Single-nuclei sequencing of uterine serous carcinoma reveals racial differences in immune signaling

Significance

Single-nuclei sequencing in uterine serous carcinoma (USC) reveals differences in tumor transcriptomes and immune responses. Black patients’ tumors display aggressive features and immune profiles indicative of immunosuppression. This study suggests that understanding these differences could lead to better treatments, especially since it shows PAX8, a gene that is involved in uterine development, suppresses immune signaling in USC. This could ultimately improve outcomes for patients from different racial backgrounds.

Abstract

Significant racial disparities exist between Black and White patients with uterine serous carcinoma (USC). While the reasons for these disparities are unclear, several studies have demonstrated significantly different rates of driver mutations between racial groups, including TP53. However, limited research has investigated the transcriptional differences of tumors or the composition of the tumor microenvironment (TME) between these groups. Here, we report the single-nuclei RNA-sequencing profiles of primary USC tumors from diverse racial backgrounds. We find that there are significant differences between the tumors of Black and White patients. Tumors from Black patients exhibited higher expression of specific genes associated with aggressiveness, such as PAX8, and axon guidance and synaptic signaling pathways. We also demonstrated that T cell populations are reduced in the tumor tissue compared to matched benign, while anti-inflammatory macrophage populations are retained within the TME. Furthermore, we investigated the connection between PAX8 overexpression and immunosuppression in USC through regulation of several cytokines and chemokines. Notably, we show that PAX8 activity can influence macrophage gene expression and protein secretion. These studies provide a detailed understanding of the USC transcriptome and TME, and identify differences in tumor biology from patients of different racial backgrounds.

In endometrial cancer (EC), significant health disparities exist between racial populations. Black women suffer from an overall 55% higher 5-y mortality risk, likely due to higher recurrence rates and frequency of aggressive subtypes (1–5). Additionally, response to traditional chemotherapy is worse for Black EC patients (6). Given that EC is the most common gynecological malignancy in the United States, addressing the differences in survival among Black and White EC patients is an urgent public health matter (7). The cause of the survival gap between Black and White patients remains unclear, although it is likely multifactorial. Differences in socioeconomic factors like access to insurance (8, 9), income (10, 11), and education level (9, 12) are reported to be significantly associated with survival; however, when these factors are accounted for in an equal care setting, disparities in survival persist (13, 14). These studies suggest that other factors, such as variation in tumor biology, contribute to differences in survival. For example, Black patients have a higher frequency of serous tumors that often harbor TP53 mutations. TP53 is differentially mutated between racial groups, with approximately 50% of EC tumors from Black patients harboring a mutation in TP53, while only 30% of tumors from White patients displayed the alteration (15). In addition to genomic differences, tumor transcriptomes differ between Black and White patients. The landmark TCGA study conducted a multiomic characterization of EC and found that tumors can be classified into one of three gene expression subtypes: immune, hormonal, and mitotic (16). Over 64% of Black patients were found to have tumors with a mitotic subtype, while only 33.7% of White patients had such a subtype (17).

The biology of USC remains understudied as this subtype is less common than endometrioid endometrial carcinomas (EEC) and represents about 10-15% of all uterine cancer cases; however, USC accounts for over 40% of uterine cancer-related deaths in the US (18). Obtaining an in-depth analysis of USC tumors will give insights into mechanisms responsible for the aggressive nature of this disease and the racial differences between Black and White patients.

Using single-nuclei RNA-sequencing (snRNA-seq) of USC tumors from patients, we identified four distinct tumor clusters in addition to several cell types present within the TME, including myeloid, lymphoid, stromal, endothelial, and ciliated epithelial cells. In addition, USC tumors from Black and White patients were compared to reveal significant differences in the USC tumor transcriptomes between racial groups. Further, differences between immune cell populations in benign and tumor tissues were characterized using multiplex-IHC. By providing a more detailed understanding of USC tumors and the TME in which they reside, this study provides biological insights that could inform treatment decisions based on patient characteristic such as race, as well as tumor biology.

Results

Single-Nuclei Characterization of Uterine Serous Carcinomas.

We performed single-nuclei RNA-sequencing on tumors from four self-identified White patients and tumors from nine self-identified Black patients (Fig. 1A). All tumors were evaluated by a pathologist and diagnosed as serous histology (SI Appendix, Fig. S2). Additionally, all tumors were screened for missense TP53 mutations via exome sequencing and were found to be positive (SI Appendix, Table S1 and Fig. S1). Six out of the thirteen TP53 mutations were located at Arg273, a well-known TP53 hotspot location. Out of the six mutations found at Arg273, 5 were found in Black patients. Additionally, we assessed the presence of clinically relevant mutations, such as those in mismatch repair and POLE mutations. No POLE mutations were found, although some mutations in mismatch repair genes were found (MLH1 and PMS2) (SI Appendix, Fig. S1). Because these patients also had TP53 mutations and were found to be histologically serous, we included them in our cohort.

Fig. 1.

Single-nuclei sequencing of uterine serous carcinoma. (A) Schematic of workflow for tumor processing and downstream analysis. (B) UMAP plot of 102,431 nuclei from 13 patients, colored by the 6 major cell types. (C) Marker genes of each cluster: PAX8 for tumor epithelium, LSAMP for stromal fibroblast, CD163 for myeloid, DNAH7 for ciliated epithelium, CD247 for lymphoid, and FLT1 for endothelium. (D) Gene set enrichment analysis of each cell type cluster using ClusterProfiler R package. Significance was limited to below 0.05 P-value and the top 200 features were used to identify enriched signaling pathways. (E) Proportion of cell types per patient. All cells were included for this calculation. (F) Contribution of each racial group to individual clusters. The Seurat object was split by racial group and 20,000 cells from each racial group were included in the UMAP plot to ensure an equal number of cells and accurate representation of distribution of cell type.

In total, 102,431 nuclei were included after filtering, with an average of 2,684 detected genes per nucleus and an average 4,617 mRNA molecules detected per nucleus (SI Appendix, Table S2). We performed dimensionality reduction and unsupervised cell clustering using methods implemented in Seurat by removing batch effects among multiple samples, and the results were presented using uniform manifold approximation and projection (UMAP) plot (Fig. 1B and SI Appendix, Fig. S2 A–C).

Transcriptomic analysis of the tumors and associated microenvironment revealed 6 major subtypes of cells which expressed cell type markers: tumor epithelium (PAX8), ciliated epithelium (DNAH7), stromal fibroblast (LSAMP), myeloid (CD163), lymphoid (CD247), and endothelium (FLT1) (Fig. 1C). Recent studies have demonstrated distinct functional roles of ciliated and unciliated uterine epithelium, and unsupervised clustering of our data confirms the transcriptional differences of the two cell types (SI Appendix, Fig. S4 D–F) (19, 20). Pathway enrichment analysis shows cell type-specific enrichment (Fig. 1D). The tumor epithelium represented the largest proportion of all cell types recovered, with 10 out of 13 tumors comprising over 60% tumor epithelium (Fig. 1E). Interestingly, significant differences in nuclei dispersion were present between patients from separate racial groups, indicating overall transcriptional heterogeneity within the tumor epithelial cells among these groups. This was most notable in the tumor cluster, in which nuclei from Black patients seemed more dispersed, while nuclei from White patients were concentrated together (Fig. 1F). To ensure accurate depiction of cell type distribution between racial groups, an equal number of randomly selected nuclei were included in UMAP plots when comparing racial groups.

Clustering of Tumor Epithelium Reveals Four Distinct Clusters and Racial Differences.

The tumor epithelium cluster was analyzed in greater detail. Clustering initially revealed 13 clusters, and although some differences existed, these clusters were collapsed into four biologically relevant clusters: tumor epithelium 1 (TE1), tumor epithelium 2 (TE2), tumor epithelium 3 (TE3), and stem cell tumor epithelium (SCTE) (Fig. 2A and SI Appendix, Fig. S4 A–E). Enrichment analysis was performed to determine the unique gene expression programs of each cluster. TE1 was enriched for longevity-regulating pathway and AMPK signaling (Fig. 2B). Gene expression of IGF1, IGF1R, and MYC drove enrichment in these terms (SI Appendix, Fig. S6A). TE2 is enriched for many neuronal gene sets, including axon guidance, dopaminergic synapse, and glutamatergic synapse. Gene expression of SEMA3A, SLIT3, PLXNA4, CHODL, and ROCK2 drove enrichment of these terms (SI Appendix, Fig. S6B). While these genes are involved with axon guidance in neurons, some studies have reported that these genes are important for angiogenesis and immune signaling in cancer cells (21–24). TE3 is enriched for cell cycle signaling and oocyte meiosis, likely due to the high expression of mitosis and chromosomal segregation factors within this cluster. This enrichment was driven by canonical proliferation markers, including MKI67, CENPF, CENPE, MCM2, and TOP2A (SI Appendix, Fig. S6C). While there were nuclei expressing MIK67 throughout other clusters, TE3 included the majority of proliferating cells. SCTE was enriched for signaling regulating the pluripotency of stem cells and TGFβ signaling, driven by expression of the inhibitor of DNA binding (ID) family of genes, including ID1, ID2, ID3, and ID4 (SI Appendix, Fig. S6D). This family of genes encodes proteins that block cell differentiation and preserves pluripotency (25–27). To confirm the stem cell phenotype, cell trajectory analysis was performed. Root node selection was performed using the get_earliest_principal_node() function from Seurat, allowing for unbiased selection of a principal node. This function identified a root located in the cancer stem cell cluster (Fig. 2C). Two main trajectories were identified, one into the TE1 cluster, and one into the TE2 cluster, indicating that these two clusters are independently derived from the stem cell cluster. The stem cell markers ID1/2/3/4 decreased as pseudotime increased (Fig. 2D). This further strengthens the notion that the cluster characterized by expression of the ID gene family represents cancer stem cells.

Fig. 2.

Transcriptomic differences between tumors from Black and White patients. (A) The tumor epithelium was subsetted from the main Seurat object and further analyzed. Data from the tumor epithelium cluster was normalized, scaled, and clustered using Seurat. The UMAP shows the four tumor epithelial clusters: TE1 in red, TE2 in green, TE3 in purple, SCTE in blue. (B) Dotplot showing GSEA results from the four tumor epithelial clusters calculated by ClusterProfiler(). Significance was limited to below 0.05 P-value and the top 200 features were used to identify enriched signaling pathways. (C) Trajectory map showing cell development pathways from the principal node. The principle node was identified using the get_earliest_principal_node() function. (D) Plot showing the expression of the ID family of genes over pseudotime. (E) Heatmap showing pathway activity inferred from gene expression. Output generated by Progeny R package. P-value was limited to 0.05 and the top 200 most significant features were used for analysis. (F) The main tumor epithelium Seurat object was subsetted by racial group to make 2 Seurat objects. These objects were downsampled to include 15,000 cells from each racial group to ensure an equal number of cells from each racial group and accurate representation of distribution of tumor epithelium cell type. (G) Significantly differentially expressed genes (DEGs) in tumor cells between racial groups. Genes on the left are higher in White patients, while genes on the right are higher in Black patients. Fold change cutoff is log1 fold change.

We analyzed gene sets to understand pathway activity by using PROGENy, an R package that reports consensus gene signatures for pathway activity based on observed gene expression differences in response to specific stimuli (28). Using this pathway activity analysis, we showed that TE1 was most enriched for p53, EGFR, and VEGF pathway activity (Fig. 2E). The TE2 cluster showed high levels of several immune and inflammatory pathways, including JAK-STAT, TGFβ, Hypoxia, NF-κB, and TNFα signaling, but low EGFR and MAPK activity. TE3 showed high MAPK and WNT signaling, but low p53 signaling. Pathway analysis of SCTE cluster showed high p53, estrogen, and TGFβ pathway activity, but low EGFR and MAPK pathway activity relative to other clusters. These data suggest the presence of multiple USC cell populations within a single tumor, likely coordinating roles relating to tumor development or progression, given the distinct gene expression programs of each cluster.

We next analyzed whether tumors from patients from different racial backgrounds have differential enrichment of these epithelial clusters. To this end, 15,000 cells per racial group were randomly selected via the subset() function in Seurat and UMAP analysis was performed (Fig. 2F). Significantly, these analyses indicate substantial heterogeneity in the proportion of epithelial cell clusters constituting tumors of different racial groups. Specifically, we find that the majority (67%) of TE1 nuclei arose from tumors of White patients, whereas the majority (88%) of TE2 came from tumors of Black patients (SI Appendix, Fig. S5F). The cancer stem cell cluster was split evenly between the racial groups, with 55% from Black patients and 45% from White patients. These findings indicate that while tumors from White patients are composed of primarily epithelial cells with high MAPK/EGFR/VEGF signaling activity (TE1), the tumors from Black patients are enriched with immunomodulatory epithelial cells that have high TGFβ/NF-κB/JAK-STAT signaling (TE2). Additionally, a number of genes, including SEMA3A, IGF1R, PAX8, and CHODL were differentially expressed between the two racial groups, indicating different molecular profiles between the group of tumors (Fig. 2G and SI Appendix, Fig. S6 F–H). Within the clusters, we found differences in enrichment between Black and White patient-derived nuclei, indicating differences in transcription even within nuclei from the same clusters (SI Appendix, Fig. S5G). Within TE1, nuclei from White patients drove enrichment in the AMPK and longevity pathways, as these nuclei were enriched for IGFR1 signaling. Additionally, reflecting the pathway activity analysis, nuclei from White patients were enriched for growth factor signaling. Nuclei from Black patients were enriched for several EC-specific enrichment terms, as well as ER protein transport and RNA gene silencing. Within TE2, nuclei from Black patients were enriched for several cadherin signaling pathways, including upregulation of CDH2 and downregulation of CDH1, as well as vascular motility and IL-13/STAT6 signaling. Nuclei from White patients were enriched for pathways associated with other types of cancer including myeloma, pancreatic, and hepatocellular carcinomas. Additionally, TP53 signaling was among the enrichment terms. These differences within enrichment indicate reliance on distinct pathways for survival and growth between racial groups.

Myeloid Cells Are Key Drivers of Immunosuppression within USC TME.

To reveal potential heterogeneity among the immune component of tumor microenvironment of USC tumors, we separately analyzed the myeloid and lymphoid cells and identified six distinct clusters of immune cells: tumor associated macrophages (TAMs), exhausted T Cells, B cells, natural killer cells (NK cells), and dendritic cells (DCs) (Fig. 3A). These clusters were defined by cell-type-specific markers: TAMs expressed CD163, T cells expressed high THEMIS, B cells expressed IGKC, NK cells expressed KLRC1, and DCs expressed FLT3 (Fig. 3B). Further characterization of immune clusters was performed with the FindMarkers() function of the Seurat software suite. T cell gene expression was characterized by several markers of T cell exhaustion, including TOX, CTLA4, and TIGIT (Fig. 3B). TOX is a transcription factor that is expressed after chronic antigen engagement, leading to a blunted cytotoxic T cell response (29). CTLA4 is a negative regulator of T cell activation and competes with antigen-presenting cells to stimulate CD28 (30). TIGIT is an immune checkpoint molecule that is expressed by T_regs and activated T cells and serves to suppress antitumor T cell function (31). The TIGIT ligand, NECTIN2, is expressed on TAMs, DCs, and many TE2 tumor cells. TAMs express several markers indicative of immunosuppression and aggressive disease, including CD163, SPP1, VEGFA, and HIF1A (Fig. 3B) (32–35).

Fig. 3.

Myeloid cells are key drivers of immunosuppression within USC TME. (A) The myeloid and lymphoid clusters were subsetted from the main Seurat object and further analyzed. Data from these clusters were normalized, scaled, and clustered using Seurat. The most expressed and specific genes of each cluster were found using FindMarker() function from the Seurat R package. From that list, the most biologically relevant genes were picked to identify the separate clusters. (B) Dotplot with cell type markers. (C) Heatmap showing pathway activity from gene expression. Output generated by Progeny R package. P-value was limited to 0.05 and the top 200 most significant features were used for analysis. (D) Enrichment analysis of the top 200 genes associated with each cell type. The Elsevier pathway collection was used to identify enriched gene sets. The top enrichment bar graph is associated with the T cell cluster, while the bottom bar graph is associated with the TAM cluster. (E) Heatmap showing outgoing signals (ligands) on the left and incoming signals (receptors) on the right. The bar graph located on the top of the heatmap indicates relative strength of all signaling relative to other clusters. The bar graph on the right side of the heatmap indicates strength of specific signaling pathway relative to others.

Pathway activation scores were calculated using the PROGENy R package (Fig. 3C). DCs and TAMs had high enrichment in several pathways, indicating high levels of activation and activity. In contrast, T cells, B cells, and NK cells showed little pathway activity, indicating low levels of activation. Enrichment analysis shows T cells were enriched for T cell receptor to NF-κB signaling, CREBBP signaling, and IL-7/ FOXO/NFκB signaling, and TAMs were enriched for genes associated with tumor-infiltrating macrophages in cancer progression and immune escape and macrophage M2-related phagocytosis (Fig. 3D).

Interactions between different cell types can be inferred based on gene expression of matched ligands and receptors. To understand the cross-talk between tumor cells and other cells within the TME, we used the R package CellChat (Fig. 3E). Tumor–myeloid interaction was of particular interest as myeloid cells were the most numerous and active immune cell type in USC tumors. Tumor cells expressed ligands including myelin protein zero (MPZ), semaphorin-3A (SEMA3A), and midkine (MK). The cognate receptors for these ligands were expressed by myeloid cells and included MPZL1, NRP1, and LRP1 respectively. Interestingly, these receptor–ligand pairs are all associated with neuronal cell morphology and function, including cell–cell adhesion, axon guidance, and synaptic signaling. The lymphoid cell cluster expressed the lowest relative number of ligands, further indicating low activity of these cells in the TME. Taken together, these data show that TAMs drive sustained immunosuppression within the TME by interacting with tumor-derived ligands and other signals within the TME, such as hypoxia. Additionally, the T cell population is largely exhausted and unable to mount a strong antitumor immune response indicated by low overall pathway activity and expression of exhaustion markers.

Multiplex IHC Reveals Spatial and Racial Differences in Immune Populations within USC TME.

To confirm mRNA data at the protein level and to visualize immune cell populations in situ, we conducted multiplex IHC on the T cell and macrophage populations in the same tumors which were subjected to single-nuclei sequencing. With this approach, we also gained spatial information of immune cells within the tumor, as well as the ability to compare immune populations within benign epithelium, stroma, and tumor epithelium. We imaged CD4+ T helper cells, CD8+ cytotoxic T cells, FOXP3+ T_reg cells, CD68+ macrophages, CD68+CD11c+ antigen-presenting cells, and CD68+CD206+ anti-inflammatory macrophages. We targeted T cells and macrophages as they were the largest immune cell populations within USC tumors. All cell populations we measured decreased in tumors compared to matched benign tissue, with the exception of CD68+CD206+ macrophages. These cells dramatically increased in percentage in tumors, from 39.9% to 64.5% of immune cells analyzed (Fig. 4A). CD206 fluorescent signal was low, but present in the majority of CD68+ macrophages in tumor tissues (Fig. 4B). The CD4+, CD8+, and FOXP3+ T cell percentages were lower in the tumor epithelium than the tumor-associated stroma (Fig. 4C). Additionally, the CD68+ and CD68+CD11C+ macrophage populations are also lower in the tumor epithelium compared to adjacent stroma. However, there was no statistically significant difference between CD68+CD206+ macrophages in the tumor epithelium vs stroma. This indicates that while most immune cell populations are excluded from the tumor body, CD68+CD206+ macrophages are not and can be observed within USC tumors (Fig. 4D). Most immune cells resided within the tumor-adjacent stroma.

Fig. 4.

Multiplex IHC reveals spatial and racial differences in immune populations within USC TME. (A) Pie chart showing percentages of each cell type indicated between total benign sections and tumor sections. All tissue compartments were included for this analysis. Percents represent percentage of immune cells analyzed. (B) IHC images showing the distribution of CD68 and CD206 on tumor-associated macrophages. (Scale bar, 200 μm.) (C) Comparisons of percents of indicated cells between tumor compartments and stromal compartments. N = 13 for each group. If a P-value is less than 0.05, it is flagged with one star (*). If a P-value is less than 0.01, it is flagged with two stars (**). ns = not significant. (D) Representative multiplex IHC images showing tumor islets surrounded by stromal regions. Each image is from a separate patient. Green regions correspond to stroma and red sections correspond to tumors. (Scale bar, 200 μm.) Fuchsia = CD4+ cells, orange = CD8+, yellow = FOXP3+, teal = CD68+, green = CD68+CD206+, red = CD11C+. If a P-value is less than 0.05, it is flagged with one star (*). If a P-value is less than 0.01, it is flagged with two stars (**). NS = not significant. (E) Cell densities of indicated cell types in stroma compartments surrounding tumors from Black or White patients.

We measured the difference between the relative density of immune cells in the stroma of tumors from Black and White patients (Fig. 4E). Significantly, the antitumor cytotoxic CD8+ T cells were present at lower levels in Black patients compared to White. In contrast, the CD68+ and CD68+CD206+ macrophages tended to be higher in tumors from Black patients. Although these differences were not statistically significant, they are in line with single-cell RNA-Seq data and suggest that tumors from Black patients may have a higher degree of immunosuppression given the lower density of CD8+ cytotoxic T cells and higher density of macrophages.

PAX8 Is Associated with Black Race and Correlates with Immunosuppressive Myeloid Cells in USC.

PAX8 is a canonical marker of endometrial epithelial cells and is commonly up-regulated in uterine cancers (36–40). As such, PAX8 was one of the most highly expressed factors in the tumor epithelium cluster (Fig. 5A). Immunofluorescence analysis further confirms the presence of PAX8 protein within the tumors (Fig. 5B). Notably, based on its unequal expression in TE1 vs. TE2, we investigated whether PAX8 was differentially expressed between racial groups. We found that PAX8 expression was higher in Black patients compared to White (Fig. 5C). To corroborate this finding, we analyzed bulk RNA-sequencing data from the Uterine Corpus Endometrial Carcinoma (UCEC) TCGA database. We found that PAX8 was more highly expressed in TP53-mutated endometrial carcinomas (Fig. 5D) and that PAX8 expression was increased in tumors from Black patients compared to White patients, in the CNV-high subtype (Fig. 5E), indicating a significant role of PAX8 in tumorigenesis of USC tumors. Supporting this, we found that patients with high PAX8-expressing tumors for all subtypes had worse overall survival compared to patients with low PAX8-expressing tumors (P-value = 0.00787, HR = 0.549) (Fig. 5F). Additionally, PAX8 is amplified more often in Black patients vs White patients in the TCGA dataset (SI Appendix, Fig. S7).

Fig. 5.

PAX8 is associated with Black race and correlates with immunosuppressive myeloid cells in USC. (A) UMAP showing distribution of PAX8-expressing cells. (B) Immunofluorescence showing protein levels of PAX8 in a representative serous tumor. (C) Violin plot showing PAX8 expression in tumor epithelium of Black and White patients. P-value is < 2.2e-16. (D) Bar graph shows the level of PAX8 expression in tumors from the TCGA database as a function of TP53 mutation. (E) Dotplot shows the level of PAX8 expression in CNV-high endometrial tumors from the TCGA database as a function of Black or White race. (F) Disease-specific survival between patients with tumors that express either high (>0.75) or low (<0.75) PAX8 expression. Survival is measured until 100 mo post diagnosis. (G) Heatmap showing pathway activity from gene expression of PAX8high and PAX8low tumor cells. Output generated by Progeny R package. P-value was limited to 0.05 and the top 200 most significant features were used for analysis. (H) Scatterplot of the average expression level of PAX8 in tumor epithelial cells per patient vs. % of all cells that are CD8+ cells (Top) or % of all cells that are CD68+CD206+ cells. Red dots represent data from Black patients and blue dots represent data from White patients.

We then investigated the functional consequences of high PAX8 expression by stratifying tumor epithelial cell clusters into PAX8^high and PAX8^low groups and performing differential pathway analysis. Results showed that PAX8 expression correlated with immune signaling pathways, including TNFα, TGFβ, and NFκB (Fig. 5G). To understand whether PAX8 expression is correlated with differential abundance of immune cell types, we assessed the correlation between PAX8 expression levels and relative abundance of immune cell population (Fig. 5H). Notably, we observed that PAX8 expression was negatively correlated with antitumor CD8+ T cell percentages in tumor-adjacent stroma, although this correlation did not reach significance (P-value = 0.31, r² = −0.2327). However, PAX8 expression was significantly correlated with CD68+CD206+ macrophage percentages in tumor-adjacent stroma, indicating PAX8 signaling may affect macrophage migration or polarization in the USC TME (P-value = 0.0057, r² = 0.6278).

PAX8 Knockdown Reveals Role in Immune Regulation.

To better understand the transcriptional implications of PAX8 activity in USC, we knocked down expression of PAX8 via siRNA in a USC cell line. ARK2 cells underwent knockdown for 24 h (PAX8 KD), and PAX8 knockdown was validated with qPCR and Western blot (Fig. 6 A–C), after which RNA was harvested and sequenced. PCA analysis showed that roughly 82% of the variance between samples is explained by PAX8 knockdown (Fig. 6D). Supporting the association between PAX8 expression and immune signaling, the enrichment analysis of DEGs showed that the top differentially expressed pathway in PAX8 KD cells is cytokine activity, along with cytokine binding (Fig. 6E). Investigation into specific pathways revealed that DEGs were associated with response to interleukin-1, regulation of myeloid differentiation, regulation of T-helper 1 type immune response, regulation of presynapse organization, and several metabolic pathways. Individual genes associated with immune signaling were differentially expressed in ARK2 vs PAX8 KD cells (Fig. 6F). Specifically, IL-1A, IL-1B, and IL-11 were significantly higher in PAX8 KD cells (Fig. 6G). Additionally, several immune factors such as MK and chemokine (C-X-C motif) ligand 1 (CXCL1) were down-regulated in response to PAX8 KD. These results indicate that PAX8 affects tumor-derived immunomodulatory factors in USC, and specifically may repress expression of inflammatory molecules.

Fig. 6.

PAX8 knockdown reveals role in immune regulation. (A) Volcano plot showing the most differential regulated genes in PAX8 KD cells. Each dot represents a genes. Red genes indicate a significant P-value and a log fold change of over 1. Blue genes indicate a significant P-value only. (B) PAX8 mRNA expression as measured by qPCR. Includes nine samples from two experimental replicates. Significance was calculated using Student’s paired t-test. If a P-value is less than 0.0001, it is flagged with four stars (****). (C) Protein from ARK2, ARK2-PAX8 OE, and PARK2 PAX8 KD was measured for PAX8 and vinculin as a control. (D) PCA plot showing variance between control samples and PAX8 KD samples. (E) Enrichment analysis of KEGG pathways modulated in PAX8 KD cells. (F) Heatmap of top 50 most differentially regulated genes in PAX8 knockdown and control samples. (G) Specific genes that were differentially regulated in PAX8 KD samples compared to control samples. Units are in transcripts per million (TPM). Significance was calculated using a paired Student’s t-test. If a P-value is less than 0.05, it is flagged with one star (*). If a P-value is less than 0.01, it is flagged with two stars (**). If a P-value is less than 0.001, it is flagged with three stars (***). If a P-value is less than 0.0001, it is flagged with four stars (****). NS = not significant.

Tumor-Derived PAX8 Signaling Polarizes Macrophages to an Anti-inflammatory Phenotype.

To test the hypothesis that PAX8 can regulate immune function, we treated differentiated THP-1 macrophages with conditioned media (CM) from PAX8 KD cells, while media from ARK2 cells with normal PAX8 expression was used as a control (Fig. 7A). We found that THP-1 cells treated with PAX8 KD CM expressed less IL-10 and TGFB1, markers that are known to be characteristic of immunosuppressive macrophages. Additionally, we found that expression of inflammatory molecules like TNFα and IL-6 was increased (Fig. 7B).

Fig. 7.

Tumor-derived PAX8 signaling polarizes macrophages to an anti-inflammatory phenotype. (A) Schematic of CM experiment. (B) Gene expression as measured by qPCR in THP-1 cells. ARG1 and TGFB1 are used as markers of anti-inflammatory macrophages and IL-6 and TNFα are used as markers of proinflammatory macrophages. Blue bars represent THP-1 gene expression as a result of 72-h treatment with media from ARK2 PAX8 WT cells. Purple bars represent THP-1 gene expression as a result of 72-h treatment with media from ARK2 PAX8 KD cells. The results shown are the result of two experimental replicates, with three technical replicates each. Significance was calculated using a paired Student’s t-test. (C) Heatmap showing results from three replicates of a cytokine array treated with media from THP-1 cells treated with either control (ARK2) media or conditioned (PAX8 KD) media. (D) Immune markers as expressed by THP-1 cells after treatment with indicated cell media. The results shown are the result of three experimental replicates. Significance was calculated using a paired Student’s t-test. Red indicates ARK2 media, while blue indicates PAX8 KD media.

In the TME, macrophages have several important roles, including antigen presentation and cytokine signaling. Because PAX8 KD RNA-sequencing revealed a role for PAX8 in cytokine signaling and myeloid cell regulation, we investigated how PAX8 signaling affected macrophage cytokine secretion. To do this, we repeated the CM experiment wherein PAX8 KD CM is collected, filtered, and used to treat THP-1 macrophages. We then collected the media from the THP-1 cells after 72 h of treatment and measured cytokine secretion with a cytokine antibody array (Fig. 7C). Analysis of the antibody array showed that PAX8 KD CM-treated THP-1 cells have increased proinflammatory cytokine signaling of factors such as IL-1α, IL-1β, and IL-6 (Fig. 7D). Additionally, anti-inflammatory factors such as TGFβ1 and TGFβ2 are decreased. Growth factors such as FGF7, FGF6, and FGF9 were also secreted less by macrophages treated with PAX8 KD CM. This indicates that tumor-derived PAX8 signaling affects the gene expression and cytokine secretion profile of macrophages within the TME. Further, the changes in gene expression and secreted factors are characteristic of an immunosuppressive microenvironment.

Discussion

In this study, we present findings from single-nuclei sequencing of the USC tumor subtype. We describe tumor cell subtypes including a stem cell cluster that gives rise to four distinct tumor clusters. We also analyze cells within the TME, both using snRNA-seq and multiplex IHC. We find significant differences in both tumor and immune populations between patients from separate racial group; these findings will need to be confirmed with additional studies and more patient samples. Additionally, we interrogate the contribution of PAX8 in regulating tumor-intrinsic immune and inflammatory signaling. We find that tumor-derived PAX8 signaling affects both the expression and protein secretion of macrophages.

Four distinct tumor clusters TE1, TE2, TE3 as well as a stem cell cluster were identified from snRNA-seq of 13 USC tumors. These subpopulations likely coexist in the same tumor as all clusters contained nuclei extracted from each patient. TE1 was enriched for AMPK signaling, while TE2 was enriched for neuronal pathways. There is some evidence that solid tumors can become innervated, which leads to worse outcomes (41, 42). To investigate this possibility, we analyzed innervation-associated genes as defined by KEGG database, but found no statistical significance between racial groups. Additionally, we did not collect any neurons in our single-nuclei experiment. However, an upregulation of axon-associated genes has been reported in several other cancer types, including breast and ovarian (43–45). In breast cancer, the axon guidance molecule SEMA3C has been shown to increase the migration rate of breast cancer cells, ostensibly contributing to their metastatic capability (46). An analysis of survival-associated genes in ovarian cancer showed an enrichment for neural activities in the genes positively correlated with poor survival, indicating neural pathways contribute to the poor survival of these patients (47). In our study, we observed elevated expression of SEM3A in the USC tumor nuclei which were sequenced. This secreted factor can interact with the receptor NRP-1, which is highly expressed on the myeloid cell cluster. Further, our ligand–receptor analysis across all nuclei validates this interaction, consistent with previous reports. Secreted SEMA3A interacting with membrane-bound NRP1, has been shown to act as a chemoattractant for macrophages in several model systems (48–50). It is therefore not surprising that TE2 is also enriched for several immune signaling pathways including TGFβ, NFκB, and TNFα. We speculate that early in the oncogenic process, there was an inflammatory response that could not be resolved, leading to a dysfunctional antitumor immune response characterized by exhausted T cells and anti-inflammatory macrophages. The immune signaling signature could be a remnant of the tumor’s initial defense against antitumor immunity, as well as a tactic to continually polarize macrophages present in the TME to support tumor growth.

Interestingly, TE1 and TE2 are not equally divided between racial groups. By analyzing the tumor clusters by race, we found that most nuclei from Black patients were in the TE2 cluster, while most nuclei from White patients were in TE1. The TE2 cluster also showed high hypoxia pathway activity, which has important implications for immune cell function. Hypoxia has been shown to exert immunosuppressive effects on macrophages, promoting anti-inflammatory phenotypes and an arrest of migration within the TME (51, 52). In fact, TAMs are the only immune cell subtype that displays high hypoxia signaling, as evidenced by our pathway activity analysis. Through multiplex IHC, we also found that CD68+CD206+ macrophages are the only immune cell type that is not excluded from the tumor body compared to adjacent benign tissue. This may indicate that the hypoxic USC TME inhibits macrophage migration out of the tumor.

Previous studies have linked PAX8 with survival, proliferation, and angiogenesis; however, no study to date has identified PAX8 as a regulator of immune function (38, 40, 53). We report that PAX8 expression is correlated with immune signaling in USC tumors. We also find that high PAX8 expression positively correlates with a high density of anti-inflammatory CD68+CD206+ macrophages in the tumor stroma. To understand whether this correlation is unique to PAX8, we performed the same analysis with SOX17 and MECOM, which are known PAX8 binding partners and key transcription factors for female reproductive tissues (54, 55). There was no significant correlation between these factors, indicating PAX8 is uniquely correlated with increasing macrophage presence in the TME. RNA-seq of PAX8 KD ARK2 cells confirmed that PAX8 regulates immune factors, such as IL-1A, IL-1B, IL-11, CXCL1, and MDK. Additionally, GSEA of DEGs from the same experiment showed that PAX8 was specifically involved with myeloid cell differentiation. Because TAMs are the most numerous and active immune cell subtype as shown by our single-nuclei RNA-seq and multiplex IHC data, we investigated whether PAX8 signaling affected TAM gene expression and protein secretion. Our data indicated that PAX8 played a role in macrophage polarization within the USC TME. PAX8 expression negatively correlated with IL-1A, IL1B, and IL-11, suggesting that PAX8 repressed the expression of these factors. These interleukins are well-known inflammatory signals to macrophages; by repressing their expression, PAX8 effectively dampens the antitumor immune response necessary for tumor clearance. PAX8-mediated repression has been shown in ovarian cancer in coordination with SOX17, suggesting binding partners may be important in directing transcriptional activation or repression (53). PAX8 and SOX17 have both been implicated in angiogenesis, so to investigate whether angiogenesis is different in Black vs White patients, we analyzed angiogenesis-associated genes as defined by KEGG database, but found no statistical significance between racial groups. Further studies are needed to identify these potential binding partners in USC.

PAX8 was expressed at a higher level in tumors of Black patients compared to White patients on average in our dataset. This finding was validated by the USC TCGA dataset, which showed the same trend. Further, patients with high PAX8 expression in the USC TCGA data had significantly worse survival than patients with low PAX8 expression.

Because PAX8 positively correlated with TAMs within the USC TME, we expected that Black patients would have higher TAMs. Additionally, previous studies with larger cohorts found differences in the density and type of macrophages in the TME between Black and White patients (56, 57). However, in our case, no statistically significant differences were observed in the density of TAMS in the stroma of USC tumors between Black and White patients. This may be due to our relatively low sample size of 13 patients, although more research is needed to determine whether this is borne out in larger studies.

Low sample size is one limitation of our study. Where possible, we have utilized the TCGA and other related studies to validate any differences in gene expression between Black and White patients in larger datasets. We also acknowledge the potential impact of covariants on our data including incidence of diabetes, hypertension, or history of other malignancies. These factors could influence data collected from patient samples, and where possible, we have included data from cell lines to provide a controlled model system. Additionally, because of the rarity of USC tumors, we relied on frozen tissues that were collected over time. Freezing could have affected the mRNA available from the smaller population of cell clusters, like immune cells, within the tumors. To compensate for any lost immune cells, and to add a spatial dimension to our understanding of immune cells within USC, we performed multiplex IHC on the T cell and macrophage populations in the same tumors analyzed with single-nuclei sequencing, allowing us to integrate the two datasets. However, this approach only allowed us to analyze macrophage and T cell populations in a detailed manner. Analysis of NK cells, B cells, and T cells subtypes will need to be done in future studies given the underrepresentation of these cells in our and others’ single-nuclei libraries (58, 59).

Taken together, our studies revealed insights into the rare and aggressive form of EC, USC. In the samples that were analyzed, we demonstrated significant differences in USC tumors and the associated TME between Black and White patients. Notably, PAX8, previously recognized as a marker for aggressive disease in various cancer types, was highly expressed in tumors of Black patients and emerged as a regulator of immune factors, particularly impacting myeloid cell gene expression and protein secretion. We demonstrate PAX8’s influence on macrophage polarization, suppressing antitumor immune responses crucial for tumor clearance. Furthermore, our observation of elevated PAX8 and its association with poorer survival outcomes underscores the clinical relevance of PAX8, especially for Black women, in USC. Our study contributes valuable insights which lay the groundwork for future research aiming to unravel the complexities of immune modulation in USC, with potential implications for therapeutic interventions and personalized medicine.

Methods

Nuclei Isolation and Single-Nuclei RNA-Sequencing Library Preparation.

Pieces of resected tumor were flash frozen in liquid nitrogen and stored at −80 °C. Frozen tumors were processed with the 10X Nuclei Isolation Kit as instructed by the manufacturer (PN-1000493). The single-nuclei suspension was then loaded onto a microfluidic chip (Chromium Next GEM Chip G, 10X Genomics) and snRNA-seq libraries were constructed according to the manufacturer’s instructions (10X Genomics). The resulting snRNA-seq libraries were pooled and sequenced on an Illumina NovaSeq instrument with 150 bp paired-end reads.

Whole Exome Sequencing.

Genomic DNA from pieces of frozen resected tumor was extracted using QIAamp UCP MinElute® columns as instructed by the manufacturer. Library preparation and sequencing was conducted by Northwestern’s sequencing core (NUseq). Paired-end 150 bp sequencing was performed. Reads were mapped to the human genome (GRCh37) using Bowtie 2 and BAM files were created using Picard. Local realignment was performed using InDels and SNPs from 100 Genomes and base quality score recalibration was performed using GATK.

Cell Culture.

ARK2 cells were obtained from the lab of Dr. Alessandro D. Santin (Yale University, New Haven, CT) and grown in DMEM supplemented with 10% FBS, penicillin, streptomycin, and 1X NEAA. All cell lines used were tested for Mycoplasma and found to be negative. THP-1 cells were cultured with RPMI supplemented with 10% FBS, 0.05 mM β-mercaptoethanol, penicillin, and streptomycin.

Immunofluorescence.

Immunohistochemistry was carried out as previously described (60). In brief, 4 μm sections were cut on coated slides and deparaffinized, with 3% H₂O₂ used to block peroxidase. Antigen retrieval was performed using a 1 M citrate buffer in a pressure cooker for 20 min. PAX8 primary antibody was diluted 1:100 into antibody diluent.

Multiplex Immunohistochemistry.

Multiplex IHC was conducted on both tumor and matched benign tissue. The matched tissue is adjacent endometrium that was collected at the same time as the tumor. One benign tissue per tumor was used in this analysis. Cell percentages were used when comparing different tissue compartments (e.g., tumor vs. stroma) as these compartments are likely to have different cell densities. Cases in which we compared the same compartment (e.g., stroma in tumors from Black vs. White patients) cell densities were used.

siRNA Knockdown of PAX8.

ARK2 cells were cultured to 70% confluency and then washed 3X with PBS. Cells were put into antibiotic-free media directly before transfection with siRNA. Lipofectamine RNAiMAX Transfection Reagent was used to transfect a pool of multiple siRNAs (Horizon L-003778-00-0010) against PAX8 into ARK2 cells. Cells were transfected for 6 h, until full media were added. Twenty-four hours after the initial knock down, cells were washed 3X with PBS and then cultured in full media for 24 h.

RNA Sequencing of PAX8 Knockdown Cells.

PAX8 expression was knocked down with siRNA for 24 h. RNA was harvested and sent to Admera Health (South Plainfield, New Jersey) for library preparation and sequencing. Raw reads were trimmed with Trimmomatic, reads were mapped using STAR, and FPKM tables were generated using StringTie. DESeq2 was used to conduct differential gene expression analysis and generate plots. GOATools and Cluster Profiler was used to conduct gene ontology enrichment analysis.

CM Experiment.

Media from ARK2 and ARK2 PAX8 KD cells were collected after 48 h. Media was filtered through a 20 μm strainer and then centrifuged to eliminate any ARK2 cells from the media. Media was then frozen at −80 °C until needed. THP-1 cells were grown until 80 to 90% confluent and then differentiated with 100 nM PMA for 24 h. Cells were washed 3X with PBS and then treated with a 1:2 ratio of RPMI media and ARK2 PAX8 KD-CM. A 1:2 mixture of RPMI and ARK2-CM was used as a control. THP-1 cells were cultured in CM and RNA was harvested at 72 h. Media was collected after CM treatment for 72 h.

Cytokine Array.

Media from THP-1 cells treated with control or PAX8 KD media was collected after 72 h of treatment. The human cytokine array C5 (Ray Biotech, AAH-CYT-5-2) was used to determine the release of 80 different growth factors and cytokines.

Statistics.

Data plotting and statistical analysis were performed with Prism 9 (GraphPad). We used ANOVA to account for multiple comparisons and t-tests (two-tailed), as indicated in the figure legends. All data points are displayed in box plots to visualize the data distribution. Analysis of sequencing data is described in Methods.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

RNA-seq from ARK2 cells (61) and snRNA-Seq data from tumors have been deposited in GEO (62).