementary Table 4). The conventional T cell clusters were further split into naive (TN), central memory (TCM), effector memory (TEM), effector (Teff), regulatory (Treg), T helper 1 (TH1)-like and exhausted (TEX) T cell clusters, which showed different tissue preference patterns (Fig. 2a,b, Extended Data Fig. 3c and Supplementary Table 5). TN cells (T01 and T06) were enriched in PB and PLN, maintaining a quiescent state. Consistent with previous studies, the majority of immunosuppressive FOXP3+ Treg (T03) cells and the HAVCR2+ exhausted CD8+ cells (T10), were predominantly enriched in both two tumor sites. The analyses by flow cytometry also suggested a higher proportion of Treg and PD-1+ T cells in tumor sites than in ascites (Extended Data Fig. 3d), further proving a more immunosuppressive status in tumor tissues compared to malignant ascites. Additionally, CXCL13+ TH1-like cells (T05) were also enriched in tumor sites, whereas CD4+ANXA1+ TCM (T02) and CX3CR1+ Teff cells (T04 and T09) were mainly detected in blood and ascites. Specifically, we identified two CD8+ TEM clusters occupying a large proportion of CD8+ T cells, with T07 ANXA2+ TEM enriched in tumor sites and T08 GZMK+ TEM enriched in ascites (Fig. 2b). Based on limited differential expressed genes, we observed that tumor-enriched ANXA2+ TEM cells expressed increased levels of genes encoding effector molecules (such as GNLY, GZMB and TNFSF10) (Extended Data Fig. 3e), indicating the intrinsic antitumor effector potential of TEM cells inside tumors. By contrast, ascites-enriched GZMK+ TEM cells exhibited higher expressions of EOMES and TCF7 (ref. ) (Extended Data Fig. 3e), which are the key transcription factor genes in progenitor TEX cells, suggesting that GZMK+ TEM cells were more likely to transit into TEX cells.

8Combined with TCR-seq and single-cell transcriptomics, we captured at least one pair of full-length productive α- and β-chains in 54,061 T cells, of which 21.12% (11,415 cells) harbored repeated TCRs of 2,386 clonotypes (Fig. 2a and Extended Data Fig. 3f,g). We then quantitatively evaluated the T cell dynamics using the previously developed STARTRAC indices upon TCR tracking (Methods). T cells carrying repetitive TCRs are defined as clonal cells. The presence of clonal cells across several different tissue sites within the same cluster implies the tissue migration (STARTRAC-migr) of indicated T cell subtypes. And clonal cells found within a T cell cluster were quantified with STARTRAC-expa index, whereas clonal cells between two different T cell subtypes referred to cell state transition (STARTRAC-tran). Among all CD8+ T cells, Teff cells showed the highest clonal expansion, migration and transition index (Fig. 2c), as expected. Additionally, expa index pointed out that clonal expansion might be a possible explanation for the TEX enrichment in tumor sites (Fig. 2c), consistent with previous findings. Notably, we observed strong TCR sharing of TEX cells among two tumor sites and ascites (Met.Ome-AS, Pri.OT-AS and Pri.OT-Met.Ome) (Extended Data Fig. 4a). Considering that exhausted T cells had poor migration capability, this was seemingly logical as these TEX cells would recognize the same tumor-derived neoantigens in different tissues.

9To decipher the potential developmental trajectories of T cells, we performed PAGA and Palantir analysis, excluding two unconventional clusters due to their distinct TCR characteristics. We noticed that ascites-enriched GZMK+ TEM (T08) was located centrally bridging TN (T06), TEX (T09) and Teff (T10) cells (Fig. 2d and Extended Data Fig. 4b), indicating their intermediate states. In addition, STARTRAC pairwise transition analysis based on TCR sharing also showed that GZMK+ TEM exhibited a high ability of transition to Teff, ANXA2+ TEM and TEX cells (Fig. 2e and Extended Data Fig. 4c), further supporting our inferred trajectory analyses. As reported, CD8+ GZMK+ T cells were defined as ‘pre-exhausted’ cells within tumors, which were accumulated by local expansion and replenishment and could further transit to terminal exhausted T cells,. Likewise, compared to other T cells, T08 GZMK+ TEM in our study also harbored a higher ability to transit into TEX cells (Fig. 2e and Extended Data Fig. 4c), suggesting transition from GZMK+ TEM as an important source of TEX cells. Given that GZMK+ TEM cells were mostly enriched in ascites, their transitions to tumor-enriched clusters (TEX and ANXA2+ TEM) might happen together with cross-tissue migration. Thus, we further checked TCR sharing between GZMK+ TEM and TEX/ANXA2+ TEM across different tissues and found that TEX and ANXA2+ TEM cells in tumor sites shared more TCR clones with ascites-derived GZMK+ TEM cells than tumor-derived GZMK+ TEM cells (Fig. 2f). The results indicated that ascites-derived GZMK+ TEM cells might serve as an important source of T cells infiltrating into tumor sites and further transit into TEX or ANXA2+ TEM. Furthermore, GZMK+ TEM in ascites shared more TCR clones with TEX or ANXA2+ TEM cells in Met.Ome than in Pri.OT (Fig. 2g), reflecting a preference of ascites-derived GZMK+ TEM infiltrating into Met.Ome. Then, we checked the TCRs shared among ascites-derived TEM (T08) and tumor-derived T07, T08 and TEX (T10) to confirm the connections between ascites TEM and tumor TEX cells. Of note, tumor TEX (T10) clones linked to ascites-derived GZMK+ TEM showed mutually exclusive patterns with tumor T10 clones linked to T07 and T08 clusters in tumors (Fig. 2h). Considering the hard-to-reverse nature of exhaustion and the lack of mobility of TEX cells, these results further support the notion that TEX cells in tumor may be derived from GZMK+ TEM in ascites, in a process including cross-tissue migration and state transition. Moreover, we checked whether the TCR clones shared by ascites TEM (T08) and tumor TEX (T10) also existed in blood or lymph nodes. We found that the majority of TCR clones shared by ascites TEM (T08) and TEX cells (T10) from primary tumor (61.73%) or metastatic tumor (77.8%) could not be detected in blood or lymph node-derived T cells (Fig. 2h and Extended Data Fig. 4d), further supporting the idea that ascites TEM cells could be an important direct source for TILs. To find the clues about where these TCR clones that are undetected in blood/lymph nodes might come from, we examined the origins of all ascites-enriched TEM cells (T08). We found that the TCRs in 15.36% ascites TEM (T08) cells could be detected in both blood and lymph nodes, whereas 9.57% and 3.34% ascites clonal TEM shared TCRs only with blood T cells or lymph node T cells, respectively (Extended Data Fig. 4e). Taken together, these findings provide insights into the cycle of CD8+ T cells in OC and suggest that ascites-derived GZMK+ TEM cells might serve as a direct source of tumor-infiltrating TEX cells.

10Similar analyses were also performed on CD4+ T cells to quantify their tissue distributions and TCR sharing. In contrast to CD8+ T cells, CD4+ T cells showed an overall lower clonal expansion. Among these clusters, CD4+ Teff cells exhibited the highest clonal expansion, migration and transition indexes (Fig. 3a), similar to the observations in CD8+ T cells. The inferred developmental trajectories also exhibited a similar branched structure. TN (T01), TH1-like (T05) and Treg (T03) cells were positioned at three different branches whereas TCM (T02) cells were located in the middle (Fig. 3b and Extended Data Fig. 5a). In addition, pairwise transition analysis based on TCR sharing (Fig. 3c and Extended Data Fig. 5b) and the shared TCR pattern among T02, T03 and T05 (Extended Data Fig. 5c,d) also revealed that TCM cells were associated with Teff and TH1-like cells, suggesting TCM as potential precursors of CXCL13+ TH1-like cells. Given that TCM cells were enriched in ascites, whereas TH1-like cells were enriched in tumors (Fig. 2b), their transition was accompanied by the ascites to tumors cross-tissue migration of CD4+ memory T cells. Then, we noticed that the TCR clones shared by tumor TH1-like cells and TCM in ascites were almost undetected in any other T cells from tumor, blood and lymph nodes (Fig. 3d and Extended Data Fig. 5e), implying that ascites-derived TCM cells might be a direct source of TH1-like cells in tumors. Additionally, we observed more shared TCR clones between TCM in ascites and TH1-like cells in Met.Ome compared to that in Pri.OT (Fig. 3e and Extended Data Fig. 5e), suggesting that ascites-derived TCM cells were more likely to infiltrate into Met.Ome. Such a tissue preference of TCM cell infiltration could be a potential explanation for the relative enrichment of TH1-like cells in Met.Ome than Pri.OT (Fig. 3f).

11Collectively, through integrated analysis of single-cell transcriptome and TCR data, we identified multiple T cell populations with distinct distribution patterns and revealed unique dynamics of T cells from ascites to tumor sites in OC. We found that ascites-enriched memory T cells (CD8+ GZMK+ TEM and CD4+ TCM) could be a potential important pool for TILs, including CD8+ TEX and CD4+ TH1-like cells, especially for Met.Ome (Fig. 3g). These results implicate a potential role of ascites in shaping the TME of OC during T cell infiltration.

12For myeloid cells, unsupervised clustering gave rise to 15 clusters with distinct gene signatures (Fig. 4a). HLAhiCD14− DC subsets (M01–M04) were further distinguished as CD1C+ DCs (cDC2), CLEC9A+ DCs (cDC1), LAMP3+ DCs and LGALS2+ DCs. Notably, the LAMP3+ DC cluster was also annotated as ‘mregDC’ for its high expression of maturation and immunoregulatory marker genes (such as CCR7, IL12B, CD274, PDCD1LG2 and LAMP3), a cellular state induced upon uptake of tumor antigens (Extended Data Fig. 6a). In line with the tissue distribution patterns reported in other cancer types, LAMP3+ DCs showed relatively comparable enrichment in tumor and lymph nodes. As LAMP3+ DCs exhibited increased expression of genes encoding a co-stimulatory molecule such as CD40, which is associated with interaction between myeloid cells and T cells, and IL12B, which promotes TH1 development(Extended Data Fig. 6a), we speculate that LAMP3+ DCs might also help potentiate the infiltration and differentiation of TH1-like cells in ovarian tumors. This could explain the higher enrichment indexes of both LAMP3+ DCs and TH1-like T cells in Met.Ome than in Pri.OT (Fig. 4b). Notably, we did not detect many conventional DCs (cDCs) in tumor tissues as reported in recent studies, but instead observed their specific relative enrichment in malignant ascites (Fig. 4b and Extended Data Fig. 6b,c). To further elucidate the functions and relationships between different myeloid clusters, we performed similarity analysis of myeloid cells in our dataset with those reported in colorectal cancer (CRC) and hepatocellular carcinoma (HCC) (Fig. 4c). As expected, both cDC1 and cDC2 from different cancer types or tissue sources were clustered together, indicating their conserved phenotypes (Fig. 4c and Extended Data Fig. 6d). We also checked the potential origins of LAMP3+ DC in tumor and observed more cDC2-derived LAMP3+ DC (Extended Data Fig. 6e), which could be associated with higher proportions of cDC2 in ascites.

13In addition, we noticed that the distribution of DC clusters was correlated with chemotherapy responses. Notably, among all DCs, the proportion of M01_DC-CD1C (cDC2) significantly increased in ascites of nonresponsive patients, whereas the M02_DC-CLEC9A (cDC1) proportion decreased (Fig. 4d). Although the previous studies reported that the protumor or antitumor responses of cDCs are uncertain among various types of tumors, our observations indicated that cDC1 and cDC2 cells in the OC ascites might function in an opposite fashion in responses to platinum-based chemotherapy, which remains to be confirmed by further studies.

14As for the monocyte/macrophage compartment, two blood-enriched clusters (M05 and M06) were characterized as CD14+ monocyte and FCGR3A+ nonclassical monocytes, respectively. The remaining clusters were all identified as macrophages (M07–M15) based on the high expression of CD68 (Fig. 4a). Notably, macrophages detected in tumor and ascites were clustered primarily by their tissue distribution. Excluding the proliferating macrophages (M13 and M15), clusters showing relatively comparable enrichment in tumor sites (M07, M10 and M12) were denoted as tumor-enriched macrophages (TeMs), whereas the remaining clusters that showed relatively preferential enrichment in ascites (M08, M09, M11 and M14) were named as ascites-enriched macrophages (AeMs) (Fig. 4b). Among TeMs, C3+ M12 was the dominant subset distributed in both Pri.OT and Met.Ome, whereas EREG+ M07 and C1QA+ M10 tended to be enriched in Met.Ome. Likewise, four AeM subsets were further marked by their featured genes, leading to the classification of FN1+ M08, FABP5+ M09, VCAN+ M11 and FOLR2+ M14.

15To further understand the heterogeneity of macrophage subsets across different tissues and tumor types, we also evaluated the similarities between macrophage clusters in our study and those reported in HCC and CRC, as mentioned above. C3+ TeMs (M12) and C1QA+ TeMs (M10) were clustered into the same branch, resembling the IL1B+ macro and C1QC+ TAMs identified in colon cancer, respectively (Fig. 4c). These clusters highly expressed C1QA and major histocompatibility complex (MHC) class II molecules associated with antigen presentation (Fig. 4a and Extended Data Fig. 6d). Notably, C3+ TeMs not only expressed genes related to phagocytosis and inflammation (C3, CCL4 and TNF), but also upregulated transcriptomic programs associated with the response to tumors (APOE, SPP1 and TGFBI), (Fig. 4a and Extended Data Fig. 6d), which was distinct from the IL1B+ macro in CRC. Conversely, EREG+ TeMs (M07) exhibited high expression of chemokines like CCL20, CCL4, CXCL10, CXCL8 and angiogenesis-related gene VEGFA, as well as low expressions of HLA-related genes, resembling the SPP1+ TAM identified in CRC (Fig. 4a,c and Extended Data Fig. 6d). Among AeM cells, FABP5+ AeM (M09), FOLR2+ AeM (M14) and FN1+ AeM (M08) were all clustered into the same branch with HCC ascites-enriched C6-MARCO, likely reflecting the environmental plasticity of macrophages. Of note, VCAN+ AeM (M11), characterized by high expression of transcripts associated with monocytes (VCAN, S100A9 and S100A12), was clustered into the same branches with tumor-enriched C5-VCAN and ascites-enriched C1-THBS1 in HCC dataset and FCN1+ mono-like cells in CRC (Fig. 4c and Extended Data Fig. 6d). These two macrophages in HCC were defined as myeloid-derived suppressor cells (MDSCs) in the same differentiation lineage. Therefore, VCAN+ AeMs (M11) in our study were more likely to be MDSCs distributed in ascites.

16We next investigated the different functional states of TeMs (M07, M10 and M12) and AeMs (M08, M09, M11 and M14). We observed that TeMs predominantly expressed MHC class II molecules and CD74, which are essential for antigen processing and presentation to CD4+ T cells. TeMs also upregulated the expressions of VEGFA, implying a role for tissue macrophages in promoting tumor angiogenesis. Moreover, we observed upregulated chemokines (such as CCL3/4/5 and CXCL10/11/12) expression in TeMs (Fig. 4e), suggesting the importance of tumor macrophages in recruiting T cells,. Cell–cell interaction analysis within tumor tissues also confirmed that TeMs participated actively in the recruitment of T cells through CXCL10/11–CXCR3, CCL3/4/5–CCR5 and CXCL12–CXCR4 signaling (Fig. 4f). In primary tumors, EREG+ macro (M07) expressed increased levels of CXCL10/11, whereas C3+ macro (M12) highly expressed CXCL12; however, in metastatic tumors, it was surprising to find that the dominant source of CXCL10/11 was switched from EREG+ M07 to C3+ M12 and C1QA+ M10 upregulated the expression level of CXCL12, indicating a reprogramming of macrophages in metastatic tumors. In addition, EREG+ TeM (M07) and C3+ TeM (M12) also showed preferential expression of molecule CCL4 and CCL5, which binds to CCR4 and CCR8, receptors highly expressed by CD4+ Treg cells. We also found very similar interaction patterns between TeMs and ascites T cells (Extended Data Fig. 7a). Collectively, our data suggested the function of TeMs in recruiting T cells and shaping an immunosuppressive niche in tumors.

17By contrast, AeMs exhibited high expression levels of S100A family (S100A8 and S100A9) associated with tumor progression and relatively lower levels of HLA-II genes (Fig. 4e), indicative of a dysfunctional state of macrophages which further contributed to a protumor environment in ascites. Moreover, AeMs also showed strong enrichment of leukocyte migration pathway, with specifically upregulated expression level of CCR1 (Fig. 4e). Notably, we also noticed that AeMs highly expressed LYVE1 and CD163 (Fig. 4e), signature genes of tissue-resident macrophages (RTMs) found in multiple human tissues, implying that RTMs might be an important source of macrophages in ascites.

18Recent studies in mice have suggested that tumor-associated macrophages could have both RTM and monocyte origins. Here, to further infer the ontogeny of TeMs and AeMs, we defined an RTM score using a set of tissue-resident relevant genes, including CD163, LYVE1, FOLR2, MRC1 and TIMD4 (Fig. 5a,b)–. Two of three TeM subsets (M07 and M12) showed much lower RTM scores compared to M10, whereas about half of cells from AeM clusters (M09 and M14) had relatively higher RTM scores (Fig. 5a). Additionally, a set of monocyte-derived macrophage-associated genes were used to complement the analysis of macrophage origins. The results displayed a similar trend, with M07 and M12 exhibiting the highest potential of monocyte-derived ontogeny (Extended Data Fig. 7b),. These findings implied that macrophages identified in OC had two possible origins, with monocyte-derived macrophages as the dominant components in tumors and RTMs accounting for a large part in ascites-enriched subsets. As reported, although RTMs in adult tissues are gradually replaced by circulating monocytes, there constantly exists a self-maintenance population of RTMs arising from embryonic precursors. To explore the extent to which embryonic peritoneal macrophages contribute to ascites-enriched RTMs, we employed Ms4a3Cre-RosaTdT monocyte fate-mapping mouse models to precisely quantify the different ontogeny of macrophages in malignant ascites of ovarian tumor-bearing mouse. Based on the flow cytometry data, nearly half of the AeMs were embryonic-derived macrophages with ~45% proportion of tdTomato− cells (Extended Data Fig. 7c). Further, ~70% CD163+TIM4+ RTMs in malignant ascites were contributed by embryonic precursors (Fig. 5c and Extended Data Fig. 7d,e). These results implied that embryonic macrophages as an important resource of AeMs, contributing to the maintenance of RTMs in the peritoneal microenvironment in OC.

19Subsequently, we characterized the distinct signatures of TeMs or AeMs with divergent ontogeny. RTM-derived M10 expressed significantly higher levels of complement C1Q genes and HLA-II related genes (HLA-DRA, HLA-DPB1 and HLA-DQA1) (Fig. 5d). By contrast, monocyte-derived M07 showed specific expression of VEGFA, IL1B and TNF. The pathway analysis also revealed a strong enrichment of complement activation and antigen processing and presentation pathways in RTM-derived M10, whereas tumor angiogenesis, response to IL-1 and NF-κB pathways were significantly increased in monocyte-derived M07 (Fig. 5d). Multicolor imaging data further confirmed the coexistence of monocyte-derived M07 EREG+ macro and RTM-derived M10 C1QA+ macro in ovarian tumors (Extended Data Fig. 7f). Next, we compared the distinct biological features of ascites-enriched RTMs (M09 and M14) and monocyte-derived AeMs (M08 and M11). RTMs in ascites exhibited higher expression levels of complement C1Q genes (Extended Data Fig. 7g), consistent with the tumor-enriched RTMs. Besides, ascites-enriched RTMs expressed specifically increased levels of FABP5, associated with tumor regulation and CCL2 molecule responsible for monocyte recruitment (Extended Data Fig. 7g). Bulk RNA sequencing of tumor-bearing fate-mapping mice models also confirmed the upregulation of C1q genes, Fabp5 and RTM signature genes, including Timd4 and Cd163 in ascites-enriched embryonic macrophages (Fig. 5e and Extended Data Fig. 7h), further confirming that embryonic macrophages might be a major source of RTMs in the ascites of patients with OC. Of note, we observed that ascites-enriched RTMs expressed lower levels of CD74 and HLA-II related genes than monocyte-derived AeMs, contrary to the observations of TeMs (Extended Data Fig. 7g), likely reflecting the different ontogeny of RTMs in ascites and tumor tissues. Furthermore, we compared the differences between RTMs distributed in tumor and ascites. Ascites-enriched RTMs (M09 and M14) exhibited specific enrichment of oxidative phosphorylation and metabolic-related pathways, whereas tumor-enriched RTMs (M10) significantly upregulated immune response and immune cell migration pathways (Extended Data Fig. 7i). Notably, RTMs in ascites also showed specific high expression of CCL2, which mediates the recruitment of CCR2+ monocytes. Taken together, our analyses establish the connections between macrophage ontogeny-specific features and their various functions in tumor growth. Further studies will be needed to fully discriminate macrophage ontogeny and to attribute the specific functional profile of these macrophages to their ontogenies.

20For nonimmune cells, we first dissected the gene signatures and tissue distributions of all 19 stromal clusters revealed in this study (Fig. 6a,b and Extended Data Fig. 8a–c), including 9 fibroblast clusters (COL1A2+ PDGFRA+), 4 mesothelial cell (MC) clusters (MSLN+UPK3B+), 4 clusters of pericytes (CSPG4+TRPC6+) and 2 vascular smooth muscle cell clusters (MYH11+CNN1+),. Among MCs, DES+ MC (S11) was the dominant stromal cluster in ascites (Fig. 6b,c), which was confirmed by multicolor immunohistochemistry (Fig. 6d). In contrast, VCAN+ MCs (S13) were highly enriched in Met.Ome (Fig. 6b and Extended Data Fig. 8b,c). It has been shown previously that MCs undergo morphological changes and detach from the peritoneal surface during OC peritoneal metastases. We therefore compared the expression levels of cell-adhesion-associated genes (CD44, ICAM1, ITGAV, ITGB1, ITGB8, VCAM1, VCAN, CADM3 and CLDN1) in tumor-derived MCs and found the lowest expression in DES+ MCs (Fig. 6e and Extended Data Fig. 8d), suggesting that DES+ MCs were more likely to fall off into the ascites from tumor tissues. Meanwhile, we observed a significantly decreased cell adhesion potential of MCs in Met.Ome compared to that in Pri.OT (Fig. 6f). These analyses indicated that the loss of cell–cell adhesions could be a reason for MCs to shed from the omentum into ascites, which provides a favorable condition for tumor cell metastasis and colonization.

21Notably, DES+ MCs showed high expression of CXCL12, CXCL13 and CXCL16 (Extended Data Fig. 8e), reminiscent of the recently reported immunomodulatory cancer-associated fibroblasts (CAFs) identified in ascites. By integrating our dataset with that of CAFs in OC ascites, we further confirmed the similarities between DES+ MCs in our study and the immunomodulatory CAFs (Fig. 6g and Extended Data Fig. 8f). We also observed that DES+ MCs had high potential to extensively interact with memory T cells and macrophages (Fig. 6h,i). One of the significantly enriched ligand–receptor pairs was CXCL12–CXCR4, which is associated with recruitment of immune cells. This could help explain the underlying reasons for the abundance of immune cells in ascites and the inflammatory milieu of ascites. DES+ MCs were also predicted to interact with macrophages and MAIT cells via C3-C3AR1 (ref. ), which would lead to the further recruitment of these cells to enhance the inflammatory response in ascites (Fig. 6h). Taken together, the results indicate that DES+ MCs might constitute a key cellular component that plays an important role in the regulation of inflammatory and immune responses in OC ascites.

22Among all endothelial cells, E07 and E08 were annotated as lymphatic endothelial cells based on the expression of canonical marker PROX1 (ref. ), whereas other clusters were identified as vascular endothelium (Fig. 7a and Extended Data Fig. 9a). It has been reported that tumor angiogenesis mainly undergoes two alternate processes, including vessel sprouting by migrating tip endothelial cells and sprout elongating, suggesting that the tip cells could accelerate angiogenesis whereas other endothelial cells were relatively more static. Here, cluster E03 showed high expression of genes associated with endothelial cell migration and matrix remodeling (Fig. 7b and Extended Data Fig. 9b), resembling the tip cells detected in lung tumor, which indicated poor prognosis of patients.

23Further deciphering the transcriptional trajectories of endothelial cells using PAGA, we found that IL13RA1+ E02 and VCAM1+ E06, two major endothelial cell clusters in tumor tissues, exhibited unique features (Fig. 7c,d). We observed that IL13RA1+ E02 showed closer connectivity with the tip-like cells (E03) and upregulated tip cells signatures, whereas VCAM1+ E06 were positioned at another branch (Fig. 7b,d and Extended Data Fig. 9b). Notably, the proportion of IL13RA1+ E02 was significantly increased in Pri.OT samples of nonresponsive patients, whereas VCAM1+ E06 was depleted in Pri.OT samples of platinum-resistant patients (Fig. 7e). Moreover, IL13RA1+ E02 expressed higher levels of SPARC, COL4A1, COL4A2, ANGPT2 and ITGB1 (Fig. 7f), genes involved in vasculature development, epithelial cell proliferation and migration pathways (Fig. 7g), suggesting that IL13RA1+ E02 could contribute to chemotherapy resistance by promoting tumor angiogenesis and migration. In contrast, VCAM1+ E06 showed preferential expression of HLA-II related molecules and ACKR1, a marker of venular endothelium and with a known role in adhesive leukocyte-endothelial interactions (Fig. 7f,g), indicating that VCAM1+ E06 might assist lymphocytes infiltration and participate in antigen processing and presentation to enhance the chemotherapy sensitivity. Therefore, we hypothesized that the relative proportions of IL13RA1+ versus VCAM1+ endothelial cells might serve as a biomarker to predict the benefit from chemotherapy. Furthermore, we also examined whether IL13RA1+ and VCAM1+ endothelial clusters were associated with the long-term prognosis of HGSOC patients using data from The Cancer Genome Atlas (TCGA). We found that patients highly expressing the top 20 signature genes of IL13RA1+ E02 had shorter overall survival (Fig. 7h), further confirming their functions in tumor angiogenesis; however, signature genes of VCAM1+ E06 were not significantly correlated with clinical outcomes of patients with HGSOC (Extended Data Fig. 9d). We also used another independent microarray dataset to validate these results (Extended Data Fig. 9e,f).

24It has been reported that ascites accumulated in patients with OC is associated with chemotherapy response and prognosis. Here, we further investigated the distinct compositions of the ascites microenvironment between responsive and nonresponsive patients. Based on the linear model analysis of all ascites-derived T cells using Milo, we noticed that MAIT cells were highly enriched in ascites of responsive patients before therapy, which was supported by the Ro/e data (Fig. 8a,b). It has been reported that MAIT cells could accumulate and function in the peritoneal cavity during a pathological process or in the tumor tissues,. In our study, MAIT cells were mainly detected in PB and ascites (Fig. 2b). We were able to detect 50 unique shared TCR clones between ascites- and blood-derived MAIT cells (Fig. 8c), suggesting PB as a potential source of ascites MAIT cells. Moreover, ascites-enriched MAIT cells upregulated homing receptors CXCR3 and CXCR4, which bind to CXCL12 and CXCL10, molecules upregulated by other ascites-enriched cells (such as cDC1 and DES+ MC) (Fig. 8d–f), further supporting the chemotaxis of MAIT cells. Ascites-enriched MAIT cells also showed preferential expression of genes related to cell activation (TMIGD2, CCL4 and CCL5) (Fig. 8d,e), suggesting an activated status. We next compared the characteristics of ascites-enriched MAIT cells from responsive and nonresponsive patients. MAIT cells captured from responsive patients overexpressed genes associated with T cell activation, such as ZFP36, JUN, DUSP1, NCR3 and KLRB–, whereas MAIT cells of nonresponsive patients highly expressed genes related to immunosuppression such as LAG3 and IFITM3 (Fig. 8g), suggesting that MAIT cells in ascites from patients with HGSOC with different responses to chemotherapy also exhibited different functions and phenotypes. Altogether, these results indicated that immune-activated MAIT cells might help patients benefit from chemotherapy, whereas MAIT cells in ascites of nonresponsive patients were more likely to be dysfunctional. Furthermore, the levels of activated MAIT cells in ascites could be a useful and noninvasive predictor of effective responses to chemotherapy.

25Despite the usage of platinum-based chemotherapy and improved survival, most patients with advanced OC undergo relapse due to chemotherapy resistance. Here, we applied scRNA-seq to five tissue types of 14 patients with OC with different sensitivities to chemotherapy and systematically dissected the complexity of TME as well as the connections among five tissues. Our analyses revealed that ascites-derived GZMK+ TEM, resembling the previously reported ‘pre-exhausted’ CD8+ T cells within tumors,,, might be a major source of tumor-infiltrating TEX cells. These findings suggest that ascites-derived memory T cells could migrate into tumor sites, acting as an additional important cell pool for TILs. As reported, pre-exhausted GZMK+ T subpopulation were regarded as pre-activated T cells which would accumulate in responsive lung cancer and melanoma tumors following immune-checkpoint-based treatment. We suspected that accelerating the migration of ascites-derived GZMK+ TEM cells into tumor sites could be a potential therapeutic strategy for OC. Moreover, we identified the proportions of MAIT cells in ascites as a potential predictive index in response to chemotherapy. Thus, our work on ascites-enriched T cells inspires us to rethink the functions of malignant ascites in shaping the tumor microenvironment. Future studies will be needed to fully understand the functional roles of these ascites T cells.

26Here, we found that cDCs exhibited specific ascites-enriched distribution patterns in OC. We hypothesized that the presence of cDCs in ascites might serve as a potential source of LAMP3+ DCs in tumor tissues as we found in T cells, which require additional in vivo lineage-tracing validation. Moreover, it has been shown that macrophages were highly heterogeneous in the tumor TME. We identified that macrophages of different origins and phenotypes coexisted within the ovarian tumor and ascites, with TeMs functioning in immune regulation and AeMs being more pro-inflammatory. RTMs in tumor tissues have been reported to provide a pro-tumorigenic niche in lung cancer and the omentum of ovarian tumors. Our data also indicated the potential function of tumor regulation and monocyte recruitment of ascites-enriched RTMs.

27Ultimately, we identified specific populations of stromal cells playing important roles in tumor progression, such as DES+ mesothelial cells in ascites and IL13RA1+ endothelial cells in tumor site. Our findings reveal that ascites-enriched DES+ MCs could help remold the microenvironment of ascites through recruiting T cells and macrophages via CXCL12–CXCR4. The chemokine CXCL12 is known to be expressed by CAFs and binds to the receptor CXCR4, mediating the recruitment of immune cells in tumors. Further, IL13RA1+ endothelial cells exhibited tip-like signatures involved in angiogenesis and were significantly enriched in platinum-resistant patients. Navigating tip cells usually lead the way during vessel sprouting, which could facilitate tumor progression and implies a worse prognosis. These observations suggest that the abundance of IL13RA1+ tip-like endothelial cells might activate angiogenesis and further influence chemotherapy resistance.

28In conclusion, we depicted a comprehensive atlas of the OC microenvironment and revealed the connections between ascites and two tumor sites. Our work provided additional insights into the biological factors that help remodel the OC TME and identified specific cell subpopulations that might serve as potential predictive markers for chemotherapy and prognostic markers of long-term survival, as well as new therapeutic targets or strategies for overcoming platinum resistance and immune suppression.

29This study complies with all relevant ethical regulations and was approved by the Ethics Committee of Xinhua Hospital Affiliated to Shanghai Jiaotong University School of Medicine and Fudan University Shanghai Cancer Center. Written informed consent was provided by all participants.

30Fourteen patients pathologically diagnosed with OC were enrolled in this study for single-cell sequencing. None of the patients had an autoimmune disorder or a history of previous cancer. Only one patient diagnosed with undifferentiated OC was treated with adjuvant chemotherapy. The disease stages of these patients were classified according to the 2018 International Federation of Gynecology and Obstetrics staging system. Fresh samples including primary ovarian tumor, omentum metastatic tumor, PLNs, malignant ascites and PB were obtained from the patients during surgery. The patients received upfront debulking surgery followed by at least six courses of platinum-based chemotherapy. Platinum resistance was defined as progression within 6 months after the last treatment course. Patients HGSOC3, HGSOC6, HGSOC7 and ECO1 were platinum-resistant (nonresponsive), whereas the other patients, except UOC1 were platinum sensitive (responsive). Patients ranged in age from 43 to 82 years old, with a median age of 62 years. Five more patients pathologically diagnosed with HGSOC (patients HGSOC11–HGSOC15) were enrolled in this study for flow cytometry analysis of T cells. The available clinical metadata of these patients are summarized in Supplementary Table 1.

31All animal experiments were approved by the Institutional Animal Care and Use Committee of the Model Animal Research Center, Xinhua Hospital, Shanghai Jiaotong University School of Medicine and were performed in compliance with the guidelines for the care and use of laboratory animals. The maximal tumor burden was not exceeded for mouse tumor experiments on the requirement of our ethics committee. All Ms4a3TdT fate-mapping C57BL/6 mice were female and sourced from Florent Ginhoux Laboratories in Shanghai Institute of Immunology. All mice were provided with water and food and maintained in a pathogen-free facility (12-h light–dark cycle, room temperature at 20–4 °C and relative humidity kept at 45–65%) at the Model Animal Research Center, Xinhua Hospital, Shanghai Jiaotong University School of Medicine. Mice were given an intraperitoneal injection with 106 ID8 cells in 500 µl sterile PBS (pH 7.4) to mimic the peritoneal spread of epithelial ovarian cancer when 4–5 weeks old. Details of cell lines are shown in Supplementary Table 6. For flow cytometry studies and bulk RNA-seq, bloody malignant ascites was collected 65 d after injection of tumor cells.

32Fresh tumor and lymph node samples were cut into approximately 1-mm3 pieces in RPMI-1640 medium (Invitrogen) with 10% fetal bovine serum (FBS; Gibco) and enzymatically digested with a MACS Tumor Dissociation kit (Miltenyi) for 30 min using a gentleMACS Octo Dissociator (Miltenyi) at 37 °C. Dissociated cells were subsequently passed through a 70-μm cell strainer (BD) and centrifuged at 400g for 10 min. The pelleted cells were then resuspended in red blood cell lysis buffer (Miltenyi) and incubated on ice for 5 min to lyse red blood cells. After washing twice with PBS (Invitrogen), cell pellets were resuspended in RPMI-1640 medium supplemented with 10% FBS. PB mononuclear cells were isolated using a leukocyte separation solution (Sigma-Aldrich) according to the manufacturer’s instructions. Malignant ascites samples were collected in 50-ml conical tubes (BD), followed by centrifugation for 10 min at 400g. The remaining pellet was washed twice with PBS and any residual red blood cells were lysed using the above-mentioned procedure. The concentration of single-cell suspensions was adjusted to about 500–1,200 cells per μl. Then, single-cell gene expression and immune repertoire measurements were conducted using the Chromium Single Cell V(D)J Reagent kit (10x Genomics) following the manufacturer’s instructions. All subsequent steps were performed following the standard manufacturer protocols. Completed libraries were sequenced on an Illumina NovaSeq6000 system.

33Low-quality cells were filtered out if cells had fewer than 200 genes expressed or >10% unique molecular identifiers (UMIs) linked to mitochondrial genes. The gene expression matrices of the remaining cells were generated with log normalization and linear regression using the NormalizeData and ScaleData functions of the Seurat package (v.3.1.4). Cells with expression of more than one major cell marker were considered as doublets and removed from each cluster individually. The remaining cells that passed the filtering criteria were considered single cells. We also identified 2,010 platelets with high expression of pro-platelet basic protein. Almost all platelets were found in PB mononuclear cell samples and they are not discussed in this study. For visualization, the dimensionality of each dataset was further reduced using UMAP with the Seurat function Run-UMAP. The principal components (PCs) used to calculate the embedding were the same as those used for clustering.

34After the main cell populations were identified by first-run clustering, we ran the Seurat pipeline for a second time. Unwanted effects caused by percentage of mitochondrial UMI counts were removed by regression in this run. The selection of the resolution on the characteristics of each dataset and the top n PCs from principal-component analysis were used for identification of clusters. For T lymphocytes, we performed extra batch correction across different samples with Harmony (v.1.0) at the default settings. Small clustering groups with expression of dual-lineage signatures, including EPCAM–PECAM1–CD3D, EPCAM–CD79A, PECAM1–CD79A and CD79A–CD3D, were removed from downstream analysis. For other cell types, we did not conduct any batch correction as no obvious clustering bias using raw transcripts per million-like expression data would affect our downstream analyses. Supplementary Table 5, showing the distribution of cell subclusters in five tissues and patients with HGSOC, was provided as diagnostic data to ensure that none of the clusters would arise from individual tissues or patients.

35Copy number variation (CNV) for individual cells was estimated using inferCNV (v.1.2.1) with a 100-gene sliding window. The method to use for smoothing was pyramidal. Genes with an average read count <0.1 among reference cells were filtered when running inferCNV. Endothelial cells, stromal cells, lymphoid cells and myeloid cells were used to define the reference. Epithelial cells were used for the observations. Downsampling was conducted for both the reference and observations to increase the speed of analysis. Epithelial cells were classified to malignant cells using a similar method previously described by Wu et al.

36We calculated the Ro/e for each cluster in different tissues to quantify the tissue preference of each cluster,. The expected cell numbers for each combination of cell clusters and tissues were obtained from the chi-squared test. One cluster was identified as being enriched in a specific tissue if Ro/e > 1. For most clusters, we used the Ro/e index (+++, Ro/e > 3; ++, 1 < Ro/e ≤ 3; +, 0.2 ≤ Ro/e ≤ 1; +/−, 0 < Ro/e < 0.2; and −, Ro/e = 0) to define the cluster preference in a specific tissue. Furthermore, when analyzing the association between each T cell subset and treatment responses to platinum-based chemotherapy, we applied miloR (v.1.5.0), a differential abundance testing framework based on K-NN graphs and generalized linear models.

37The TCR sequences for each single T cell from 10x Genomics were processed using CellRanger (v.3.0.2) with the manufacturer-supplied human VDJ reference genome ‘GRCh38-alts-ensembl’. If two or more cells had the same identical α/β chain pair, the α/β chain pair were identified as clonal TCRs and these T cells were considered to originate from the same clonotypes, identified as clonal cells. After integrating TCR results with the gene expression data of 10x Genomics data, we identified TCR α/β-chain pairs for 59,334 cells. We then presented three STARTRAC indices to analyze different aspects of T cells based on paired single-cell transcriptomes and TCR sequences using STARTRAC (v.0.1.0) as previously described. STARTRAC-expa, STARTRAC-migr and STARTRAC-tran are designed to measure the degree of clonal expansion, tissue migration and state transition of T cell clusters upon TCR tracking, respectively. The MAIT cells (T11) and γδT cells (T12) were not included in these types of analyses because they have distinct TCRs.

38To characterize the developmental origins of CD4+ and CD8+ T cells, respectively, we performed the partition-based graph abstraction method PAGA, a part of the single-cell analysis package Scanpy (v.1.7.2) in Python (v.3.6.13), to infer the potential differential trajectory. Moreover, we used PAGA to assess the most likely trajectories of cell progression among endothelial cells in OC. The computations were carried out using default parameters. The edge connectivity between each subpopulation node for all edges are further compared by using an unpaired two-sided Student’s t-test.

39We also applied Palantir to complement the trajectory analysis using default parameters.

40For an unsupervised comparison of the myeloid clusters identified from multiple datasets, we identified the top 2,000 highly variable genes across different clusters, calculated the mean expression of these genes in each cluster and performed hierarchical clustering using the distance defined as (1 − Pearson correlation coefficient)/2. Here, we used the batch-corrected expression value from the CCA function of the Seurat package. For comparison of stromal cell clusters reported in OC ascites and that detected in ascites in our study, we used the top 1,000 highly variable genes.

41The significantly overexpressed marker genes for clusters were identified using the FindAllMarkers function of Seurat. Genes with adjusted P value < 0.05 by Wilcoxon rank-sum test were defined as cluster-specific signature genes. For two different clusters, we used the Wilcoxon test to evaluate the significance of each gene, with multiple hypothesis correction using the BH procedure. Genes with adjusted P value <0.05 were considered as differentially expressed genes that were further used for GO enrichment analysis with the clusterProfiler package (v.3.14.3). GO terms with adjusted P values <0.05, using the BH procedure, were considered significant.

42To identify the origins of macrophages enriched in tumors and ascites, we used a panel of genes associated with tissue-resident macrophages/monocytes to define the signature of macrophages in our study. The RTM/monocyte-like phenotype of each macrophage cluster was defined as the mean expression of gene signatures. P values were measured by two-sided t-test using Rstatix (v.0.7.0).

43We used cellphoneDB (v.3.0.0) based on cellphoneDB database v.2.0.0 to infer cell–cell interactions of selected ligand–receptor pairs between tumor-enriched macrophages and T cell subsets, DES+ mesothelial cells and indicated immune cell subsets, as well as DC clusters and MAIT cells. The potential interaction strength between two cell subsets was predicted based on the expression of ligand–receptor pairs. The enriched ligand–receptor interactions between two cell subsets were calculated based on a permutation test. We extracted significant ligand–receptor pairs with a P value <0.01.

44The TCGA OC data and microarray data of GSE9891, GSE19829 to GPL8300 were used to evaluate the prognostic performance of gene sets derived from different EC clusters. We used the mean expression of signature genes for distinct cell clusters to evaluate the enrichment of corresponding EC types in patients diagnosed with HGSOC. Specifically, for IL13RA1+ and VCAM1+ EC clusters, we used the top 20 differentially expressed genes of these two clusters as signature genes to define their signatures, (provided in Source Data Fig. 7f). We performed survival analysis using the Cox proportional hazards model implemented in the R package survival (v.3.2.3) to correct patient age and plotted Kaplan–Meier survival curves using the R function ggsurvplot.

45Fresh human tumor samples were cut into approximately 1-mm3 pieces in RPMI-1640 medium (Invitrogen) with 10% FBS (Gibco) and enzymatically digested with the MACS Tumor Dissociation kit (Miltenyi) for 30 min using a gentleMACS Octo Dissociator (Miltenyi) at 37 °C. Dissociated cells were subsequently passed through a 70-μm cell strainer (BD) and centrifuged at 400g for 10 min. The pelleted cells were then resuspended in red blood cell lysis buffer (Miltenyi) and incubated on ice for 5 min to lyse red blood cells. After washing twice with PBS (Invitrogen), cell pellets were resuspended in FACS buffer and kept on ice until staining. Human ascites samples were collected in 50-ml conical tubes (BD) and mouse ascites samples were collected by syringe extraction from terminally anesthetized mice. The samples were centrifuged at 400g for 10 min to obtain cell precipitation. The pelleted cells were then lysed using the above-mentioned procedures. After washing twice with PBS (Invitrogen), cell pellets were resuspended in FACS buffer and keep on ice until staining. Antibodies used to analyze T cells included PerCP-conjugated CD45 (1:200 dilution, Invitrogen), BV570-conjugated CD3 (1:100 dilution, BioLegend), SFV430/780-conjugated CD4 (1:100 dilution, Yuanqi), PerCP-iF710-conjugated CD8 (1:100 dilution, Yuanqi), BV480-conjugated CD25 (1:100 dilution, BD), PE-Cy5-conjugated CD127 (1:100 dilution, BioLegend) and SB702-conjugated PD-1 (1:100 dilution, Invitrogen). Antibodies used to gate macrophages in mouse ascites included BV510-conjugated CD45 (1:200 dilution, BD), BV785-conjugated Ly6G (1:200 dilution, BD), PerCP-Cy5.5-conjugated CD11b (1:200 dilution, BD), BV650-conjugated F4/80 (1:200 dilution, BioLegend), BV421-conjugated TIM4 (1:200 dilution, BD), PE-CY7-conjugated CD163 (1:200 dilution, BioLegend) and 7AAD Viability Staining Solution (BD). The tdTomato signal was detected via the PE channel. Cells were maintained at 4 °C and analyzed on Cytek Aurora flow cytometer (Cytek Biosciences). Data were collected in SpectroFlo (v.3.0.0) and analyzed in FlowJo (v.10.6.2). Gating strategies of T cells are shown in Extended Data Fig. 10a–c and gating strategies for macrophages used for proportion analysis are presented in Extended Data Fig. 10d.

46FACS analysis was used to isolate macrophages with or without tdTomato signals from mouse malignant ascites, performed on a BD Aria III instrument. Antibodies used in this section were APC/cyanine7-conjugated CD45 (1:200 dilution, BD), FITC-conjugated Ly6G (1:200 dilution, BioLegend), PE/cyanine7-conjugated Siglec F (1:200 dilution, Invitrogen), BV785-conjugated Ly6C (1:200 dilution, BioLegend), BV650-conjugated CD11b (1:200 dilution, BioLegend), Alexa Fluor 647-conjugated F4/80 (1:200 dilution, BioLegend) and 4,6-diamidino-2-phenylindole (Invitrogen). Expression levels of these molecules were gated by their negative controls of unstained cells and positive controls of cells stained by each antibody. Expression levels of tdTomato were gated by negative controls of wild-type mice without tdTomato signals. Gating strategies are presented in Extended Data Fig. 10e. Based on FACS analysis, macrophages were sorted into 96-well plates (Axygen) chilled to 4 °C, prepared with lysis buffer with 1 μl 10 mM dNTP mix (Invitrogen), 1 μl 10 μM Oligo dT primer, 1.9 μl 1% Triton X-100 (Sigma) and 0.1 μl 40 U μl−1 RNase Inhibitor (Takara). The cell lysates were sealed and stored frozen at −80 °C immediately. Transcriptome amplifications were performed according to protocol. The External RNA Controls Consortium spike-in controls (Ambion; 1:4,000,000 dilution) were added into each well before reverse transcription. Amplified cDNA products were purified with Agencourt XP DNA beads (Beckman). Quality control (QC) was performed following the first round of purification, which included the detection of GAPDH by qPCR and fragment analysis by Fragment Analyzer (AATI). For single-cell samples of high quality after QC (cycle threshold < 30), the DNA products were further purified with 0.5× Agencourt XP DNA beads and the concentration of each sample was quantified using Qubit HsDNA kits (Invitrogen). Multiplex (384-plex) libraries were constructed and amplified using the TruePrep DNA Library Prep Kit V2 for Illumina (Vazyme Biotech). The libraries were then purified with Agencourt XP DNA beads and pooled for quality assessment by Fragment Analyzer. Purified libraries were then analyzed by an Illumina Hiseq 4000 sequencer with 150-bp paired-end reads. Fastp was used to get the clean reads. Read mapping was performed using STAR (Bulk RNA-seq, v.2.7.2a) using the mouse reference genome (mm10). Gene level quantification was completed using Subread featureCounts (bulk RNA-seq).

47Human tissue specimens, including tumor samples and ascites cell precipitation, were provided by Xinhua Hospital Affiliated to Shanghai Jiaotong University School of Medicine. The specimens were collected within 30 min after the tumor resection and fixed in formalin for 48 h. Dehydration and embedding in paraffin was performed following routine methods. Paraffin blocks were cut into 5-μm slices and adhered to slide glass. Sections were then placed into a paraffin oven at 70 °C for 1 h before deparaffinization in xylene and successive rehydration in 100%, 90%, 70% alcohol. Antigen was retrieved by citric acid buffer (pH 6.0) in a water bath at 95 °C for 20 min. Endogenous peroxidase was inactivated by incubation in 3% H2O2 for 15 min. Following preincubation with 10% normal goat serum to block nonspecific sites for 30 min, sections were incubated with primary antibodies in a humidified chamber at 4 °C overnight. The antibodies used in the validation of DES+ mesothelial cells in ascites cell precipitation were anti-MSLN (1:250 dilution, Cell Signaling Technology), anti-UPK3B (1:20 dilution, Abcam), anti-WT1 (1:100 dilution, Cell Signaling Technology) and anti-DES (1:2,000 dilution, Abcam). The antibodies used in the validation of two macrophage subtypes (M07_Macro-EREG and M10_Macro-C1QA) in ovarian tumor tissue were anti-CD68 (1:400 dilution, Cell Signaling Technology), anti-SPP1 (1:2,000 dilution, Abcam), anti-EREG (1:100 dilution, Lifespan Biosciences), anti-IL1B (1:100 dilution, Cell Signaling Technology), anti-C1QA (1:1,000 dilution, Abcam), anti-RGS2 (1:200 dilution, Abcam) and anti-MARCO (1:200 dilution, Lifespan Biosciences). After the sections were washed with PBS twice for 5 min, the antigenic binding sites were visualized using the PhenoImager Fusion (Akoya) with the Phenochart viewer software (v.1.10) according to the manufacturer’s protocol.

48No statistical methods were used to predetermine sample sizes and the experiments were not randomized. The investigators were not blinded to allocation during the experiments and outcome assessments. Data collection and analysis were not performed blinded. No data were excluded from the analyses. Statistical analyses were performed using R (v.3.6.1) and GraphPad Prism (v.9.0). One-sided or two-sided unpaired Student’s t-tests, two-sided Wilcoxon tests, two-sided unpaired limma-moderated t-tests and Kruskal–Wallis tests were used to evaluate significance, as indicated in figure legends. P < 0.05 was considered statistically significant. Data distribution was assumed to be normal, but this was not formally tested.

49Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.