Portal de Periódicos da CAPES

Deciphering functional tumor states at single‐cell resolution

2021; Springer Nature; Volume: 41; Issue: 2 Linguagem: Inglês

10.15252/embj.2021109221

1460-2075

Rolando Vegliante, Ievgenia Pastushenko, Cédric Blanpain,

Cancer Genomics and Diagnostics

Review17 December 2021free access Deciphering functional tumor states at single-cell resolution Rolando Vegliante Rolando Vegliante Laboratory of Stem Cells and Cancer, Université Libre de Bruxelles, Brussels, Belgium Search for more papers by this author Ievgenia Pastushenko Ievgenia Pastushenko Laboratory of Stem Cells and Cancer, Université Libre de Bruxelles, Brussels, Belgium Search for more papers by this author Cédric Blanpain Corresponding Author Cédric Blanpain [email protected] orcid.org/0000-0002-4028-4322 Laboratory of Stem Cells and Cancer, Université Libre de Bruxelles, Brussels, Belgium WELBIO, Université Libre de Bruxelles, Brussels, Belgium Search for more papers by this author Rolando Vegliante Rolando Vegliante Laboratory of Stem Cells and Cancer, Université Libre de Bruxelles, Brussels, Belgium Search for more papers by this author Ievgenia Pastushenko Ievgenia Pastushenko Laboratory of Stem Cells and Cancer, Université Libre de Bruxelles, Brussels, Belgium Search for more papers by this author Cédric Blanpain Corresponding Author Cédric Blanpain [email protected] orcid.org/0000-0002-4028-4322 Laboratory of Stem Cells and Cancer, Université Libre de Bruxelles, Brussels, Belgium WELBIO, Université Libre de Bruxelles, Brussels, Belgium Search for more papers by this author Author Information Rolando Vegliante1,†, Ievgenia Pastushenko1,† and Cédric Blanpain *,1,2 1Laboratory of Stem Cells and Cancer, Université Libre de Bruxelles, Brussels, Belgium 2WELBIO, Université Libre de Bruxelles, Brussels, Belgium † These authors contributed equally to this work as first authors *Corresponding author. Tel: +32 2 555 4175; E-mail: [email protected] The EMBO Journal (2022)41:e109221https://doi.org/10.15252/embj.2021109221 This article is part of the Cancer Reviews series. PDFDownload PDF of article text and main figures. ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions Figures & Info Abstract Within a tumor, cancer cells exist in different states that are associated with distinct tumor functions, including proliferation, differentiation, invasion, metastasis, and resistance to anti-cancer therapy. The identification of the gene regulatory networks underpinning each state is essential for better understanding functional tumor heterogeneity and revealing tumor vulnerabilities. Here, we review the different studies identifying tumor states by single-cell sequencing approaches and the mechanisms that promote and sustain these functional states and regulate their transitions. We also describe how different tumor states are spatially distributed and interact with the specific stromal cells that compose the tumor microenvironment. Finally, we discuss how the understanding of tumor plasticity and transition states can be used to develop new strategies to improve cancer therapy. Introduction Solid cancers are composed of tumor cells (TCs) and their stroma, which includes cancer-associated fibroblasts (CAFs), vascular cells, extracellular matrix, and immune/inflammatory cells. TCs do not usually constitute a homogeneous cell population. They are rather composed of functionally heterogeneous populations that present different cellular states dynamically evolving over time. The notion that tumors are composed by heterogeneous subpopulations of TCs with different histology, karyotype, growth rates, enzymes, and response to cytotoxic drugs has been known for decades (Heppner, 1984). Initially, tumor heterogeneity has been attributed only to genetic diversity arising from clonal evolution, which is discussed in the review by Swanton and colleagues (Vendramin et al, 2021). This gene-centric hypothesis has been challenged by the discovery of functional diversity of TCs. Functional assays, such as in vitro clonogenic assays, transplantation, and in vivo lineage tracing, have suggested that some tumors are hierarchically organized and present a population of cells called cancer stem cells (CSCs) that sustain tumor growth by giving rise to TCs with more restricted proliferative potential (Lapidot et al, 1994; Beck & Blanpain, 2013; Prager et al, 2019). These two concepts are not mutually exclusive, as populations of CSCs can exhibit substantial intra-tumoral genetic heterogeneity (Shipitsin et al, 2007). Historically, similar to the primary tumors, metastasis-initiating cells have been initially dominated by the genetic model of tumor evolution. Although additional driver mutations can be found during metastatic dissemination (Yates et al, 2017; Nayar et al, 2019), metastases usually do not present driver mutations exclusively found in metastasis and not in primary tumors, suggesting that other mechanisms besides genetic evolution can drive the metastatic dissemination (Birkbak & McGranahan, 2020; Massague & Ganesh, 2021). Growing evidence indicates that intra-tumoral heterogeneity in primary tumors and metastasis is not only determined by genetic and epigenetic features in cancer cells but can also be influenced by the tumor microenvironment. Historically, researchers had studied functional heterogeneity in cancers using different functional assays. Inspired by developmental and stem cell biology, cancer biologists had identified cell surface markers that are heterogeneously expressed within a given tumor and separated TCs into distinct subpopulations and assessed their clonogenic and differentiation potential by transplantation experiments using limiting dilutions. Using this approach, Dick and colleagues demonstrated for the first time, the existence of a small population of leukemic cells that were much more efficient at forming secondary leukemia than the bulk of cancer (Lapidot et al, 1994; Bonnet & Dick, 1997) and called this subpopulation leukemic stem cells. This approach has been used to identify, characterize functionally and molecularly CSCs in many different cancers (Al-Hajj et al, 2003; Nassar & Blanpain, 2016). Whereas isolation followed by transplantation in vivo or clonogenic assays in vitro had been widely used to characterize CSCs, these assays have their limitations as cancer cells are dissociated from their surrounding microenvironment and transplanted very often in a heterotopic site. To study tumor heterogeneity within its native microenvironment, researchers have used lineage tracing to label TCs in situ and studied their ability to change fate and their clonogenic potential over time. These studies in different types of solid tumors have demonstrated that not all TCs are equal and clonogenic, and only a small population survives long term and fuels the tumor growth (Nassar & Blanpain, 2016). Other methods such as pulse-chase experiments have been used to characterize proliferation heterogeneity within the tumor (Fillmore & Kuperwasser, 2008; Pece et al, 2010; Roesch et al, 2010; Schober & Fuchs, 2011; Brown et al, 2017). The same approaches using cell surface markers have been used to identify TC populations with increased metastatic potential called metastasis-initiating cells (Celia-Terrassa & Kang, 2016; Pascual et al, 2017; Massague & Ganesh, 2021) or to deconvolute tumor heterogeneity occurring during epithelial-to-mesenchymal transition (EMT) and identify TC populations enriched for metastasis-initiating cells (Pastushenko et al, 2018). However, these approaches are biased toward the availability of cell surface markers that recognize the different TC populations, or their physical or chemical characteristics such as the expression of aldehyde dehydrogenase (Moreb, 2008; Luo et al, 2012). Nevertheless, bulk RNA and DNA sequencing of human tumors allowed the identification of genetic and transcriptional heterogeneity between different tumors, giving rise to more clinically relevant, molecular-based classifications (Verhaak et al, 2010; Cancer Genome Atlas, 2012a, 2012b, 2015; Brennan et al, 2013). However, bulk sequencing approaches average the genetic and expression profiles of the different tumor subpopulations and are not very powerful in identifying the distinct tumor states that compose the tumors. The development of new technologies based on single-cell sequencing opened new avenues to capture different tumor states and understand intra-tumoral heterogeneity with unprecedented resolution and scale. Single-cell RNA sequencing (scRNA-seq) allows to define the transcriptome of individual TCs and to identify clusters of cells (cell states) presenting similar gene expression profiles within a tumor. During the last years, several tumor states, including proliferative, differentiated, invasive, hypoxic, metastatic, and stress-like states, have been identified and functionally characterized. Each tumor state is associated with different hallmarks of cancer, such as tumor progression, metastasis, and resistance to therapy. Single-cell profiling also allows the identification of stromal cells including CAFs, immune cells, and endothelial cells that compose the tumor microenvironment (Fig 1). Single-cell “-omics” (genomics, transcriptomics, epigenomics, and proteomics) technologies are evolving at a very rapid pace (see Box 1). In this review, we will summarize the most recent single-cell studies that allowed the identification of tumor functional states (Table 1), their spatial organization, and their dynamics during tumor progression, metastasis, and response to therapy. Figure 1. Single-cell RNA sequencing to unravel tumor heterogeneity Schematic showing how tumors derived from different models can be used for transcriptional analysis at the single-cell level of both malignant and stromal cells. CAF: cancer-associated fibroblasts; EMT: epithelial-to-mesenchymal transition; GEMM: genetically engineered mouse model; PDX: patient-derived xenograft. Download figure Download PowerPoint Table 1. Selected studies that used single-cell RNA sequencing to understand tumor cell states. Reference/Year Species Model Tumor type scRNA-seq analysis on Tumor cells Stromal cells Kinker et al (2020) Human Cancer cell lines 198 cell lines from 22 cancer subtypes X Kumar et al (2018) Mouse Syngeneic tumors Melanoma, breast, lung, colon carcinomas, and fibrosarcoma X X Cook and Vanderhyden (2020) Human Cancer cell lines Lung, prostate, breast, and ovarian cancer cell line X Chung et al (2017) Human Primary, metastatic lymph nodes Breast cancer X X McFaline-Figueroa et al (2019) Human Cancer cell line X Davis et al (2021) Human PDX, tumor, lymph node, and lung metastasis X Deshmukh et al (2021) Human Cancer cell line X Karaayvaz et al (2018) Human Primary tumors Triple-Negative Breast Cancer X X Pal et al (2021) Human (healthy, pre-neoplastic, tumoral) tissue + lymph nodes Mammary duct and breast cancer X X Li et al (2017) Human Primary tumors CRC X X Lee at al. (2020) Human Primary tumors X X Gojo et al (2020) Human Primary tumors, PDX Ependymoma X X Gillen et al (2020) Human Primary tumors X Neftel et al (2019) Human Primary tumors Glioblastoma X Patel et al (2014) Human Primary tumors X Pine et al (2020) Human Organoids, PDX x Filbin et al (2018) Human Primary tumors, PDX Glioma X Venteicher et al (2017) Human Primary tumors Gliomas (Oligodendroglioma and Astrocytoma) X Hovestadt et al (2019) Human Primary tumors Medulloblastoma X Tirosh et al (2016) Human Primary tumors Oligodendroglioma X Yao et al (2020) Mouse 4-NQO induced primary tumors Esophageal SCC X Wu et al (2018) Human Primary tumors Esophageal SCC and ADC X Puram et al (2017) Human Primary tumors and metastatic lymph nodes Head and Neck cancer X X Chen et al (2020) Human Primary tumors Nasopharyngeal carcinoma X X Zhao et al (2020) Human Primary tumors X X Kim et al (2020) Human Primary tumors, pleural fluids, lymph node, and brain metastasis Lung adenocarcinoma X X Laughney et al (2020) human primary tumors + metastasis X X Quinn et al (2021) Human PDX cancer cell lines, primary tumors, and metastasis Lung cancer X Ireland et al (2020) Mouse Cell lines derived from primary tumors Small Cell Lung Cancer X Tirosh et al (2016) Human Primary tumors and lymph nodes + distant metastasis Melanoma X X Wouters et al (2020) human cell lines derived from primary tumors X Pastushenko et al (2018) Mouse Primary tumors Skin SCC X Ji et al (2020) Human Primary tumors, PDX X X Hu et al (2020) Human Primary tumors Ovarian and endometrial cancer X X Izar et al (2020) Human Cells isolated from ascites Ovarian cancer X X Peng et al (2019) Human Primary tumors Prostate carcinoma X Chen et al (2021) Human Primary tumors, one lymph node metastasis Prostate carcinoma X X Young et al (2018) Human Primary tumors Renal cancer X Box 1 Single-cell sequencing has rapidly evolved over the last 10 years and has revolutionized the ability to interrogate tumor heterogeneity. These approaches enable to investigate the transcriptional, genomic, epigenomic, and proteomic features of thousands of individual cells within a tumor. Although scRNA-seq has largely contributed to the identification of TC states, analysis of other layers of cell features, either on their own or integrated with transcriptomic interrogation, have greatly expanded our understanding of cancer biology. Single-cell whole-genome sequencing (scWGS) is a powerful tool to decipher cell heterogeneity in a biological sample. Several single-cell genome amplification techniques have been developed, such as DOP-PCR, multiple displacement amplification (MDA), multiple annealing and looping-based amplification cycles (MALBAC), and linear amplification via transposon insertion (LIANTI). These techniques can accurately call copy number variations, indels, and single-nucleotide variations (Navin, 2015; Mallory et al, 2020). High-dimensional single-cell DNA sequencing in a clinical context is limited by its high cost. One option is to apply single-cell targeted sequencing of the genetic aberrations of interest identified by bulk sequencing. Techniques of such type carry the bias of seeking pre-identified mutations leaving out the discovery of new ones (Rodriguez-Meira et al, 2019). Epigenetics plays a pivotal role in cell biology as different modifications on DNA (e.g., methylation) and histones, as well as varying chromatin accessibility and organization, tune gene expression (Kelsey et al, 2017). Single-cell methods that analyze the epigenetic landscapes of thousands of single cells have rapidly expanded. Single-cell chromatin immunoprecipitation sequencing (scChIPseq) detects direct binding of transcription factors onto DNA (Rotem et al, 2015); high-throughput chromosome conformation capture (Hi-C) determines high-order chromatin organization (Nagano et al, 2015); DNA modifications such as methylation can be identified by single-cell bisulfite sequencing (sc-BSseq) or inferred by the use of restriction enzymes whose activity depends on the methylation state of DNA (Guo et al, 2013; Cheow et al, 2015). Single-cell chemical-labeling-enabled C-to-T conversion sequencing (CLEVER-seq) enables identification of 5-formycytosine (5fc), where C is read as a T after specific chemical labeling (Zhu et al, 2017). A wide application in single-cell -omics has been found for the assay for transposase-accessible chromatin sequencing (ATAC-seq) that interrogates chromatin accessibility based on the ability of the Tn5 transposase to add sequencing adapters into open chromatin regions (Buenrostro et al, 2015; Cusanovich et al, 2015). As opposite to single-cell technologies detecting nucleic acids, analysis of the proteome in a cell progresses at a slower pace. Mass cytometry, commercialized as CyTOF (mass cytometry by time of flight), relies on metal-isotope-conjugated antibodies for immunolabeling, and analyses them by mass spectrometry (MS) and has emerged as a powerful tool in the field of single-cell proteomics as it can measure 40 to 100 parameters in each cell (Bandura et al, 2009; Bendall et al, 2011). Increased throughput and peptide quantification are the advantages of an improved mass cytometry-based platform called SCoPE-MS (Budnik et al, 2018). Based on the same technology, imaging CyTOF has been developed and can be applied to tissues on slides to combine proteomics and spatial architecture (Giesen et al, 2014). Similarly, MALDI imaging mass spectrometry relies on a matrix that coats a tissue sections, extracts molecule from the tissues, and generates mass spectra that can be matched with histological staining of the section (Norris & Caprioli, 2013). Importantly, many of the techniques described above have been integrated into single-cell multi-omics. Generally, scRNA-seq is included in most scMulti-omics studies as gene expression data represent a necessary tool to decipher cell processes. Simultaneous characterization of different layers within a single cell has already taken off and will continue to provide valuable information about cell identities. For a detailed and comprehensive discussion on the integrative approaches used for scMulti-omics, please refer to reviews (Ma et al, 2020; Longo et al, 2021). Tumor states in solid tumors Certain tumors hijack the developmental programs of normal tissues and mimic their cellular hierarchy leading to TCs with high self-renewing capacities known as CSCs and TCs that are more differentiated and have decreased tumor-initiating capacity. Single-cell sequencing studies have described the similarities of tumor subpopulations and developmental lineages. Developmental programs in brain tumors A pioneering work using scRNA-seq identified different tumor states in primary human glioblastomas. These states include cycling, quiescent, hypoxic, and a continuum of stem-like tumor states. Interestingly, malignant cells invading the surrounding tissue express a transcriptional program characterized by low hypoxia, low proliferation, and high migratory capacity (Patel et al, 2014). In glioblastoma, four distinct tumor states were found that recapitulate the different neural cell types including neural progenitor-like, oligodendrocyte-like, astrocyte-like, and mesenchymal-like. Each state contains proliferative cells, although higher proliferation is observed in the neural- and oligodendrocyte progenitor-like states (Neftel et al, 2019). Mass cytometry, immunostaining, and xenografting experiments demonstrated that among the different glioblastoma subpopulations, glial progenitor cells are more proliferative and possess higher tumor formation capacity than TCs with astrocyte and neuronal lineage differentiation. This highly proliferative state is also the most resistant to the alkylating agent temozolomide (Couturier et al, 2020). Importantly, specific genetic alterations promote the relative abundance of the different tumor states. For example, mutations in EGFR promote the astrocyte-like state abundance, whereas TCs harboring chromosome 5q deletion preferentially differentiate into a mesenchymal-like state. In human IDH1 or IDH2 mutant oligodendroglioma and astrocytoma, the majority of cancer cells differentiate along two glial transcriptional programs: oligodendrocytes (characterized by OLIG1, OLIG2, or OMG expression) and astrocytes (characterized by APOE, ALDOC, or SOX9 expression) (Tirosh et al, 2016b). In addition, a rare subpopulation of undifferentiated TCs, associated with a neural stem cell expression program, is enriched for proliferation signature, suggesting that this stem-cell-like population fuels tumor growth (Tirosh et al, 2016b). Further analysis of scRNA-seq data showed that these undifferentiated TCs exhibit a strong similarity in gene expression profile between the two tumor histotypes, raising the possibility of a shared cell of origin for IDH-mutant oligodendroglioma and astrocytoma (Venteicher et al, 2017). Similarly, in H3K27M glioma, the stem-like state proliferates more actively and presents higher clonogenic capacity upon transplantation as compared to the more differentiated states (Filbin et al, 2018). Medulloblastoma, a childhood cerebellar tumor, comprises four molecular subgroups associated with different oncogenic mutations and transcriptional landscapes (Liu et al, 2020). WNT, SHH, and Group 3 Medulloblastoma contain both stem-like and more differentiated tumor states. The differentiation of the SHH subgroup resembles cerebellar granule neurons, whereas Groups 3 and 4 resemble neuronal-like cells in different proportions (Hovestadt et al, 2019). In ependymoma, a similar hierarchical organization of cell states including stem-like and differentiated tumor states was found. The stem-like state is associated with a poorer prognosis compared to tumors with differentiated-like states (Gojo et al, 2020). Further investigation of ependymoma subgroups revealed the presence of two subpopulations with ependymal differentiation features (cilia function and cellular transport) and an undifferentiated subpopulation associated with clinical aggressivity (Gillen et al, 2020). Overall, it appears that most brain tumors follow well-defined hierarchical architectures, with stem-like tumor states driving tumor growth and giving rise to more differentiated TCs. Understanding the molecular mechanisms underlying differentiation of TCs can be exploited as a valuable strategy to hinder tumor growth and lead to clinical benefits. Inspired by the treatment of acute promyelocytic leukemia with retinoic acid and arsenic trioxide that triggers leukemia cell differentiation and disease elimination (de The, 2018), new insights in the molecular basis of differentiated tumor states would help designing pro-differentiation therapies in solid tumors. Normal differentiation and tumor states As with brain tumors, pairwise scRNA-seq analyses of tumors and matched healthy tissues show that the different tumor states recapitulate the spectrum of differentiation found in the normal tissue. For example, in colorectal cancer, tumor states include enterocytes, goblet cells, and stem cells (Li et al, 2017). In normal fallopian tube epithelium, which is the cell of origin of ovarian cancer, four different secretory cell types, including EMT-like cluster and ciliated cell cluster, were identified by scRNA-seq. The ciliated subtype is enriched in low-grade ovarian cancer, whereas the EMT subtype is associated with high-grade serous ovarian cancer and poor overall survival (Hu et al, 2020). In triple-negative breast cancer (TNBC), clustering of single cancer cells using bulk RNA-seq-derived signatures of normal basal, luminal progenitors, and differentiated luminal cells of the mammary gland showed that most TCs express the luminal progenitor signature, consistent with these cells being the cell of origin of TNBC (Lim et al, 2009; Molyneux et al, 2010; Van Keymeulen et al, 2015). Similar results were obtained with unsupervised clustering showing the presence of a basal-like signature and that most tumors contain a subpopulation resembling luminal differentiated cells (Karaayvaz et al, 2018). Another study analyzed the cellular composition and heterogeneity of non-pathologic mammary gland and breast cancers in a large cohort of patients. Epithelial cell populations do not differ across healthy patients, whereas the tumor microenvironment composition—particularly fibroblasts—changes between pre- and post-menopausal subjects. The largest proliferative population is found in TNBC rather than ER+ and Her2+ breast cancers. EMT-expressing TCs did not appear as a discrete cluster but were scattered throughout different subclusters across the three breast tumor subtypes. Among EMT-related genes, TNBCs and ER+ tumors expressed higher vimentin than HER2+ (Pal et al, 2021). Pancreatic ductal adenocarcinomas display two types of ductal tumor states expressing typical ductal lineage markers. Nevertheless, the type 2 ductal tumor state expresses much higher levels of pancreatic adenocarcinoma markers, whereas type 1 ductal cells express higher level of genes regulating normal pancreatic functions, including digestion, pancreatic secretion, and bicarbonate transport. However, these cells can be further clustered into two groups, one resembling normal ductal cells and the other similar to malignant type 2 ductal cells (Peng et al, 2019). Lineage trajectory analysis suggests that type 1 ductal cells are the cells of origin giving rise to type 2 malignant TCs. The malignant ductal cell markers were used to cluster human pancreatic tumor samples. Proliferative ductal markers are more enriched in two of the three pancreatic adenocarcinoma clusters that are associated with lower survival rate. CDK1, PLK1, and AURKA are markers of proliferative ductal cells, and the pancreatic cancer cell line MIA PaCa-2 growth was suppressed by inhibitors targeting these three proteins (Peng et al, 2019). ScRNA-seq of normal lung epithelia and lung adenocarcinoma identified four common differentiated lineages (including alveolar epithelial cell types 1 and 2, ciliated and club cells). In lung adenocarcinoma, two new tumor states that are observed during the regeneration of severely injured lung can be identified (Laughney et al, 2020). These two states include SOX2-derived KRT5+ basal-like cells, which exhibit increased RAS signaling and mesenchymal gene enrichment associated with wound response and the other one corresponds to SOX9-expressing alveolar epithelial progenitors. (Vaughan et al, 2015; Zuo et al, 2015; Laughney et al, 2020). Different states of melanoma Whereas using bulk sequencing, melanoma could be classified as MITF-high or AXL-high, at the single-cell level every tumor contains malignant cells corresponding to both states. Melanoma cells with the AXL program are selected and enriched following treatment with RAF and MEK inhibitors. CAF abundant tumors are enriched for the AXL-high signature, suggesting that CAFs promote the AXL-high tumor state (Tirosh et al, 2016a). Melanomas comprise cycling states and non-cycling states, the latter showing high expression of the histone demethylase KDM5B. Within the cycling population, a high-cycling state with unique high expression of cyclinD3 differs from the low-cycling state. In cell lines characterized by a low percentage of AXL+ cells, the treatment with BRAF/MEK inhibitors promotes the enrichment of AXL+ cells leading to drug tolerance, suggesting that a rare tumor-resistant state exists before treatment (Tirosh et al, 2016a). Moreover, gene regulatory network and trajectory inference in patient-derived cancer cell lines indicate that the increase in AXL expression occurs during the transition from the melanocytic tumor state (characterized by the expression of SOX10 and MITF transcription factors) to the mesenchymal tumor state (characterized by the expression of SOX9 and AP-1) via an intermediate state expressing SOX6 and displaying simultaneous melanocytic (pigmentation) and mesenchymal-like (increased migration) phenotypes. Consistent with a role of SOX10 in regulating the melanocytic state, SOX10 knockdown results in the downregulation of makers of the melanocytic lineage and the increase in the expression of mesenchymal-like genes (Wouters et al, 2020). Altogether, these studies in melanoma show that scRNA-seq analysis allows the identification of dynamically regulated cellular states with different properties. Insights from mouse models Tumor mouse models can be helpful in deciphering functional heterogeneity, allowing longitudinal studies from tumor initiation to late stages of tumor progression. Single-cell analyses performed during tumor initiation in a chemically induced mouse model of esophagus squamous cell carcinoma (SCC) revealed different tumor states across different stages of tumor development including hyperplasia, dysplasia, and carcinoma. This study proposed a model where, during carcinogenesis, proliferating cells either switch to a malignant state progressively acquiring gene signatures typical of EMT, angiogenesis, immunosuppression, and invasiveness, or differentiate. The malignant switch is associated with the increase in transcription factors such as Snai3 and Ets1 and the decrease in tumor suppressor gene expression (Trp53, Pit1, and Bclaf1) (Yao et al, 2020). The progression through the malignant phenotype was accompanied by gradual decrease in Notch1 expression, consistent with the high proportion of Notch1 mutations in SCCs (Dotto, 2009; Sanchez-Danes & Blanpain, 2018; Yao et al, 2020). A study in a mouse model of small cell lung carcinoma (SCLC)—a neuroendocrine tumor classified into four molecular subtypes (SCLC-A, SCLC-N, SCLC-P, and SCLC-Y)—provided deeper insights in the understanding of the mechanisms underpinning cell phenotype plasticity. The authors showed that mutated-MYC overexpression in a pulmonary neuroendocrine cell of origin, in the context of Trp53 and Rb1 deletion, promotes the SCLC-N and SCLC-Y, but not SCLC-P, molecular subtypes (Ireland et al, 2020). scRNA-seq and pseudotime ordering of four tumors developed in this mouse model showed that MYC drives the evolution of SCLC fate from neuroendocrine to non-neuroendocrine states. Mechanistically, MYC increases NOTCH signaling to destabilize the neuroendocrine identity during SCLC evolution. Therefore, this work sheds light on the mechanisms that determine SCLC subtypes, demonstrating that genetic (MYC mutation), cell of origin (neuroendocrine cell), and cell plasticity matter in SCLC evolution (Ireland et al, 2020). In conclusion, mouse models allow to interrogate the tumor states at different stages of tumor progression. In addition, the genetic overexpression or downregulation of candidate genes and transcription factors in vivo allows the identification of precise molecular mechanisms regulating cell state transitions in the native tumor environment. Spatial organization and