A pooled single-cell genetic screen identifies regulatory checkpoints in the continuum of the epithelial-to-mesenchymal transition

Abstract

Integrating single-cell trajectory analysis with pooled genetic screening could reveal the genetic architecture that guides cellular decisions in development and disease. We applied this paradigm to probe the genetic circuitry that controls epithelial-to-mesenchymal transition (EMT). We used single-cell RNA sequencing to profile epithelial cells undergoing a spontaneous spatially determined EMT in the presence or absence of transforming growth factor-β. Pseudospatial trajectory analysis identified continuous waves of gene regulation as opposed to discrete 'partial' stages of EMT. KRAS was connected to the exit from the epithelial state and the acquisition of a fully mesenchymal phenotype. A pooled single-cell CRISPR-Cas9 screen identified EMT-associated receptors and transcription factors, including regulators of KRAS, whose loss impeded progress along the EMT. Inhibiting the KRAS effector MEK and its upstream activators EGFR and MET demonstrates that interruption of key signaling events reveals regulatory 'checkpoints' in the EMT continuum that mimic discrete stages, and reconciles opposing views of the program that controls EMT.

Figure 1:. Pseudospatial trajectory reconstruction of spontaneous EMT reveals the transition as a continuum of epithelial-mesenchymal states

a) Schematic of spontaneous confluence-dependent EMT assay, cell isolation and pseudospatial trajectory reconstruction using Monocle2. Red circle denotes the area that defines inner and outer cells for macro-dissection. b-c) t-SNE embedding of cells from our spontaneous EMT assay. Cells are colored by the fraction from which they were isolated (b) or expression of the mesenchymal marker VIM (c). d) Trajectory of inner and outer MCF10A cells upon spontaneous EMT progression. Arrow denotes progression of pseudospace. Insert: density of cells across pseudospace. e) Left: Stitched brightfield images of an MCF10A colony at the end of our spontaneous EMT assay (2,000 μm scale bar). Right, top to bottom: E-cadherin and DAPI staining of cells from the center, middle and edge of the MCF10A colony (50 μm scale bar, representative fields from 6 images across 3 independent samples). f) Expression of epithelial and mesenchymal markers across pseudospace. Cells are colored as in b. g) Hierarchical clustering of kinetic curves for dynamically regulated genes across pseudospace for all 5,004 cells in our experiment (likelihood ratio test, FDR q < 1 × 10⁻¹⁰ and AUC > 10). Rows represent row centered dynamics of gene expression. h) Gene-set analysis using the Gene Ontology Biological Processes and MSigDB Hallmarks gene-set collections of gene clusters from g (hypergeometric test FDR, q < 0.05). i) Expression of epithelial and mesenchymal markers across pseudospace in primary human mammary epithelial cells (HuMEC). Cells are colored as in b.

Figure 2:. Alignment of spontaneous and TGF-ß-driven EMT pseudospatial trajectories identifies discrete waves along the EMT continuum

a) Dynamic time warping of pseudospatial trajectories allows for comparison of the dynamics of EMT progression along a common axis. b) Epithelial and mesenchymal markers expression across warped pseudospace (cells are colored by treatment). c) Hierarchical clustering of kinetic curves for dynamically regulated genes that vary significantly between spontaneous (5,004 cells) and TGF-ß-driven (4,237 cells) EMT trajectories (likelihood ratio test, FDR, q < 1 × 10⁻¹⁰ and ∣ΔAUC∣ > 0.02). Rows represent row centered dynamics of gene expression. At left: Gene-set analysis on gene clusters using the Oncogenic Signatures gene-set collection (hypergeometric test FDR, q < 0.05). Red and blue arrows denote association with increased or decreased activity, respectively. At right: Gene-set analysis on gene clusters using the GO-BP and Hallmarks gene-set collections (hypergeometric test FDR, q < 0.05). d-f) Pseudospatial expression dynamics of EMT-associated genes that increase in expression at the end of spontaneous and TGF-ß-driven trajectories (d), towards the end of the TGF-ß-driven trajectory (e) and towards the middle of the TGF-ß-driven trajectory (f). g-h) Boxplots of early and late EMT scores of MCF10A cells at early and late positions in pseudospatial trajectories (Mock = 1,020 cells, TGF-ß = 772 cells) and HNSCC tumors (6 = 80 cells, 20 = 321 cells, 5 = 41 cells, 18 = 140 cells, 22 = 119 cells, 25 = 54 cells, 17 = 330 cells, 16 = 56 cells). Boxplots depict the median score (bold line within box) with lower and upper hinges depicting the 25th and 75th percentiles, respectively. i) Density of cells across EMT trajectories after k-nearest neighbor projection of HNSCC tumor cells to MCF10A cells under spontaneous and TGF-ß-driven conditions.

Figure 3:. Multiplexed loss-of-function screening of EMT-associated genes recovers deficiencies in TGF-ß-induced EMT

a) Schematic of pooled approach to determine regulators of distinct EMT states. Red circle in right panel denotes the area that defines the boundary between inner and outer cells for macro-dissection. b) Collection of EMT associated cell surface receptors and transcription factors included in our CROP-seq screen. c-d) t-SNE embedding of sgRNA containing cells from our TGF-ß-exposed CROP-Seq experiment colored by EMT marker expression (c) or expression of NTC or TGFBR2 sgRNAs (d). e) Fraction of cells within VIM low clusters expressing NTC, TGFBR1 or TGFBR2 sgRNAs from our TGF-ß-exposed CROP-Seq screen. f) Total number of cells expressing NTC, TGFBR1 or TGFBR2 sgRNAs from our TGF-ß-exposed CROP-Seq screen. g) Expression of FN1 and VIM across cells expressing a sgRNA to NTC (943 cells), TGFBR1 (219 cells) or TGFBR2 (2299 cells) from our TGF-ß-exposed CROP-Seq experiment. Point within the violin depicts the mean expression level for each group with violin spanning the minimum and maximum expression value across cells.

Figure 4:. Accumulation of knockout cells across spontaneous and TGF-ß-driven EMT trajectories identifies regulators of discrete checkpoints across the EMT continuum

a-b) Enrichment of knockouts whose distribution is significantly altered across pseudospace, and therefore EMT progression, in our spontaneous (11,908 cells) (a) and TGF-ß-driven (9,951 cells) (b) conditions. The distribution of cells expressing sgRNAs against EMT genes was compared to the distribution of NTC controls using Chi square (empirically determined FDR < 10%). For targets whose distribution is altered enrichment across each region was determined by calculating the odds ratio. c) Percent E-cadherin (top panels) or vimentin (bottom panels) positive cells in MCF10A colonies exposed to MEK (U0126) and PI3K (LY294002) inhibition after spontaneous (left panels) or TGF-ß-driven (right panels) EMT. Error bars denote standard deviation from the mean (n = 3, two-tailed Student’s t test). d) Percent E-cadherin (top panels) or vimentin (bottom panels) positive cells in MCF10A colonies exposed to EGFR (Erlotinib), MET (Crizotinib), FGFR (Infigratinib) and ITGAV (Cilengitide) inhibition after spontaneous (left panels) or TGF-ß-driven (right panels) EMT. Error bars denote standard deviation from the mean (at left: spontaneous EMT control/EGFRi/ITGAVi n = 7, METi/FGFRi n = 4 independent samples; at right: TGF-ß-driven EMT control n = 4, EGFRi/METi/FGFRi/ITGAVi n = 3 independent samples, two-tailed Student’s t test). e) Inferred EMT regulatory network and putative regulators identified in this study. f) Model depicting the MEK dependent EMT regulatory checkpoint created and its effects on the development of intermediate EMT phenotypes.