Over the past decades our understanding of how tumors form, grow and progress has shifted sharply from the early idea that a phenotypically static cancerous cell state is the driving force of oncogenesis, to appreciating that tumors are heterogeneous, dynamic entities that evolve over time and change in response to external cues, including therapy. Technological developments in tumor profiling and modeling, such as the increased availability of human samples and the use of sophisticated animal and patient-derived tumor models, have been revealing not only an unanticipated level of inter- and intratumor heterogeneity, but also the exquisite plasticity of cancer cells. This multifactorial attribute has emerged as a hallmark of cancer, as well as a moving target for therapy: depending on genomic status, tissue context and exogenous conditions, cellular plasticity underpins dedicated cell subpopulations that drive not only tumor development and progression, but also resistance to, and relapse after, therapy1,2. Three papers published in this issue of Nature Cancer approach this topic from different perspectives by highlighting the complexities of cancer cell plasticity in distinct cancer types.

Working on diffuse intrinsic pontine glioma (DIPG), a rare, hard-to-treat pediatric cancer type with a very poor prognosis, Xi and colleagues delineate a molecular mechanism through which tumor cells can be reprogrammed to a differentiated state to inhibit tumor growth3. They show that signaling via bone morphogenic protein (BMP) is downregulated in one of the most frequent subtypes of DIPG that has been reported to reside in a prolonged stem-cell-like state and harbors a lysine-to-methionine substitution in histone H3.3 (H3.3K27M) but is wild-type for the gene ACVR1, which encodes a BMP type I receptor. They attribute the limited BMP pathway activity to the action of CHRDL1, a protein known to antagonize binding of BMP ligands to their receptors and that has high expression in this particular DIPG subtype. The authors go on to demonstrate that tumor growth is reduced when BMP signaling is activated either through treatment with BMP ligands or with the HDAC inhibitor Dacinostat, which has been shown to upregulate this pathway, and identify the induction of CXXC5 expression as a key node in the tumor-suppressive activity of BMP signaling in this setting. Thus, by elucidating the molecular underpinnings of the prolonged stem-like state of the mutant H3.3K27M and wild-type ACVR1 DIPG and by demonstrating the plasticity of these tumor cells toward a BMP-induced differentiated state, the investigators identify a potential approach to targeting this currently incurable DIPG subtype.

In a separate study, Mu and colleagues delineate the molecular mechanism that underlies the plasticity-mediated resistance of metastatic castration-resistant prostate cancer cells to androgen receptor–targeted therapies, such as enzalutamide4. Starting from the prior observation that resistance of prostate cancer cells with loss-of-function TP53 and RB1 mutations and upregulation of SOX2 can emerge through the acquisition of a lineage-plastic state that permits survival independently of androgen receptor signaling, the authors identify the ectopic activation of JAK–STAT signaling as a key mediator of this process, with SOX2 orchestrating a positive feedback loop. They demonstrate that the JAK–STAT pathway drives resistance specifically in a stem-like subset of prostate cancer cells with gene-expression programs indicative of multilineage potential, but not in cells displaying neuroendocrine lineage transcriptional programs. Importantly, they show that pharmacological inhibition of JAK–STAT signaling reduces tumor growth in mice, and also reverses the lineage-plastic state that underlies therapy resistance and thus resensitizes resistant prostate cancer cells to enzalutamide. An accompanying News & Views article by Brady and Barbieri further contextualizes these findings and the potential of harnessing cell plasticity for therapeutic purposes5.

Moving to the study of tumor relapse after therapy, Batlle and colleagues identify a drug-tolerant persister colorectal cancer cell subpopulation that is marked by expression of Mex3A and is able to regenerate the disease after chemotherapy6. The authors first identify the Mex3A+ cells as a latent subpopulation that is, however, chemoresistant and can fuel tumor regeneration, by using mouse and patient-derived colorectal cancer organoids to study the effect of different growth conditions, aiming to emulate distinct tumor niche environments. They go on to study the role of this cell subset in mouse models of metastatic colorectal cancer and observe that although Mex3a+ cells make a minor contribution to metastatic outgrowth, they are able to fuel tumor recurrence after chemotherapy by downregulating WNT signaling–mediated stem cell gene expression and transiently switching to a transcriptional program similar to that of YAP+ fetal intestinal progenitor cells. In the absence of Mex3a, chemotherapy-treated colorectal cancer cells fail to survive. Although further work is needed to fully characterize this cell population and its role in driving disease relapse, these findings highlight the transient adoption of plastic states as a key step in driving tumor regeneration after therapy. The authors further discuss the implications of their work in an accompanying Research Briefing7.

Seen together, these findings underscore the dynamic, plastic nature of tumors and the complexity of characterizing plastic states determined by distinct genetic, tissue and environmental contexts in different settings: plasticity induced by unleashing the action of a signaling cue to inhibit tumor growth; plasticity induced by the ectopic action of a signaling pathway to drive therapy resistance; and plasticity induced to drive disease relapse after therapy. Importantly, they also highlight the potential of utilizing this knowledge for therapeutic purposes, providing exciting routes for further preclinical and translational work.