The Stress-Like Cancer Cell State Is a Consistent Component of Tumorigenesis

Abstract

Transcriptional profiling of tumors has revealed a stress-like state among the cancer cells with the concerted expression of genes such as fos, jun, and heat-shock proteins, though this has been controversial given possible dissociation-effects associated with single-cell RNA sequencing. Here, we validate the existence of this state using a combination of zebrafish melanoma modeling, spatial transcriptomics, and human samples. We found that the stress-like subpopulation of cancer cells is present from the early stages of tumorigenesis. Comparing with previously reported single-cell RNA sequencing datasets from diverse cancer types, including triple-negative breast cancer, oligodendroglioma, and pancreatic adenocarcinoma, indicated the conservation of this state during tumorigenesis. We also provide evidence that this state has higher tumor-seeding capabilities and that its induction leads to increased growth under both MEK and BRAF inhibitors. Collectively, our study supports the stress-like cells as a cancer cell state expressing a coherent set of genes and exhibiting drug-resistance properties.

Keywords: cancer cell states; drug-resistant states; melanoma; single-cell RNA-seq; spatial transcriptomics; stress-like.

Figure 1.. Single-cell RNA-Seq on zebrafish melanoma.

(A) Eight tumor biopsies were processed from three distinct tumors using scRNA-Seq. (B) tSNE analysis of 7,278 individual cells from tumor 1. Color indicates the inferred cell type. (C) PCA on the cancer cells revealed three transcriptional cell states, indicated by the colored circles. (D) Heatmap showing the Pearson’s correlation coefficients between the three cell states across all eight biopsies. Biopsy samples cluster according to states and not tumor or animal of origin. (E) Serial biopsies were taken from the same tumor (tumor 1), at one week intervals. The tumor at each time point is shown in the micrographs. The stacked bar plot indicates the proportions of the transcriptional cell states detected in panel (C) for each biopsy.

Figure 2.. Transcriptional program underlying melanoma cancer cell states.

(A) Normalized expression levels of the differentially expressed genes across the cancer cells of tumor 1. Genes are colored by function based on GO annotations indicated at the bottom. Bottom panel - expression of mitfa, egfra, ngfra, ngfrb and crestin, all associated with melanoma cell lines program (B) PCA of bulk melanocyte differentiation from previously reported data (Mica et al., 2013), colors indicate expression levels of SOX2 (purple) and DCT (green). In the heatmap, the Pearson’s correlation levels are shown between the three human melanoma cell type programs and the developmental transcriptomes of stem cells, neural crest, and mature melanocytes. (C,D) Comparison between zebrafish melanoma transcriptional program and (C) human melanoma transcriptional program (Tirosh et al., 2016a) and (D) patient-derived xenografts (Rambow et al., 2018; Tirosh et al., 2016a) (*, P<10⁻²; ***, P<10⁻⁴; ****, P<10⁻⁵).

Figure 3.. The transcriptional programs of the cancer cell states are enriched in cancer areas and detected at the protein level across cancer types.

(A) Hematoxylin and eosin stain of a zebrafish transplanted tumor section. Red and blue dotted lines mark cancer and non-cancer areas, respectively. (B) Gene expression profiles of the indicated genes obtained by spatial transcriptomics performed on a section adjacent to the one shown in panel A. (C) Violin plots indicating the enrichment of each gene (Man-Whitney test, −log₁₀ of the P-value) in each of the indicated gene programs. Genes shown in panel B are indicated by arrows in each program. Negative control represents a randomly selected set of 200 genes. (D) FOS protein is localized in the cell nuclei (white arrows) as shown by the DAPI nuclear staining (blue, left), FOS immunofluorescence staining (green, middle) and a merged image (right). (E) PCA on PDAC (Moncada et al., 2020), TNBC (Kim et al., 2018), oligodendroglioma (Tirosh et al., 2016b) and melanoma (Tirosh et al., 2016a) tumor cancer cells. Color indicates normalized expression levels of the stress-like program, significantly enriched in one vertex (*, P<10⁻²; **, P<10⁻⁴).

Figure 4.. Zebrafish ubb^high cells form higher burden tumors and induction of the stress-like state increases drug resistance.

(A) The ZMEL1-GFP;ubb-tdTomato system to track and select for the cells in the stress-like state. (B) ZMEL1-GFP;ubb-tdTomato cells were sorted to high and low levels of tdTomato intensities and injected into zebrafish for tumor initiation assay followed by quantification of GFP intensity (additional representative images can be found in Figure S4J). (C) Boxplot of tumor burden quantified by GFP intensity of the two different levels of tdTomato compared to parental unsorted cells as a control (mix of population with no selection). Tumor sizes were significantly higher when high tdTomato cells were injected (Mann-Whitney test; *, P<0.05). (D) Schematic of ZMEL1-GFP;ubb-tdTomato cells cultured in optimal and heat shock conditions and exposed to different drug concentrations. (E) Bar plot of cell viability across culturing conditions (optimal/heat shock) and drug treatments. Cell viability is significantly higher under drug treatment when cells were cultured in heat shock conditions.