Coordinated protein modules define DNA damage responses to carboplatin at single-cell resolution in human ovarian carcinoma models

Abstract

Tubo-ovarian high-grade serous carcinoma (HGSC), the most lethal gynecologic malignancy, initially responds to platinum-based chemotherapy, but due to frequent defects in the DNA damage response (DDR), most tumors develop resistance. The molecular mechanisms underlying clinical platinum resistance remain poorly defined with no biomarkers or targeted therapies to improve outcomes. Here, applying mass cytometry, we quantify phosphorylation and abundance of DDR proteins in carboplatin-treated HGSC cell line models. Despite similar levels of intranuclear platinum, a proxy for carboplatin uptake, cells follow divergent fates, reflecting DDR heterogeneity. Unsupervised analysis reveals a continuum of DDR states, and matrix factorization identifies eight protein modules. The activity of one module, containing canonical DDR proteins, increases in carboplatin-sensitive cells. Resistant cells engage a broader DDR protein module. These findings demonstrate the ability of single-cell proteomics to identify functional DDR states and reveal a DDR sensitivity module as a promising biomarker for clinical stratification and therapeutic decisions in HGSC.

Keywords: CIOV1; CIOV2; CIOV3; DNA damage response; UWB1.289; Uniform ManifoldApproximation and Projection/UMAP; carboplatin resistance; mass cytometry/CyTOF; non-negative matrix factorization; partition-based graph abstraction; single cells; tubo-ovarian high grade carcinoma.

Graphical abstract

Figure 1

Characterization of the DNA damage response by CyTOF (A) Pathway map showing CyTOF antibody panel designed to measure proteins participating in the carboplatin-mediated DNA damage response. Proteins marked in gray were not measured due to unavailable antibodies. “β-catenin” is non-phospho-β-catenin. Table S1 provides detailed information about each antibody. (B) Schema of the experimental approach. Foundational experiments were performed using TYK-nu and UWB1.289 HGSC cell lines. Validation was performed using three spontaneously immortalized continuous HGSC cell lines (CIOV1, CIOV2, and CIOV3) derived from patient tumors.

Figure 2

Characterization of DDRs to genotoxic and microtubule-targeting agents in HeLa cells and to carboplatin in TYK-nu cells (A and B) HeLa cells respond with distinct DDRs depending on treatment and cell-cycle phase. (A) UMAP embedding colored by cell-cycle phase highlights DDR differences within the same cell-cycle phase. (B) UMAP embedding colored by treatment demonstrates DDR variation by agent. (C) TYK-nu cells treated with carboplatin, talazoparib, or the combination. Pie charts depict the distribution of cells across cell-cycle phases for different treatments. (D) Biaxial plots of ¹⁹⁵Pt levels, (carboplatin uptake) plotted against c-PARP to distinguish apoptotic from non-apoptotic cells over time. (E) Box plots showing ¹⁹⁵Pt uptake in single intact cells (yellow) and isolated nuclei (blue) over time. CyTOF enabled characterization of apoptotic populations at early times when frequencies of apoptotic cells were low (∼200–1,000 cells). Notches indicate 95% confidence intervals for median values of ¹⁹⁵Pt uptake. Whiskers depict range 1.5 × IQR ± hinges. (F) Fold change in median ¹⁹⁵Pt levels comparing whole cells to nuclei across time points and populations.

Figure 3

Overview of the unsupervised data analysis approach CyTOF data were normalized, arcsinh-transformed, and gated prior to unsupervised analysis according to the schema.

Figure 4

Identification of DDR protein modules in TYK-nu cells (A) UMAP embedding of 721,579 TYK-nu cells based on expression of 29 DDR proteins measured across all time points. Leiden cell clusters are overlaid on the UMAP and colored. (B–E) Partition-based graph abstraction (PAGA) plots show connectivity between Leiden clusters. The plots are colored for (B) cell fate, (C) treatment, (D) time point, and (E) cell cycle. Each cluster is represented by a pie chart indicating the proportion of cells from different conditions that share a similar DDR. (F) DDR protein modules discovered by non-negative matrix factorization (NMF). The matrix of expression levels for 29 DDR proteins in 721,579 cells was decomposed into two matrices. One matrix discovered the most frequently co-occurring proteins in eight DDR modules (number of modules user-selected). The contribution of each protein within a specific module is given by its Z-score, and this matrix is depicted on a hierarchically clustered heatmap. (G) The second matrix describes the activity of each module in individual cells and is overlaid on the UMAP. (H) Box plots depict the activity of each module over time.

Figure 5

Identification of DDR modules in UWB cells (A–E) (A) PAGA plot of Leiden clusters for UWB cells shows connectivity of clusters in high-dimensional space. Cluster nodes are colored based on an additional round of Leiden clustering to identify highly interconnected clusters. PAGA plots are colored for (B) time point, (C) treatment, (D) cell fate, and (E) pseudo-time. 0 and 1 represent the scale on the pseudo-time plot. 0 denotes the starting point of the trajectory (untreated cells), and 1 denotes the endpoint of the trajectory (cells committed to apoptosis). (F) BRCA1 status. (G) Cell-cycle phase. (H) DDR modules discovered by NMF as described in caption to Figure 4F. The contribution of each protein within a module is given by its Z score depicted on a hierarchically clustered heatmap. (I) Module activity is depicted on the PAGA plot. Each Leiden cluster is colored with a pie chart to show the proportion of cells that recruit a specific module. Modules with less than 10% median activity in a cluster were excluded. (J) Box plots depict the activity of each module over time. Whiskers depict range 1.5 × IQR ± hinges.

Figure 6

Characterization of patient-derived CIOV1, CIOV2, and CIOV3 cell lines (A) Violin plots depict expression levels of tumor-associated proteins using the CyTOF tumor cell antibody panel described previously. The color key (top right) denotes the individual cell lines and is applicable to (A), (D), and (E). (B) Cell-cycle distributions vary across cell lines in response to treatment. (C) Pie charts indicate minimal apoptosis under treatments used. (D) Box plots depict ¹⁹⁵Pt uptake. Box colors correspond to cell lines shown in (A). Whiskers depict range 1.5 × IQR ± hinges. For each cell line, treatments are ordered (left to right) as carboplatin, carboplatin + paclitaxel, and carboplatin + rucaparib. (E) Violin plots depict changes in E-cadherin (E) and vimentin (V) expression levels across cell-cycle phases in response to treatment. EV scores were computed according to the following formula: EV Score = arcsinh E-cadherin counts/arcsinh E-cadherin counts + arcsinh vimentin counts). Scores range from 0 to 1 with a score of 1 indicating a purely epithelial phenotype (high E-cadherin, low vimentin) and a score of 0 indicating a purely mesenchymal phenotype (low E-cadherin, high vimentin).

Figure 7

Validation of TYK-nu NMF modules in CIOVs, JHOS22, and OVCAR3 cell lines In an independent experiment CIOV1–3, JHOS2 and OVCAR3 cell lines were treated with carboplatin, a carboplatin-based combination, or vehicle control for 48 h and then analyzed by CyTOF using the same DDR antibody panel (Table S1). (A) Patient treatment timelines reconstructed from electronic health record data illustrating clinical treatment and interventions (further information is provided in Table S5). CIOV1 and CIOV3 cell lines were cultured from malignant ascites, and CIOV2 was cultured from disaggregated tumor tissue. DOD, date of death. (B) Hierarchically clustered heatmap depicting module 6 activity (identified in TYK-nu) in CIOV1-3, JHOS2, and OVCAR3 with TYK-nu as a control. The values in the heatmap are the increase in the percentage of module 6 activity relative to vehicle control under conditions shown.