Cell state transitions: catch them if you can Free

The question of how cell state transitions are controlled is fundamental for biology, as a substantial fraction of cells at any given time in both developing and adult organisms undergo controlled transitions between states or identities. Before delving into the mechanisms and uncovering emerging approaches for studying cell state transitions, we first consider what is a cell state. It might appear trivial, but particularly with the current prevalence of single-cell transcriptomics, which allows cells to be grouped based on their transcriptional similarity, it has become important to consider whether all such groups comprise a cell state or even a cell type. Following extensive discussion, we came to the consensus that, minimally, a cell state is defined by the cellular ability to perform a specific function(s) and that a transition between states entails a detectable change in function.

However, to enable the transit from one functional state to another, Austin Smith proposed that a distinct state of competence may be required. A prime example of applying such a functional definition of a cell state transition is the specification of mouse primordial germ cells – the progenitors of the gametes. Although mouse embryonic stem cells are not competent to respond to primordial germ cell inductive cues, such as bone morphogenetic proteins (BMPs), their transition into an epiblast-like state allows them to respond to BMP and to form primordial germ cell-like cells (Hayashi et al., 2011; Mulas et al., 2017; Smith, 2017). Similarly, during spermatogenesis, germ stem cells undergo a fibroblast growth factor 5 (FGF5)-dependent process of licensing, which is needed for their subsequent differentiation into gametes (Kitadate et al., 2019). The identity of the mechanisms regulating the state of cellular competence, as well as the stability of these intermediate states, remain unresolved. Importantly, the definition of a distinct cellular state is not exclusive to differentiation, but also extends to cells that change their functional behaviors, such as those that undergo epithelial-to-mesenchymal transition (Wang et al., 2022a) or exit and/or enter quiescence (Koester et al., 2021). Investigation of the previous state of competence in these types of state transitions will be of interest.

Cell state transitions are controlled by coordinated molecular regulatory networks with complex feedback behavior (Casey et al., 2020). Thus, lineage bifurcations tend to require downstream consolidation of molecular identities, a process that restricts the landscape of what is transcriptionally possible. Under this framework, the expression of a set of key transcription factors, a core regulatory network, is required for proper cell state maintenance; a change in their expression could facilitate a state transition. A prime example is the binding of transcriptionally silent chromatin by pioneer transcription factors that remodel the otherwise repressive chromatin environment to allow significant transcriptional changes, which drive cell fate transitions (Cirillo et al., 2002). In this context, Anna Philipott described how expression levels of the pioneer transcription factor Ascl1, a key transcriptional regulator in both embryogenesis and adult stem cell homeostasis, results in profound genome rewiring and altered gene expression patterns (both repression and activation) to promote neurogenic differentiation. Importantly, however, this depends on a competent cellular context (Woods et al., 2022).

Although it is clear that transcriptional programs are primary determinants of functional cellular competence and thus must be reconfigured during cell state transitions, transcription itself occurs within a highly organized nuclear environment that is dependent on the three-dimensional nuclear and chromatin organization. In this respect, Ana Pombo delivered a thought-provoking talk on the importance of understanding the cell type-specific genomic landscape when describing cell states. Specifically, Ana discussed the unique genomic architecture of terminally differentiated post-mitotic cells with highly specialized functions in the mouse brain, where chromatin topology plays a much more central role in cell state maintenance than it typically does in mitotically active cells (Winick-Ng et al., 2021). In specific brain cell types, cell state transitions are associated with large-scale alterations in the 3-dimensional (3D) chromatin architecture and chromatin interactions over long distances, indicating that high-order chromatin structure is crucial for consolidating terminally differentiated cell states. It is thus important to keep in mind that the precise interplay between 3D genome topology, gene expression regulation and state transitions and/or function is highly cell type and/or state specific.

Cell types have classically been defined by histology. Even in the era of the prevalence of global gene expression and other genomic aspects of cell fate determination, it is apparent that the morphology of a cell can be used as a readout of its molecular and functional properties (Bakal et al., 2007; Yin et al., 2013). Ewa Paluch and Jianhua Xing proposed the use of single-cell morphometrics as a powerful new tool for cell state characterization (Andrews et al., 2021; Wang et al., 2022b). Along these lines, Dominique Bergmann showed that a decrease in cell size serves as a key trigger for differentiation in plant epidermis (Gong et al., 2022 preprint), while Edouard Hannezo discussed the idea that the material properties of a cell could control morphogenetic competence, as well as cell-state transitions (Hannezo and Heisenberg, 2022). These concepts were supported by Sara Wickstrom, whose work in the skin epidermis shows that changes in cortical tension, and cell-cell and cell-extracellular matrix (ECM) adhesions direct cellular differentiation and also allow dynamic coupling of cell states and their position within the tissue (Miroshnikova et al., 2018). This concept of biophysical guidance of cell fate specification was further reinforced by the work of the Chalut laboratory, which has recently implicated dynamic differential cellular fluidity in early embryonic lineage sorting (Yanagida et al., 2022).

Importantly, as our understanding of cellular states/fates and the mechanisms of transitions evolves, it is crucial to keep in mind the difference between those transitions that actually occur within living systems in vivo versus those that can be experimentally induced via forced expression of key transcription factors (Di Stefano et al., 2014; Takahashi and Yamanaka, 2006; Weintraub et al., 1989) or by extreme in vitro culture conditions. This is especially significant in the context of revising the fundamental knowledge of cell fate bifurcations to generate cells for therapy, as these perturbations may have unappreciated long-term consequences beyond the initial expected outcomes.

Decades of work in adult stem cells of the gut, skin and other highly self-renewing tissues where cells have to adapt to the constantly changing needs of an organism indicate that cell fate decisions are likely not genetically hardwired at the level of individual cells (Beumer and Clevers, 2016; Chacon-Martinez et al., 2018; Gonzales and Fuchs, 2017; Wabik and Jones, 2015). It is plausible that the observed heterogeneity within cells of the same type (as seen by variability in gene expression, signaling and other cellular activities) reflects the capacity of cell collectives to perform context-dependent decision making. On this backdrop of heterogeneity, how do we begin to predict how cells decide their fate? Despite being complex, biological systems are subject to physical laws. Thus, many isolated aspects of biological phenomena can be captured with mathematical equations and modeled in silico to drive hypothesis generation.

James Briscoe introduced a conceptual framework using catastrophe theory to describe a time-evolving dynamic system (Saez et al., 2022). In this framework, cell states are represented as attractor equilibrium points that attract any biochemical state around its neighborhood. These attracting regions are called basins of attraction and the probability of a cell escaping the neighborhood of their initial state is determined by the size and stability of their initial basin. In the context of mouse pluripotent stem cell differentiation, the levels of Wnt and FGF signaling determine the size and stability of the different basins of attraction. The probability of a given cell undergoing a transition from an initial state and landing in another state is biased by the concentration and duration of the morphogens applied to them. Similar concepts are also exemplified by the micropatterning studies of Aryeh Warmflash where, depending on the time of exposure to BMP, human pluripotent stem cells either remain pluripotent or differentiate into either primitive streak derivatives or extra-embryonic fates (Camacho-Aguilar et al., 2022 preprint). A key challenge that remains is understanding how cells interpret and respond to signals combinatorically to maintain precise and reproducible self-organization in space and time.

However, although morphogens and niches (including cell-cell interactions, cell-ECM interactions, systemic factors, etc.) within them play a fundamental role in guiding cell state transitions, in some instances these decisions can be hardwired at the level of individual cells. For example, cells disassociated from the pre-somitic mesoderm of zebrafish embryos faithfully recapitulate their cell fate trajectories in vitro without any extrinsic signaling factors (Rohde et al., 2021 preprint). Furthermore, a spatiotemporal model for the pre-somitic mesoderm, in which cells sequentially express different transcription factors before differentiating, reproduces the gene expression patterns observed in in vivo (Negrete et al., in preparation). This transition from the mesoderm to the somite state is fast and irreversible, implying that it is genetically hardwired at the level of individual cells. This notion diverges from the model described by Briscoe and colleagues, and compels us to ask whether we can we make a general association between cell states and specific mathematical objects, such as equilibrium points or phase portrait trajectories. This is a difficult question to address due to the inherent variability in gene expression and transcriptional noise between different systems (Tsimring, 2014). Transcriptional noise may set a minimal threshold for a signal that must be met before cell state transition can occur. Thus, noise can induce transitions by creating ‘jumps’ between co-existing dynamic attractors (as in the model for mouse pluripotent stem cell differentiation) or, alternatively, cell state transitions can occur independently of noise (as in the model for somitogenesis), where it is a mere source of variability. Sophie Jarriault proposed that intermediate cell states might correspond to a set of weak attractors (i.e. being easy to escape from), whereas cell fates would correspond to strong attractors, potentially via more potent elevation of cell fate-specific transcriptional activity over transcriptional noise.

The true power of theoretical models is the ability to generate experimentally testable hypotheses that are otherwise difficult to derive. However, limitations also exist. Cell states could arise from emergent behaviors in biological systems, which implies that mathematical models might be limited in predicting them. Although it remains challenging to faithfully predict symmetry-breaking events, such as the emergence of patterning or tissue architecture, by relying purely on empirical discovery methods, sophisticated machine-learning approaches are becoming sufficiently powerful to predict many aspects of emergent phenomena, as demonstrated by the work of Assaf Zaritsky and others (Zamir et al., 2022). Another challenge for mathematical modeling, pointed out by Ben Simons, is the difference between static and time-evolving events, which introduce complications in our intent to pinpoint precise mathematical definitions for cell-state transitions. A final limitation of modeling cell-state transitions discussed during the meeting was the insufficiency of differential equations to faithfully recapitulate the complexity behind cell differentiation processes. Despite these challenges, the use of mathematical models can be powerfully complemented by model-free predictions, such as sophisticated machine-learning techniques (Pathak et al., 2018).

Although it is essential to understand how individual cells make decisions, it is also crucial to reveal how organismal morphogenesis and cell-fate decisions are coordinated to ensure that correct cell types are specified at the right time and place. Undoubtedly, studies on isolated cells are fundamental to uncovering the mechanisms that drive cell-state transitions. However, what a cell can do on its own may not be the same as what a cell does in the context of a tissue. In the embryo, cells are exposed to a highly dynamic environment that contains a plethora of physical and chemical cues. Sensing mechanochemical cues allows tissues to adjust their behavior, and therefore confers developmental robustness against perturbations – a fundamental feature of organisms with a regulative development (Martinez Arias et al., 2013). Likewise, adult tissues display a certain degree of plasticity that allows them to respond to environmental challenges, such as injury, and thereby restore homeostasis (Shivdasani et al., 2021). A paradigmatic example is the dedifferentiation of intestinal cells upon elimination of the stem cell compartment (Murata et al., 2020), as discussed by Ramesh Shivdasani.

The physicochemical characteristics of the ECM also impact cell fate decisions, both in the embryo and the adult (Engler et al., 2006; Li et al., 1987). For example, ECM stiffness upon aging compromises stem cell function in hair follicle stem cells and oligodendrocyte progenitor cells (Koester et al., 2021; Segel et al., 2019). Cell-ECM adhesions crosstalk and modulate cell-cell adhesions, which also control different aspects of cell behavior, including cell-fate specification (Miroshnikova et al., 2018; Punovuori et al., 2019). Globally, adhesion systems modulate cell shape, and the mechanics and material properties of tissues (Petridou et al., 2021). But how do these morphological and physical changes affect cell fate? We are just starting to get a glimpse of the answer. Work by the groups of Sara Wickstrom and Yekaterina Miroshnikova have dissected the molecular mechanism that allows epidermal cells to respond to stretching by changing their epigenetic marks and global chromatin organization (Le et al., 2016; Nava et al., 2020). Similarly, recent studies have shown that cell-surface mechanics and tension regulate signaling in embryonic stem cells, thereby affecting pluripotent state transitions (Bergert et al., 2021; De Belly et al., 2021). In turn, pluripotent state transitions transcriptionally enable tissue morphogenesis (Shahbazi et al., 2017), highlighting the complex feedback loops that exist between cell fate and shape during development.

Intercellular communication also needs to be included in the picture when discussing cell state transitions at the tissue level. A form of cohort communication recently described is through competition for mitogen factors. In the mouse testis, this mechanism maintains spermatogonial stem-cell homeostasis (Kitadate et al., 2019). Pioneering work by John Gurdon in the 1980s described a community effect during Xenopus development, whereby the fate of a cell was dependent on its neighbors (Gurdon, 1988). The concept of coherent cell state transitions was discussed during the meeting by Sally Lowell and others. This effect has been observed in various model systems, such as micropatterned cultures of human embryonic stem cells (Nemashkalo et al., 2017), and in the context of ferroptosis, a form of necrotic cell death (Riegman et al., 2020). We are just beginning to elucidate the molecular mechanisms that coordinate individual cellular behaviors at the tissue level to generate a collective response. For example, during wound healing, a propagation wave of activation of the protein kinase ERK determines the direction of migration (Aoki et al., 2017). Likewise, recent work from Assaf Zaritsky's lab demonstrates how endothelial cell monolayers achieve synchronized multicellular calcium mechanosensing through the gradual enhancement of information transfer from single cells to multicellular scales (Zamir et al., 2022). Finally, a typical physical strategy for coordination is given by the coupling of units with intrinsically oscillatory dynamics in the context of establishing embryonic patterning (Morelli et al., 2012). For example, during neurogenesis, neural stem cells show oscillatory expression of the transcription factor HES1 during quiescence and self-renewal, but these oscillations are arrested upon differentiation (Imayoshi and Kageyama, 2014). Similarly, in the mouse spinal cord, single-cell dynamics of the transcription factor HES5 determine the tissue-level spatiotemporal organization of neurogenesis (Biga et al., 2021). Future studies are needed to dissect the global and environment-specific molecular mechanisms behind these crucial communication events.

The study of cell state transitions requires experimentation at the interface of cell biology, tissue architecture, epigenetics and modeling, thus requiring techniques that bridge scales of biological organization. Although single-cell RNA-sequencing techniques have proven powerful for identifying cell states based on the global transcriptome, Allon Klein discussed the limitations of single-cell transcriptomics in the study of dynamic cellular histories (Wagner and Klein, 2020). Commonly used proxies for assessing dynamics, such as pseudotime analyses, can only provide hypotheses of the real-time dynamics, which then need to be tested by methods such as barcode tracking or time-lapse imaging with high temporal and spatial resolution. We discussed the opportunity to map the tissue context using spatial transcriptomics (Marx, 2021), and to go beyond RNA expression by capitalizing on recent advances in high-throughput imaging and image analyses, which enable us to perform single-cell morphometrics at unprecedented resolution (Andrews et al., 2021). We also need to capture tissue-level effects by deconstructing the complexity and dynamics of the native ECM and dissecting its contribution to the chemical and physical cues that a cell is experiencing while undergoing a fate transition. The use of in vivo models to uncover mechanisms that can then be mechanistically dissected within fully chemically defined and tunable hydrogels has proven to be particularly useful (Koester et al., 2021; Labouesse et al., 2021; Ranga et al., 2016). Indeed, incorporating the third dimension into standard 2D cultures has had a dramatic effect on the field, as several aspects of cellular physiology are altered when moving into three dimensions, from genome architecture to protein concentration, cell phenotypes and behaviors, as beautifully demonstrated by many labs in the organoid and embryoid fields (Shahbazi et al., 2019; Simian and Bissell, 2017). Finally, computational modeling and machine-learning approaches are increasingly used as tools for hypothesis generation and testing, complementing experimental work (Sharpe, 2017). We thus anticipate that the resolution of puzzles regarding cellular decision making, at the single-cell and population levels, will emerge from collaborative, interdisciplinary and international efforts.