Anna Pirttikoski

Homeostatic gene expression programs persist through oncogenesis, creating new malignant states that adapt to therapies. To reveal how fallopian tube epithelia (FTE) cell states transform into intra-tumoral heterogeneity in ovarian high-grade serous carcinoma (HGSC), we analysed treatment-naive tumors ( n = 50) and normal FTE ( n = 14) by single cell transcriptomics, identifying tissue-specific transcriptional modules gained, lost, or conserved through carcinogenesis. Of these, two conserved, opposing modules associated with survival and showed genetic biases, suggesting that oncogenic drivers employ existing epithelial programs. Challenging the cancer stem cell paradigm, a developmentally early oxidative-phosphorylation module was associated with improved survival, whereas the developmentally late MAPK/NF-κB module defined a stable, chemoresistant state that had lost the MHC II antigen presentation of the corresponding normal FTE state. Using these modules to predict perturbations capable of shifting cell-state dynamics, we demonstrate in patient-derived organoid models that restraining ER stress via SERCA2 activation drives a resistant, NF-κB–active phenotype, whereas inhibiting the DNA repair regulator CCDC6 promotes the opposite, platinum-sensitive state. Our study suggests that conserved tissue-of-origin states, rather than de novo cancer states, drive the key cell state dynamic and clinical response heterogeneity of HGSC and can be targeted in vitro to modulate platinum response.

Supp. Fig. S1 shows signature robustness. Supp. Fig. S2 displays the association of all signatures with homologous recombination and BRCA1/2 mutations. Supp. Fig. S3 depicts the association of signatures with whole genome duplication, amplification of CCNE1, KRAS, MYC, and MECOM and the relationship with survival. Supp. Fig. S4 shows how complex structural variants manifest and in which signatures they are prevalent. Supp. Fig. S5 explains the SCN-G segmentation aberration. Supp. Fig. S6 presents the validation dataset results in terms of survival and genomic characteristics. Supp. Fig. S7 displays transcriptomics analysis results for the proliferative HRP subtype. Supp. Fig. S8 depicts the transcriptomic results for the EMTHRP subtype. Supp. Fig. S9 contains the results from the transcriptomic analysis for the core HRP subtype.

Summary of biological associations and characteristics attributed to each signature.

Subtype discovery. The SCN signature activity profiles have been clustered into six subtypes with ConsensusClusterPlus and annotated in different colors beneath the dendrogram. The accompanying barplot illustrates signature activity values, with each color representing a distinct signature. Annotations below the main barplot include mutational HR status and the presence of specific gene alterations, such as CDK12, BRCA1, and BRCA2 mutations, as well as CCNE1, MYC, and AKT2 amplifications. A heatmap also displays the counts of complex structural variations (SV) using a color-coded scale. The purple bar plot below indicates the number of breakpoints, the red track with light blue dots represents the tumor mutational burden (TMB), and the amount of long interspersed nuclear elements (LINE) in each sample is shown as a heatmap in the bottom annotation. Notably, AKT2 amplification and CDK12 mutation are mutually exclusive to HRD clusters, whereas CCNE1 and MYC amplifications are present in both HRP and HRD compartments. Moreover, HRD subtypes and the proliferative HRD subtype are characterized by high TMB and number of breaks, and the proliferative subtype is enriched in pyrgos and chromoplexy. Further visualization of the clusters is available at GenomeSpy_6.

Study and signature overview. A, Schematic summary of the data and integration workflow. B, Feature summary: median of the value of each feature in the cohort, with error bars representing the quantiles 0.25 and 0.75. The features are colored according to the feature class and displayed in the boxes above the distribution. Feature classes derived from continuous features are connected below, with their distributions transformed on log scale. Colors match classes in A. The red dashed lines represent discretization thresholds determined using the Jenks natural breaks algorithm. CNV, copy number variant; LINE, long interspersed nuclear elements; SV, structural variant. C, Feature profiles of extracted signatures, derived with SigProfilerExtractor and using the features in A. The signatures exhibit distinct patterns, emphasizing their differences, and can be further observed in the interactive visualization at GenomeSpy_2.

Graphical and textual explanation of the features used for the summarization of genomic profiles.

Associated publication for each single-cell RNA sample.

Biological and clinical associations of the SCN signatures. A, Correlation plot between COSMIC mutational and extracted signatures. Cell color indicates correlation strength and direction. Significance is assessed using Spearman’s correlation with FDR correction. B, Distribution of signature SCN-E (top) and SCN-A (bottom). Each panel displays the signature activity across HRP samples, HRD samples without the respective BRCA mutation (BRCA1 for SCN-E and BRCA2 for SCN-A), and samples with the corresponding BRCA mutation. Samples with BRCA1 deficiency exhibit significantly higher SCN-E signature activity, whereas BRCA2-mutated samples show the same trend with SCN-A signature activity. Interactive visualization is available at GenomeSpy_3. C, Kaplan–Meier survival analysis of HRD-related signatures with log-rank P-values. Signature activities are first summarized per patient using the sample median. Patients are then divided into high and low signature activities using the median signature activity among all patients as a threshold. Elevated HRD-related signature activity is associated with improved treatment response. PFI, platinum-free interval. D, SCN-F signature activities for patients bearing CDK12 mutation and patients not having it. Mann–Whitney U-test with FDR correction significance level is reported (P = 8.49e−10). E, Correlation between copy number alterations and signature activity. Bar plots display the number of amplifications (left), gains (middle), and deletions (right) derived from GISTIC, whose copy number–ploidy ratio correlates with the activity of each SCN signature. The length of each bar segment represents the number of copy number events in each correlation category. SCN-I and SCN-J display the highest number of positive correlations with amplified regions. Similarly, SCN-C exhibits a high number of positive correlations with gains and deletions, excluding HRD-related signatures. In contrast, SCN-K exhibits negative correlations with most deletion and gain peaks. F, Difference in breakpoints abundance in chromosome 19 between CCNE1amp samples and normal CCNE1 samples. The counts are referred to bins of 0.5 Mb. G, GenomeSpy visualization of the signature SCN-K displaying the segmentation profiles of the five samples with the highest value for SCN-K activity, also available at GenomeSpy_5.

Subtype characterization. A, Uniform Manifold Approximation and Projection (UMAP) of scRNA-seq cancer cell profiles from the full dataset of 57 patients, colored by patient. B, UMAP of scRNA-seq profiles from the 57 patients colored by cell type. C, Pathway activities of proliferative subtype samples from analysis of the primary and metastatic tissues. The bar color represents the regulation direction of the pathway, the white stripes pattern indicates nonsignificant results, and the stroke highlights the characterizing pathways (pathways that exhibit the same directional trend in both tissue compartments). D, Proliferation score distribution across the core, EMT, and proliferative subtypes, calculated as the proportion of cells in S and G2/M cell-cycle phases per scRNA sample. Colors represent quantiles, white diamonds indicate median scores, and the red dashed line marks the proliferative HRP reference. Gray densities show the overall score distribution within each subtype. E, Progeny-derived weights and t-values from differential expression analysis (DEA) of hypoxia target genes in metastatic samples of the proliferative subtype. Red dots represent significantly upregulated genes, whereas blue dots indicate downregulated genes. F, Distribution of epithelial ovarian cancer (EOC), fibroblast, and immune cell counts across tissue areas (primary and metastatic) for the HRP subtypes. Notably, metastatic EMT subtype samples display a significantly higher proportion of fibroblasts than the other HRP subtype (Mann–Whitney U test, P = 0.02). G, IC50 values for HRP cell lines from the GDSC1 drug screen (ES2, NIHOVCAR3, FUOV1, JHOS2, and OV90). Compounds targeting the G2/M cell-cycle checkpoint and histone deacetylase (HDAC) inhibitors are highlighted. HRP cells exhibit sensitivity to both classes of compounds, suggesting a potential vulnerability associated with G2/M checkpoint regulation. H, Viability of patient-derived organoids (PDOs) after prexasertib delivery in two concentrations. Notably, core HRP and BRCA2-like HRD subtypes are sensitive, whereas EMT HRP is resistant.

Additional detail description of the methods used in the main text