Understanding the nature of cancer initiation and progression has experienced wild fluctuations, from the initial chaos of phenomenological description of cancer to the attempts of simplifying cancer regulation to single molecular events, and then followed by indefinite complexity of genetic and phenotypic heterogeneities and regulatory mechanisms (Weinberg, 2014). At present, about five million cancer literatures have been published. They showed that almost all aspects of biological research play a role in cancer, and nearly every gene is associated with cancer (de Magalhães, 2022). Some hypotheses/concepts have been proposed to generalize the rules behind the complexity of tumorigenesis, including notably the mutation theory, aneuploidy and chromosome instability, Warburg effect, epithelial-to-mesenchymal transition (EMT), etc. However, none of them can integrate the complexity of cancer because they explain tumorigenesis in some aspects but meet serious challenges in others (Soto and Sonnenschein, 2011; Paduch, 2015; Hanselmann and Welter, 2016). Notably, cancer was proposed as a disorder of developmental dynamics (Rubin, 1985), and the key signaling pathways involved in early embryogenesis also play important roles in cancers, for instances, the TGFβ, Wnt, FGF, Notch pathways. But how the complex process of embryogenesis is intrinsically linked with the complex process of tumorigenesis has remained a challenging question, particularly when considering that a normal embryogenesis needs a process of fusion of gametes. Cancer research is not the business only for biological and medical researchers. Astrophysicists proposed cancer as an atavistic reversion effect (Lineweaver et al., 2021). Moreover, quantum physics, the study on the discrete units of matter and energy throughout the Universe, is also suggested as the governing rule of cancer (Hameroff, 2004; Bordonaro, 2019; Laster et al., 2019). Despite the universality of physics, it is critical to find a particular link to integrate different data and phenomena, distinguish causality and associations, and figure out causal and supporting factors from the observations of cancer research. In the review, I will discuss my findings about the central role of neural stemness in cell tumorigenicity and pluripotent differentiation potential. This gives rise to a conceptual paradigm that might integrate different features of tumorigenesis. I propose that tumorigenesis might represent a distorted replay of neural induction and subsequent tissue differentiation during embryogenesis, a critical process required for neural development and normal body axis formation.

How the nervous system is formed and body axis is established had been a primary focus of embryological study. Almost a century ago, a paramount work done by Spemann and Mangold showed that a small group of cells, the dorsal blastopore lip of a newt gastrula embryo, were capable of inducing formation of secondary body axis or a conjoined twin when they were transplanted to the ventral side of a host gastrula embryo. The secondary body axis, which contained neural tube, somites, pronephros and gut, was derived from the host, whereas the transplanted dorsal blastopore differentiated mostly into notochord (Spemann and Mangold, 1924; Spemann and Mangold, 2001). The dorsal lip was then named as the “Spemann-Mangold organizer”. An ectopic organizer induces formation of a conjoined twin containing neural tissues in a gastrula embryo. Vice versa, various studies showed that an embryo in the absence of organizer activity forms a “belly piece” only that contains no neural and dorsal structures (Spemann, 1938; Gerhart, 2001; De Robertis, 2009; Sosa et al., 2019).

Subsequent pursuit of the mechanisms underlying the induction of neural tissue and body axis by organizer was a failure in nearly 6 decades after the dorsal blastopore transplantation experiment (De Robertis, 2009). Revival of the research began until the end of 1980s because of a critical observation. Amphibian blastula ectodermal explants differentiated into only epidermal tissues when they were cultured in neutral saline in vitro. Surprisingly, when the explants were disaggregated into single cells for a few hours first and then re-aggregated again, ectodermal cells differentiated into neural cells exclusively (Godsave and Slack, 1989; Grunz and Tacke, 1989; Sato and Sargent, 1989), suggesting that removal of an extracellular signal is required for the ectoderm to adopt a neural fate, and neural fate might be the “default fate” of ectodermal cells. Afterwards, genes with localized expression in the organizer that showed the activity of neural induction and secondary axis formation were identified, including noggin, chordin, cerberus, etc. (Smith and Harland, 1992; Sasai et al., 1994; Bouwmeester et al., 1996; Harland, 2000; De Robertis and Kuroda, 2004; De Robertis, 2006). They encode for secreted proteins that inactivate the signaling pathways promoting epidermalization of ectoderm and ventralization of body axis, particularly the BMP signaling, via direct binding to the BMP ligands (Harland, 2000; De Robertis and Kuroda, 2004; De Robertis, 2006; Anderson and Stern, 2016). Moreover, inhibition of the receptor for activin, a BMP-related ligand of the TGF-β family, led to neuralization of ectoderm in absence of inducing factors and rescue of ventralized embryos (Hemmati-Brivanlou and Melton, 1994). In contrast to the initial aim for finding neural inducers, these lines of evidence demonstrate that the fate of ectoderm is neural by default and epidermal fate is induced, and the organizer promotes neural fate by inhibiting the signals that promote epidermal fate in ectoderm. This is the “neural default model” of ectoderm (Weinstein and Hemmati-Brivanlou, 1997; Muñoz-Sanjuán and Brivanlou, 2002).

Neural induction is a prerequisite for body axis formation, a process including differentiation of not only the nervous system, but also differentiation of mesodermal and endodermal tissues such as somite and gut. Neural induction means activation or upregulation of a spectrum of neural genes, forming a regulatory network defining neural plate, the undifferentiated precursor tissue of the central nervous system. Ectopic organizer activity led to the formation of conjoined embryos. Likewise, ectopic stimulation of genes with specific or enriched expression in embryonic neural cells, for example, the proto-oncogenes eed, yy1, ski, egfr, erbb2, erbb4 in Xenopus, gelsolin and msxB in zebrafish, also causes formation of a partial secondary body axis or a conjoined twin, which contains both neural and non-neural tissues (Amaravadi et al., 1997; Satijn et al., 2001; Kanungo et al., 2003; Nie and Chang, 2006; Phillips et al., 2006). These genes are components of the regulatory network for embryonic neural cells. Their ectopic expression activates the neural regulatory network, leading to gain of neural fate in non-neural cells and formation of a second body axis. By contrast, disruption of embryonic neural genes causes defects in neural and axial differentiation in mouse embryos, ultimately leading to developmental arrest at early stages (Faust et al., 1995; Gassmann et al., 1995; Threadgill et al., 1995; Berk et al., 1997; Britsch et al., 1998; Donohoe et al., 1999). Moreover, neural plate specifies somite size and is required for somite development during early embryogenesis (Mariani et al., 2001). These results suggest the critical importance of embryonic neural genes and neural precursor cells in tissue differentiation. Neural stemness, refering to the collective property of primitive neural stem cells (NSCs), neural crest cells and adult NSCs, represents the general stemness, defining tumorigenicity and pluripotent differentiation potential, a notion that has been experimentally clarified (Clarke et al., 2000; Tropepe et al., 2001; Xu et al., 2021; Cao, 2022; Zhang et al., 2022). This unique property of neural stemness might be predestined by the evolutionary advantage of neural regulatory networks (Xu et al., 2021; Cao, 2022). Contribution of neural stemness to formation of the nervous system is self-evident. However, its contribution to non-neural differentiation is not. Neural crest cells are pluripotent and share regulatory network with cleavage stage embryos, differentiating into peripheral nervous system and many types of non-neural tissues/cells, such as melanocytes, skeletal and connective tissues, and medulla cells of the adrenal gland, etc. Locating between neural plate and epidermal ectoderm, neural crest is induced by interactions between neural plate and adjacent tissues (Selleck and Bronner-Fraser, 1995; Knecht and Bronner-Fraser, 2002; Buitrago-Delgado et al., 2015; Gilbert and Barresi, 2016; Pla and Monsoro-Burq, 2018). This means that pluripotency of neural crest cells is ultimately derived from neural plate cells. Neuromesodermal progenitors in the most posterior region of elongating embryos give rise to both spinal cord and paraxial mesoderm. These cells are presumably originated from anterior neural plate (Henrique et al., 2015; Sambasivan and Steventon, 2021). Therefore, neural induction generates neural precursor cells, which give rise to the differentiation of not only neural tissues, but non-neural tissues as well.

The studies above on neural induction and body axis formation were primarily performed with amphibian species, the newts and African clawed frog (Xenopus laevis). The functional homologue has been identified in all classes of vertebrates, such as fish, bird and mammalian embryos, which is known as the node. It exhibits the activity of neural induction and body axis formation similar to the organizer through conserved molecular mechanisms (Gerhart, 2001; Thisse and Thisse, 2015; Martinez Arias and Steventon, 2018). The neural default state exhibited by amphibian blastula ectodermal cells is also exhibited by mammalian embryonic stem cells (ESCs). Amphibian blastula ectodermal cells are the equivalents of ESCs, since they have the potential of differentiation into cell types of all three germ layers. ESCs are usually cultured in medium containing high-concentration of fetal bovine serum. They adopt a neural fate and turn into primitive NSCs when cultured in defined serum-free medium (Tropepe et al., 2001; Ying et al., 2003a; Smukler et al., 2006). In this cell fate transition, BMP signaling plays a critical role in inhibiting neural fate in ESCs (Ying et al., 2003b; Malaguti et al., 2013), a similar mechanism as observed in amphibian ectodermal cells.

In summary, either extracellular signals by the organizer or node or ectopic activation of neural genes in non-neural cells can cause the gain of neural fate in non-neural cells during gastrulation, i.e., neural induction, and leads to the formation of a secondary body axis or a conjoined twin. This is the paradigm for understanding how neural tissue and body axis are initiated to form during early embryogenesis. Neural induction is fundamental for neural development and embryogenesis. Nevertheless, aberrant occurrence of neural induction or a similar process might be associated with some most sophisticated pathological effects. It was proposed that a conjoined twin is formed when a secondary organizer-like activity is present in a gastrulating embryo, such as a human embryo (Levin, 1999). A neural induction-like process could also occur erroneously in cells of postnatal animals, which might be the general cause of tumorigenesis.

As analyzed above, neural induction during embryogenesis means that non-neural (ectodermal) cells turn into neural precursor cells in response to either an extracellular signal inhibiting non-neural cell property or intracellular stimulation of embryonic neural genes. Ectopic neural induction during gastrulation causes formation of a conjoined twin. Neural stemness is characterized as the essential property of tumorigenic and pluripotent cells, and tumorigenesis could be envisioned a progressive loss of original cell identity and gain of neural stemness in postnatal cells (Southall et al., 2014; Cao, 2017; Li et al., 2020; Cao, 2022; Zhang et al., 2022), reflecting a neural induction-like effect. One obvious example for the comparability of tumorigenesis as a conjoined twin formation should be the teratocarcinomas/teratomas, which are composed of disorganized but histologically identifiable tissues or organs derived from all three germ layers, such as undifferentiated neural epithelial tissue and differentiated nerves from ectoderm, gut and glandular tissues from endoderm, and cartilaginous and muscle tissues from mesoderm. Teratocarcinomas/teratomas are usually found in the gonads, but they can also form in extragonadal tissues/organs (Chao et al., 2004; Gatcombe et al., 2004; Singhal and Jhavar, 2008; Agrawal et al., 2010). The mechanism underlying teratocarcinoma/teratoma formation has been rarely reported. In mouse, an inactivation mutation in the gene Dnd1, which encodes a master regulator for vertebrate germ cell development, causes progressive loss of germ cells and incidence of testicular teratoma (Youngren et al., 2005; Liu and Collodi, 2010). The pluripotent property of embryonal carcinoma cells (ECs) derived from teratocarcinoma was well characterized, which enlightened subsequent studies on pluripotency of ESCs (Andrews, 2002; Solter, 2006; Barbaric and Harrison, 2012). ECs form teratocarcinomas when transplanted into immunodeficient mouse hosts, and contribute to formation of chimeric embryos when transferred into blastocysts. EC pluripotent cell lines are characteristic of neural precursor or progenitor cells, they can be differentiated into neurons when treated with retinoic acid (Pleasure and Lee, 1993; Bain and Gottlieb, 1998; Negraes et al., 2012), a reagent inducing neuronal differentiation from NSCs. Neural stem/progenitor cells, which are tumorigenic and are capable of teratoma formation in immunodeficient mice (Xu et al., 2021), were isolated from teratocarcinoma/teratoma (Hasgekar et al., 1996; Kim et al., 2019). Therefore, ECs are characteristic of neural stemness, tumorigenicity and pluripotent differentiation potential. Like neural induction during embryonic development, teratocarcinomas/teratomas are the consequence of progressive loss of original cell identity and gain of neural stemness, which contributes to tumorigenicity and pluripotency (Xu et al., 2021; Cao, 2022; Zhang et al., 2022), in either germ or somatic cells.

A much broader range of tumorigenesis is the formation and progression of cancers that have been found in most postnatal tissues/organs. Growing evidence has shown that cancer cells, or generally tumorigenic cells, are characteristic of NSCs. Like NSCs and ECs, cells from different cancer types are capable of neuronal differentiation in response to inhibition of endogenous cancer promoting factors, which have a specific or enriched expression in embryonic neural cells during vertebrate embryogenesis, and play essential roles in maintaining neural stemness in both cancer cells and NSCs (Lu et al., 2017; Zhang et al., 2017; Lei et al., 2019; Chen et al., 2021; Zhang et al., 2022). In general, most (if not all) cancer-promoting genes are neural stemness genes or genes with enriched expression in embryonic neural cells during vertebrate embryogenesis. Cancer cells share both regulatory networks and cell property with NSCs or embryonic neural cells (Zhang et al., 2017; Cao, 2022). By contrast, non-neural tissue-specific genes and genes promoting differentiation are downregulated/silenced in cancer cells, and a substantial part of tumor suppressor genes are non-neural genes during embryogenesis (Cao, 2017; Zhang et al., 2017; Cao, 2022). This mode of expression change of cancer related genes reflects the progressive loss of original cell identity and gain of neural stemness, and consequently, the gain of tumorigenicity and pluripotency in cancer cells during tumorigenesis. For example, when the key muscle differentiation gene Myod1 was knocked out in myoblast cells, the cells lost their myoblast identity and gain of neural stemness, tumorigenicity and pluripotent differentiation potential (Xu et al., 2021). Correspondingly, a recurrent mutation in MYOD1 leads to a dominant-negative product that inhibits the function of wild-type protein in rhabdomyosarcomas (Kohsaka et al., 2014; Rekhi et al., 2016). Intestinal stem cells in Drosophila turned into a NSC-like state in response to the loss of a transcription repressor, and consequently, caused the formation of neuroendocrine tumor (Li et al., 2020). Neurons can also be dedifferentiated into an NSC-like state when a factor repressing NSC and cell cycle genes and maintaining neurons in a differentiated state was removed, leading to acquirement of tumorigenicity and tumor formation (Southall et al., 2014).

Unlike teratocarcinomas/teratomas that are composed of a haphazard mixture of adult tissues and deformed organs, most other tumors usually do not contain well differentiated tissues and organs. Nevertheless, they are composed of cell types with distinct functional features and/or expression of tissue- or cell type-specific markers, indicating intratumor phenotypic heterogeneity. Two mainstream models are proposed to explain how phenotypic heterogeneity is generated. The clonal evolution model emphasizes that phenotypic heterogeneity is a result of genetic heterogeneity arising from Darwinian-like evolution. Nevertheless, how genetic heterogeneity causes phenotypic heterogeneity seems to be not understood at all and not testified experimentally. The cancer stem cell (CSC) model proposes that differentiation of CSCs generates phenotypic heterogeneity (Clevers, 2011; Marusyk et al., 2012; Beck and Blanpain, 2013; Burrell et al., 2013; Meacham and Morrison, 2013; McGranahan and Swanton, 2015; Quintanal-Villalonga et al., 2020), which has been validated in many studies. In vitro generated cells with CSC property can differentiate into cell types expressing neuronal, endothelial and muscle cell markers (Scaffidi and Misteli, 2011). CSCs of glioblastoma give rise to tumor endothelium and vascular pericytes, supporting tumor growth (Ricci-Vitiani et al., 2010; Wang et al., 2010; Cheng et al., 2013). CSCs of colorectal cancer revealed the capacity of multilineage differentiation (Vermeulen et al., 2008). A consensus conceptual framework could not be deduced from these studies about the nature of CSCs except that they can differentiate. It was not clear whether CSCs of different cancer types have the property of stem/progenitor cells of their respective tissues of cancer origin, or all CSCs might have a common property of stemness, or CSCs might be of “cancer-specific” nature that is not comparable to any known stem cell types. In fact, similar to ECs, cancer cells are also pluripotent because xenograft tumors formed by cancer cells show expression of markers of tissue/cell types derived from all three germ layers, for example, SOX1-expressing cells representing cells with neural stemness and derived from ectoderm, ACTA2-expressing cells derived from mesoderm, and AFP-expressing cells derived from endoderm (Xu et al., 2021; Cao, 2022; Zhang et al., 2022). These cells can be widely detected in different cancer types. Public databases show that a majority of transcripts and their protein products have low cancer specificity and are present in many cancer types (www.proteinatlas.org) (Uhlen et al., 2017). For example, BMI1, CDH2, DCLK1, FGFR4, MSI2, and SMARCA4 representing neural stemness; MAP2, NEUROG2, and TUBB3 representing neuronal differentiation; AFP, FOXA3, GATA6, and KRT8 representing endodermal tissue differentiation; and ACTA1, ACTA2, COL1A1, FXR1, and MEF2D representing mesodermal tissue differentiation (Figure 1). Noticeable is that ACTA2 and COL1A1 are also the markers of cancer-associated fibroblasts (CAFs). Therefore, different cancer types contain basic elements representing cell/tissue differentiation, similar to the case of embryogenesis. Cancer cells at different stage of tumorigenesis exhibit different degree of tumorigenicity. They are more similar to the cells of cancer origin and exhibit weak tumorigenicity at the beginning stage of tumorigenesis. Cancer cells at later stage are more dissimilar to the cells of cancer origin and show stronger tumorigenicity. Neural stemness and differentiation potential of cancer cells grow progressively and simultaneously with the progression of cancer, as suggested by a serial xenotransplantation assay of cancer cells (Zhang et al., 2022). Although cancer cells share the regulatory networks and cell property with neural stem or embryonic neural cells, some essential disparities still exist, including extensive defects in genes (differentiation genes in particular) and genome in cancer cells and the difference in the microenvironments with which cancer cells or embryonic neural cells communicate, leading to chaotic differentiation of cancer cells. In some cases, however, tissue/organ formation can be still observed, such as osteoid and bone formation in various cancers (Hoorweg et al., 1997; Goto et al., 2010; Dekkers et al., 2019; Kattepur et al., 2021; Tian et al., 2021). Cancers are degenerated forms of teratomas/teratocarcinomas or embryoid bodies, and tumorigenesis resembles an ectopic neural induction process and consequently embryonic tissue differentiation in postnatal animals. Cancer cells were demonstrated to induce secondary axis formation when transplanted into the appropriate position of a blastula stage of zebrafish embryo, implying a neural induction-like activity (Hendrix et al., 2007). Mutations in Wnt pathway that leads to stabilization and hence extra nuclear accumulation of β-catenin in cells are usually the cause of some cancers, e.g., colorectal cancer. Meanwhile, these mutations are able to cause the formation of a twinned mouse embryo (Brickman and Burdon, 2002), strengthening the intrinsic link between neural induction, conjoined twin formation and tumorigenesis. It is thus plausible that nearly every gene is associated with cancer (de Magalhães, 2022).

EMT has been considered as a general rule governing malignant transformation of epithelial cells. EMT is described as a phenotypic change, in which a polarized epithelial cell loses its polarity and adhesion with neighboring cells, and assumes a mesenchymal phenotype with a motile property. At molecular level, the loss of epithelial phenotype is reflected by downregulation of the epithelial marker CDH1, and gain of mesenchymal phenotype is driven by a core set of EMT transcription factors, SNAI1, SNAI2, TWIST1, ZEB1, and ZEB2. EMT is believed to be the key driver of carcinogenesis and has been extensively investigated (Yang et al., 2020). Nevertheless, it has been heavily controversial because of its essential flaws, which were discussed in details previously (Cao, 2017). The main point is that both epithelial and mesenchymal states are highly heterogeneous among different tissues/organs and cannot be defined universally by a few “epithelial and mesenchymal” markers, and the meaning of “mesenchymal marker” expression change during malignant transformation of epithelial cells should be a misinterpretation (Cao, 2017). Moreover, some cancers are not originated from epithelial cells, but from mesenchymal cells, such as sarcomas. After 2 years of discussion, the EMT International Association (TEMTIA) in 2020 made a consensus statement about the guidelines and definitions of EMT research (Yang et al., 2020). The statement pointed out that the definition of classical EMT cannot reflect the complicated intermediate states between the binary switch from fully epithelial to fully mesenchymal state. Therefore, TEMTIA recommends that the definition of EMT should be more flexible and use “EMT plasticity” (EMP) to describe these intermediate states (Yang et al., 2020). The revised term can now smoothly fit all possible situations encountered during EMT research. However, the mechanisms underlying the boundless plasticity or how the plasticity is derived have not been understood at all. Furthermore, “while the characteristics of fully epithelial cells are relatively clearly defined, our current knowledge does not allow us to define the mesenchymal state with specific cellular characteristic or molecular markers that are universal end-products of all EMT programmes” (Yang et al., 2020). This means that EMT actually represents a transition from a relatively known to an unknown state. Many molecular mechanisms have been described for the regulation of EMT transcription factors in cancer or regulation of cancer progression by EMT transcription factors in numerous publications (Yang et al., 2020). But what the “mesenchymal state” is has remained unknown. In EMT or EMP, either the undefined plasticity (or “dynamics” in other literatures (Brabletz et al., 2021)) or the undefined mesenchymal state is used as a standard reference to define the characteristics of cancer cells. How this dilemma could fit for the regularly advocated logics of biological research, e.g., rigor, precision, and physiological relevance, is intriguing and worthy of pondering.

Cancer research was dominated by a cancer cell-intrinsic view before 1980s since mutations in oncogenes and tumor suppressor genes were seemingly sufficient to determine cancer initiation and progression. This view could not explain smoothly the mechanisms governing cancer metastasis. Studies on tumor microenvironment (TME) and interactions between different cell types in the TME and between tumor and normal tissues might provide reasonable explanations, leading to a shift from the cancer cell-centric view to a tumor environment-centric view (Vogelstein and Kinzler, 1993; Maman and Witz, 2018; Garner and de Visser, 2020). Tumor-host interaction and crosstalks in TME is important for tumor growth and cancer progression. However, understanding the causal and supporting factors involved in the interactions and crosstalks is crucial not only for cancer biology, but also for the development of more efficient strategies of cancer therapy.

A tumor consists of heterogeneous populations of cells. A widely held view is that normal tissue cells infiltrate tumors or cancer cells acquire magic power to hijack normal cells, for example, nerves, immune cells, fibroblasts, blood vessels, etc., and recruit them into TME to promote cancer progression (Maman and Witz, 2018; Saxena and Bhardwaj, 2018; Brandao et al., 2019; Cervantes-Villagrana et al., 2020; Lugano et al., 2020). The crosstalks between cancer cells and recruited normal tissue cells have been a major topic of study in cancer biology. How the cells in the TME are originated and how the functions of the cells are related with tumorigenesis have remained elusive.

Tumorigenicity and pluripotency are key cell properties for tumorigenesis and embryogenesis. The proposal that neural stemness represents the ground state of and unifies cell tumorigenicity and pluripotency implies that tumorigenesis and embryogenesis are driven by a same general principle, the neural induction or a similar process. This provides fresh insights into the essence of tumorigenicity and tumorigenesis. A key property of ESCs is that they differentiate into various cells/tissues and contribute to chimeric formation when transplanted into blastocysts. Similar chimeric formation effects can be observed for NSCs (Clarke et al., 2000; Tropepe et al., 2001). Importantly, various types of cancer cells, including teratocarcinoma, leukemia, neuroblastoma and melanoma cells, can also contribute to chimeric formation or be induced to differentiate into different types of cells when transplanted into an embryo. The differentiated offspring cells are similar to host cells and not tumorigenic anymore (Brinster, 1974; Papaioannou et al., 1975; Illmensee and Mintz, 1976; Cooper and Pinkus, 1977; Gootwine et al., 1982; Podesta et al., 1984; Webb et al., 1984; Gerschenson et al., 1986; Wells and Miotto, 1986; Kulesa et al., 2006; Hendrix et al., 2007). A recent study highlights that the differentiation stage of cancer cells affects the consequence of tumorigenicity (Baggiolini et al., 2021). Moreover, transplantation of the nuclei of different cancer cells into enucleated oocytes led to development of normal embryos (King and DiBerardino, 1965; McKinnell et al., 1969; DiBerardino et al., 1983; Li et al., 2003; Hochedlinger et al., 2004), suggesting the pluripotent nature of cancer cells. Therefore, characterization of cancer cells and NSCs implies that variants of pluripotent state can be numerous and are present throughout the life of an animal from a pre-implantation blastocyst to adult stage. Chimeric formation indicates that pluripotent cells, including cancer cells, NSCs and embryonic pluripotent cells, can be induced to differentiate into normal cells in an embryonic milieu and integrated into the development of an embryo. Instead, they form embryoid structures that cannot integrate into normal differentiated tissues or organs in the absence of embryonic differentiation signals during tumorigenesis in a postnatal animal or under the skin of an immunodeficient mouse. The different behavior of pluripotent cells in embryonic milieu and in a postnatal animal suggests that tumorigenicity, a pathological property, is actually an aberrant manifestation of pluripotent state in a postnatal animal. These studies, together with a recent study, also suggest that differentiation of cancer cells induced by embryonic inducing factors could be an efficient therapeutic strategy of cancers (Yang et al., 2021). Different cancer cell features, such as rapid cell cycle and proliferation, stemness, dysregulated metabolism, dysregulated epigenetics, immunoevasion, etc., are all integral constituents of neural stemness (Cao, 2017; Xu et al., 2021; Cao, 2022; Zhang et al., 2022). In conclusion, neural stemness and its regulatory network is the unique base on which an embryo or a tumor is built up. Figure 3 summarizes that neural induction, a process of loss of original cell identity and gain of neural stemness or restoration of neural ground state in cells, drives embryogenesis in gastrulating embryos but drives tumorigenesis in a postnatal animal. Tumorigenicity and pluripotency are both but different manifestations of the same cell property: neural stemness, in different stages of an animal life.

It is increasingly clear that malignant transformation of cells is a process of progressive loss of original cell identity and gain of the property of NSCs, a cellular property that determines both tumorigenicity and pluripotency, and unifies different malignant features, such as fast cell cycle/proliferation, motility, evasion of apoptosis and immune surveillance, dysregulated epigenetics and metabolism, therapy resistance, etc. (Cao, 2017; Xu et al., 2021; Cao, 2022; Zhang et al., 2022). Neural stemness is the prime cellular property that determines tumorigenesis. TME plays important roles during tumorigenesis and has been a main focus of cancer research. Nevertheless, how the cells in TME are derived and how the cells function in cancer have remained elusive. The TME of teratocarcinoma and the pluripotency of teratocarcinoma cells are inspiring. In addition to cells originating from the host, the pluripotent differentiation potential of cancer cells (or generally, tumorigenic cells) should be the prime factor to consider when deciphering TME and the functions of TME components in cancer.

As discussed above, great progresses have been achieved in cancer therapy, in particular the targeted and immune therapies. But many challenges persist, such as the limited number of patients who can benefit from the therapies and the insurmountable obstacle of therapy resistance. Many mechanistic studies have revealed that factors conferring cancer cells with therapy resistance are generally neural stemness or embryonic neural factors, suggesting that neural stemness is also the cell property responsible for resistance. Neural stemness is the key property of cancer cells, targeting neural stemness might be potentially an efficient strategy to suppress cancer and reduce resistance effect. That cancer cells can be induced to differentiate into non-tumorigenic cells in embryonic environment implies that inhibition of tumorigenesis might be achieved via differentiation, hence the loss of neural stemness in cancer cells. Differentiation therapy was suggested 50 years ago (Pierce and Wallace, 1971; Pierce, 1983). A variety of differentiation inducing agents, including neural growth factors, all trans retinoic acid, arsenic trioxide, butyric acid or cAMP, have been shown some degree of differentiation-inducing capability in vitro and/or in vivo experiments (de Thé, 2018). The best-known case of differentiation therapy is the treatment of acute promyelocytic leukemia with all-trans retinoic acid (Huang et al., 1988). But differentiation therapy has been not applied as widely as other therapies. If the key property of cancer cells is not understood, it will be difficult to find an appropriate approach to achieve efficient differentiation. The neural stemness property and pluripotency of cancer cells implies that they can be induced to differentiate by key differentiation factors, particularly those for embryonic tissue differentiation that can mimic embryonic differentiation environment. Indeed, preclinical studies have shown that pancreatic cancer cells can be induce to differentiate by a key pancreatic differentiation factor Ptf1a, and breast cancer cells can be induced to post-mitotic adipocytes, and leading to suppression of the cancers (Ishay-Ronen et al., 2019; Krah et al., 2019). My preliminary study also demonstrated that non-neural pro-differentiation factors inhibit tumorigenicity of cancer cells via inhibition of neural stemness (Cózar et al., 2021). Whether differentiation via targeting neural stemness can be widely applied for cancer therapy is worth further studies.