Intertwined with the concept of cell lineage is that of cell commitment. Cell lineage follows the normal fate of a cell and its daughters, leading to the formulation of genealogical trees of cells with increasingly restricted cell fate choices as development proceeds. For many years, comprehensive lineage reconstructions had been possible only in lower invertebrates, such as the nematode Caenorhabditis elegans (Sulston et al., 1983), or in basic chordates as Ciona intestinalis (Tassy et al., 2006). However, recent technological developments have proved to be valuable to address the lineages of organisms with non-stereotypic development. By reconstructing different cell lineages, we can now determine the functional cell transitions that distinct cell populations undergo, the impact of morphogenesis in the spatial distribution of progenitor cells, and the dynamics of the whole cell population. In other words, they provide the cellular data to complement the well-described GRNs involved in cell specification and differentiation. Multiple efforts have been deployed to developing tools for cell lineage analysis. These tools can be classified into: (i) cell birth-dating, aimed to identify when cells are born; (ii) cell fate mapping, to reveal the developmental potential of progenitor cells at later developmental stages; (iii) clonal analysis, to decipher the derivatives from a progenitor cell; and (iv) cell lineage tracing, to describe the mitotic connections between two or more genealogically related cells, allowing the assessment of cell lineages and cell behaviors in the whole organ context. They can be applied either to single cells in a mosaic manner or to an entire cell population (Garcia-Marques et al., 2021).

Imaging-based strategies provide an excellent spatial resolution, allowing to determine genetic clonal relationships based on the mitotic history, such as twin-spot Mosaic Analysis with a Repressible Cell Marker (twin-spot MARCM) in the nervous system of Drosophila (Yu et al., 2009), or Mosaic Analysis with Double Markers (MADM) in mice (Zong et al., 2005; Gao et al., 2014). The first enables the visualization of sister-paired clones from the same progenitor in two different colors, while the latest, permits to identify different recombination events by single or combined segregation of two fluorescent proteins (GFP and RFP) in daughter cells, therefore enabling the tracing of such derivatives in a total of three colors (green, red, and yellow). To improve their limited clonal resolution, multicolor labeling tools such as Confetti (Snippert et al., 2010), Brainbow (Livet et al., 2007; Weissman et al., 2011; Cai et al., 2013), StarTrack (García-Marqués and López-Mascaraque, 2013; Figueres-Oñate et al., 2016) or MAGIC (Loulier et al., 2014), rely on a stochastic and combinatorial expression of different fluorescent reporter genes induced by recombinases, which results in the generation of multiple color hues that label clonally related cells in the same color palette (for detailed reviews on cell lineage tools, both imaging and sequencing-based, see Espinosa-Medina et al., 2019; Figueres-Oñate et al., 2020; Garcia-Marques et al., 2021). Although the development of such multicolor strategies has been a major step forward in the cell lineage tracing field, they are not scalable, and in many cases they do not provide temporal resolution.

Noteworthy, the development of high-resolution 4D imaging paired with Light Sheet Fluorescence Microscopy (LSFM) using zebrafish transgenic embryos set up the path for understanding early embryonic development and assess cell lineages and behaviors at high spatiotemporal coverage and resolution (Keller et al., 2008; Olivier et al., 2010; Luengo-Oroz et al., 2011; Keller and Ahrens, 2015; Dyballa et al., 2017). This was accompanied with the development of cell-tracking tools, instrumental to reconstruct cell lineages and cell rearrangements upon time (Amat et al., 2014; Faure et al., 2016; Wolff et al., 2018; Wan et al., 2019). Although this approach provides valuable temporal information about cell lineages —and therefore cell hierarchies— in the context of the whole cell population, they are not scalable and need high computing power and specific know-how for the tracking analyses.

In addition to these imaging-based cell lineage tools, the development of CRISPR/Cas and the high-throughput and highspeed sequencing revolution have pushed forward the emergence of sequencing-based lineage strategies, which enable the establishment of cell connections upon unique genomic landmarks, also known as barcodes. Whether it is by Cas9/sgRNA induced genomic mutations (Cotterell et al., 2020; CARLIN, Bowling et al., 2020; ScarTrace, Alemany et al., 2018; LINNAEUS, Spanjaard et al., 2018) or by the insertion of exogenous arrays of DNA with multiple and inducible CRISPR/Cas target sites such as scGestalt (McKenna et al., 2016; Raj et al., 2018), barcoding tools label individual cells with a unique combination of scarred sequences. This cumulative stochastic barcode editing provides a unique DNA scar combinatory that will prevail in derivative cells, while adding up new generated ones. This enables the “tracing” of such derivatives after a transcriptomic analysis, using pseudo-time scales to generate cell trajectories and infer relationships between progenitor cells and their progeny. Similarly, other methods such as MEMOIR by engineered Mutagenesis with Optical In situ Readout, asses trajectories by combining barcoding elements and sequential rounds of multiplexed in situ hybridization (Frieda et al., 2017). Its improved version intMEMOIR goes further allowing the differentiation between cellular states due to the higher number of integrated barcode combinations as the result of the array’s inversion after genomic recombination (Chow et al., 2021). Although these strategies provide valuable single-cell transcriptional signature maps and atlases, the barcoding and omics combination still fails to represent cell behavior at the tissue level, neither provides cell division rates nor kinetics. Moreover, no functional relationships (circuits) are obtained.

To overcome such limitations, there is an urgent need for 4D tools that allow cell lineage relationships and temporality within the morphological context to be scalable. A few strategies have recently emerged that comply with the requirements of such need. One of these is CLADES (Cell Lineage Access Driven by an Edition Sequence), a genetic tool in Drosophila that enables cell lineage tracing coupled with birth-dating information. It is based on the sequential activation of a cascade of different reporters in progenitor cells by CRISPR/Cas9 induction, which are inherited by their differentiated progeny. This enables the cell lineage tracing without losing the temporal input, since early born and late-born cells will be labeled in different colors of the reporter cascade (Garcia-Marques et al., 2020). However, while clonal resolution is not an issue in Drosophila with a few and highly stereotypic lineages, cell lineages in vertebrates remain incompletely characterized due to the higher tissue complexity and the larger size of embryos. Therefore, there is still the demand to incorporate new strategies that couple temporal and clonal information preserving the anatomical context to fill the remaining gap between cell biology and genetic determinism. To strengthen the importance of such factors in a tissue context, in the next section we discuss how time and space shape differently two paradigmatic structures of the CNS, the brain cortex and the spinal cord.

The characterization of the adult spinal cord according to morphology, molecular markers, neuronal connectivity and axonal projections revealed a modular organization with stereotypical position of specific neurons (Sagner and Briscoe, 2019). During spinal cord formation, long-range morphogen signals emanating from the roof and floor plates pattern the tissue along the DV axis by regulating cell fate through transcription factor expression. This transcription factor code defines 11 molecularly distinct neural progenitor domains –six dorsal and five ventral–, each of which gives rise to one or more different neuronal subtypes (Alaynick et al., 2011; Lu et al., 2015). As development proceeds, progenitors in each domain specify in a spatiotemporally ordered manner and either amplify, or give rise to the corresponding type of post-mitotic neurons (Jessell, 2000; Dessaud et al., 2008; Le Dréau and Martí, 2013). Thus, the stereotypical position of cells –and the spatial regulation of gene expression– is crucial in the formation of neuronal circuits. However, several large-scale molecular profiling studies provided catalogs of gene expression, revealing a higher complexity of cell types (Delile et al., 2019; Rayon and Briscoe, 2021). For instance, single-cell RNA-sequencing experiments in mice embryonic and adult spinal cord suggest the existence of at least several dozen of molecularly different neuronal subtypes (Delile et al., 2019). This reveals the sequential upregulation or induction of sets of transcriptions factors that underpin the identity of the derivative arising neurons, generating a temporal stratification of neuronal subtypes from each domain. Thus, complementary to the positioning, time also plays a role in neuronal identity acquisition and in the generation of neuronal diversity (Sagner and Briscoe, 2019).

In the spinal cord, distance from the DV poles seems to dictate progenitor competence since neural progenitors acquire distinct identities in response to opposing morphogen gradients (Jessell, 2000; Ribes and Briscoe, 2009; Le Dréau and Martí, 2013). These morphogens, Sonic hedgehog (Shh) ventrally and Bone Morphogenetic Proteins (BMP) and Wnts dorsally, induce the expression of several homeodomain and bHLH transcription factors in discrete domains along the DV axis (Kutejova et al., 2016). Since their combinatorial expression confers neuronal identity to progenitors, they are exquisitely regulated. They cross-regulate forming a well-defined GRN accountable for the response to morphogen gradients by modular enhancers (Peterson et al., 2012; Oosterveen et al., 2013). During neurogenesis, the spinal cord continues to grow along its DV axis and expand the mantle zone with the differentiated neurons, raising the questions of how discrete progenitor domains and specific gene expression territories remain stable and scalable upon being challenged by cell proliferation and how DV patterning and neurogenesis are intertwined. Lately, a two-phase model has been proposed for explaining the growth and patterning of the spinal cord (Kicheva and Briscoe, 2015). The pattern of neuronal progenitor domains would be established at early developmental stages, when the position of a cell within the morphogen gradients grid can be precisely decoded (Sagner and Briscoe, 2019). These progenitors would maintain certain cell plasticity such as they could switch identities (Dessaud et al., 2007, 2010), facilitating the transition along different progenitor states. Upon tissue growth, the pattern of progenitor domains would be maintained by GRN cross-repressive interactions and unequal neuronal differentiation rates would determine domain sizes (Kicheva et al., 2014). However, how the neuronal differentiation dynamics of different progenitor populations is regulated has not been revealed. Interestingly, the regulatory programs and neuronal cell types are highly similar in different vertebrates, despite the distinct developmental time scales across species. Differences in protein turnover play a role in interspecies differences in the tempo of motoneuron differentiation (Rayon et al., 2020); however, whether similar mechanisms may operate to specifically regulate the differentiation rate of the distinct progenitor domains is not known yet.

Dynamics of morphogen signaling and cell division mode have been linked in the spinal cord, since the onset of neurogenesis in dorsal interneurons and ventral motoneurons is controlled by BMP/SMAD- and Shh-signaling, respectively (Saade et al., 2013; Le Dréau et al., 2014). As example, in motoneuron progenitors (MNp) Shh maintains self-expanding symmetric proliferative divisions, while preventing progenitors from switching to neurogenic divisions. A reduction in Shh activity results in reduction of symmetric proliferative cell divisions, coinciding with the developmental time of motoneuron generation (Saade et al., 2013, 2017). While clones of the MNp domain grow equally in both axes, the rest of the domains show more elongated cell clones in DV, resulting in an inferior net growth rate DV/AP (Kicheva et al., 2014). A 3D computational simulation of the spinal cord DV growth shows that the differences in the spread and shape of MNp clones, and the isotropic growth, can be explained by the higher differentiation rate of these progenitors (Kicheva et al., 2014; Guerrero et al., 2019). Overall, these studies demonstrate that DV progenitor position influences proliferative capacity, cell fate and growth in the spinal cord. However, they do not explain why MNp differentiate at a higher rate than progenitors in adjacent domains. As neurons differentiate, they delaminate toward the mantle zone. This active displacement of neurons shapes the spinal cord in such a manner that progenitor cells are kept in the ventricular zone and differentiated neurons allocate in the adjacent medial domain, and the tissue grows without dramatic morphogenetic changes.

In several systems temporal cues regulate neuroblast competence –and therefore expansion of neural diversity– by specifying distinct neuronal fates using combinatorial temporal patterning (Bayraktar and Doe, 2013). In the spinal cord, the birth order of neurons also underlies specificity in neuronal connectivity and circuit formation (McArthur and Fetcho, 2017; Pujala and Koyama, 2019; Wan et al., 2019). As previously mentioned, several works have stressed the importance of the temporal transcription factor code for subdividing neurons through the DV axis (Delile et al., 2019). Although these results suggest that the temporal transcriptional factor code is functionally important, it seems that temporal cues would work within a given neuronal population to help to expand its diversity. Thus in the spinal cord, the precise position of neural progenitors serves as a functional ground for neuronal subtype determination.

Similar to other regions of the CNS, neural progenitors in the spinal cord give rise first to neurons and later to glial cells. This temporal switch relies on the sequential induction of SoxE and NFI factors and is regulated by several signaling pathways (Deneen et al., 2006; Kang et al., 2012). As an example, studies in zebrafish embryos demonstrate that motoneurons and oligodendrocytes emerge from the same ventral progenitor domain, the MNp (Zannino and Appel, 2009; Esain et al., 2010). In mice, most of the oligodendrocytes are generated after motoneurons in a Shh-dependent manner (Soula et al., 2001; Fogarty et al., 2005). However, 5% of the total population arises in a Shh-independent manner from a dorsal Dbx1-expressing region at early postnatal stages and distribute to the lateral white matter, radially opposite to their site of origin. In contrast, pMN-derived oligodendrocyte cells usually distribute in the gray matter (Fogarty et al., 2005). Similarly, DV position also determines the astrocytic subtype since the expression of Pax6 and Nkx6.1 confers positional identity defining three distinct astrocyte subpopulations. Each of these progenitor domains displays a specific code for Reelin and Slit guidance molecules, resulting in a correlation between the origin of astrocyte subtype and their final position within the neural tube (Hochstim et al., 2008). Thus, in oligodendrocytes and astrocytes, both birth-dating and DV position within the spinal cord influence their final location within the adult structure.

Since Cajal’s descriptions of the brain cortex cytoarchitecture and laminar distribution, the development of clonal analysis and cell lineage tools has fastened and accurately unveiled its organization and composition. The cerebral cortex evolves from a dense and packed single cell sheet composed solely by progenitors –the embryonic forebrain– to a stratified tissue remarkably conserved across most mammals. The neocortex is organized into six distinct layers, each of them with neuronal heterogeneity that emerges from sequentially born progenitors. The ventricular zone (VZ) harbors the soma of progenitor cells, followed by the subventricular zone (SVZ) as the main area of cell amplification. The cortical plate consists of several cell layers that sequentially accumulate on top (LI-LVI, being LI the uppermost and LVI the deepest layer), in an ‘inside-out’ manner, where early born neurons locate in deep layers, whereas newly born neurons migrate and position in upper layers (Angevine and Sidman, 1961; Rakic, 1974; Takahashi et al., 1999). In contrast, their glial counterparts organize in a stochastic manner along the apicobasal extent of the cortical plate (Zhang X. et al., 2020).

Neuroepithelial cells are the early progenitors populating the VZ, which divide in a symmetric proliferative manner prior to neurogenesis (Subramanian et al., 2017). As neurogenesis starts, NEC transition to RGC that are classically defined by the combination of several features: (i) an elongated morphology with contacts in the apical and basal surfaces of the neuroepithelium; (ii) the maintenance of the apicobasal polarity, (iii) the expression of astroglial markers such as glutamate transporter (GLAST) (Hartfuss et al., 2001), glial fibrillary acidic protein (GFAP), glutamine synthase (GS), and brain-lipid-binding protein (BLBP) (Feng et al., 1994; Arellano et al., 2021). Nascent RGC may undergo symmetric proliferative cell divisions to expand the progenitor pool, and later they transition into the neurogenic state and asymmetrically divide, thereby producing a self-renewed RGC and a differentiated cortical neuron (Noctor et al., 2004). Ventricular RGC might as well give rise to one RGC and either an IP or to another progenitor type, the basal radial glial cell (bRGC) (Miyata et al., 2001; Haubensak et al., 2004; Noctor et al., 2004) that migrates to the SVZ becoming the major contributor to neuronal diversification (for reviews see Huttner and Kosodo, 2005; Penisson et al., 2019). These bRGC differ from apical RGC in their retraction of the ventricular processes before their division (Miyata et al., 2001), and in their division mode, since they usually generate either two bRGC by symmetric proliferative division or two daughter neurons by symmetric differentiative division (Haubensak et al., 2004; Noctor et al., 2004).

By consecutive waves of neurogenesis, distinct cortical layers are formed in the ‘inside-out’ fashion (for reviews see Rakic, 2009; Taverna et al., 2014; Hansen et al., 2017). Despite its pluripotency, cortical RGC undergo a progressive fate restriction over time (Desai and Mcconnell, 2000), since they lose the capacity to generate deep cortical layer neurons, limiting their derivatives to upper layers (McConnell and Kaznowski, 1991; Frantz and McConnell, 1996). Most clonal analysis studies suggest that the RGC behavior can be predictable across all developmental stages. RGC in the neurogenic phase do not undergo terminal differentiation in a stochastic manner but rather follow a defined non-random program of cell cycle exit resulting in eight to nine neurons produced by one RGC (Gao et al., 2014). In the same line, MADM clones induced in RGC at later developmental stages, right before the onset of gliogenesis, show more neurons in the upper layers (Zhang X. et al., 2020). Similarly, other studies demonstrate that as more fate restricted the cortex progenitor cells are, less neuronal cell types they are able to generate. For instance, when IP are early targeted, the derived neurons locate mainly in deeper layers instead of covering the entire translaminar area (Mihalas and Hevner, 2018); when they are targeted even later in development, they mainly produce neurons that locate in upper layers instead (Tarabykin et al., 2001). These results indicate that RGC constitute a pretty homogeneous cell population. However, a recent report shows that a limited number of progenitors display a stochastic neuronal output to account for the diverse clone types (Llorca et al., 2019). When they map the lineage of genetically labeled progenitor cells focusing on progenitors that start generating neurons early during development, they observe that early born neurons locate in deep layers as expected, and that a substantial group of neurons are confined either to the deep or superficial layers. They propose that heterogeneous lineage configurations can arise directly from neurogenesis and contribute to diverse neuronal types (Llorca et al., 2019). Overall, these observations suggest that the laminar position allows a crude classification of projection neurons and dictates their connectivity, although the progenitor population might not be so homogeneous.

Radial glial cells can also produce glial cells, both astrocytes and oligodendrocytes, which organize dispersedly along the apicobasal extent of the cortical plate. Once the neurogenic capacity of the remaining progenitors decreases, intrinsic and extrinsic signals set the start of gliogenesis (for review see Kessaris et al., 2008). However, the mechanisms of lineage progression from neurogenesis to gliogenesis remain largely unexplored. Astrocytes arise few days later than neurons as the result of remaining RGC detaching from apicobasal poles and retracting their projections (Noctor et al., 2004). Although astrocytes show some layer-specific features (Lanjakornsiripan et al., 2018; Batiuk et al., 2020), a clonal analysis study using MAGIC Markers combinatorial labeling, demonstrates that astrocytes do not follow an ‘inside-out’ pattern; instead, they distribute along the cortical area and acquire their fate in a stochastic manner (Clavreul et al., 2019; Zhang X. et al., 2020). It is well known that oligodendrocytes arise from both NG2-positive (Mo and Zecevic, 2009) and NG2-negative (Gensert and Goldman, 2001) oligodendrocyte progenitor cells; however their birth-date is still unclear. There are several reasons for this such as (i) oligodendrocytes found in the cortex are a result of competing waves emanating both locally and from other brain areas (Spassky et al., 1998; He et al., 2001; Tekki-Kessaris et al., 2001; Gorski et al., 2002; Kessaris et al., 2006), and (ii) fully differentiated and functional myelinating oligodendrocytes maturate during postnatal stages (for reviews on intrinsic and extrinsic factors driving oligodendrocyte development and maturation see Meijer et al., 2012; Baydyuk et al., 2020). Neuronal layer inversion studies suggest oligodendrocytes indeed need the correct sequential positioning of neurons to acquire their characteristic asymmetric distribution along the cortical area (Tan et al., 2009). Thus, temporal cues regulate the successive generation of layered postmitotic neurons and glial cells. Although we have a framework for RGC lineage progression, there are still open questions such as (i) how heterogenous the pool of RGC is, (ii) whether deterministic and stochastic modes of neuronal production coexist, and (iii) how cortex morphogenesis and cell fate acquisition are coordinated.

Recent advances have challenged that time and spatial location are the main determining factors for cell specification and cell fate acquisition for the generation of cell diversity in the CNS. High-throughput transcriptional profiling studies have allowed the envisioning of new horizons for cell characterization based on their individual RNA profile and challenged the “cell fate” concept proposing cellular state as the accurate terminology (Figure 1A). This brings on the table an open debate that goes beyond nomenclature: cell fate or cell state?

As previously stated, new technological approaches mainly based in single-cell molecular profiling elicit new arguments about what a cell type is. Cell types are the basic building blocks of multicellular organisms, determined and maintained by gene regulatory programs; however, cell type classification schemes remain ambiguous (Arendt et al., 2016). A classic discrete cell type categorization from progenitor cells to their differentiated derivates has been challenged lately as such description considers that cells follow discrete steps in a linear path to acquire their fate. Instead, it has been suggested cells navigate through different states toward differentiation (Figure 1A). Conceptually, cell fate comprises the future identity of a cell and it is determined by multiple factors such as gene expression, cell–cell interactions and external cues –both mechanical and biochemical. Therefore, cell fate would be the discrete and final step that defines the type of a cell. In contrast, the state of a cell implies a temporary feature, as it undergoes transitions over time from a starting point in space to the next one in a continuum of a dynamic system. Such scenario is usually envisioned as a space of states (Trapnell, 2015) and has recently become one of the most fervent debates in biology (see Clevers et al., 2017 for disparate views on cell identity, cell fate and cell state concepts).

Although a cell type is characterized by morphology, function, position and gene expression, even homogeneous cell type populations display high heterogeneity in their transcriptional profile. The current challenges biology faces are the untangling and the interpretation of this heterogeneity at the individual cell level, and for this, many genetic and transcriptomic profiles are carried out to characterize individual cell signatures. Posterior cell clustering and mapping of such libraries identify intermediate cell states, as single cells with a similar transcriptomic profile are likely to be closely related. However, the question poses: can we predict from the transcriptome of a progenitor cell, the identity, connectivity and function of their derivatives?

Several studies have reported the existence of transitory cell states within the CNS using these approaches. For instance, single-cell RNA-sequencing analysis of FACS sorted cortical apical progenitors identifies different transcriptional states according to the pseudo-developmental stage (Telley et al., 2019). While at early mice embryonic stages apical progenitors are characterized by the expression of genes related with cell intrinsic transcriptional programs, at later stages they progress to a more environmental-sensing transcriptomic signature, which already suggests that environmental cues may play a key role in refining the neuronal heterogeneity arising from cortical apical progenitors (Telley et al., 2019). In the zebrafish hindbrain, a single-cell RNA-sequencing study at different embryonic patterning stages identified discrete cellular states that differ on their transcription factor expression, and recapitulate the transition of progenitors to neuronal differentiation (Tambalo et al., 2020). This can suggest that cells display different competence states, and/or the existence of a heterogeneous progenitor pool. Barcoding systems have contributed exponentially to the untangling of new cellular states in the CNS by generating cell trajectories (Raj et al., 2018), although in none of the cases cell hierarchies could be established. As we discussed in Section “Framing the question: progenitor cells vs. differentiated neurons” of this review, the impact of single-cell transcriptomics on the characterization of gene programs for neuronal diversification is undeniable; however, these tools are based solely on transcriptomic signatures with neither spatial organization nor cell–cell contact inputs since they require tissue dissociation. Moreover, since they rely on pseudo-time parameters they lack the developmental history of cells, raising the question of whether cell identity can be defined by a single signature pattern of gene expression. Although the integration of imaging approaches that maintain the 3D tissue conformation might solve the first issue, such as MERFISH (Chen et al., 2015; Zhang M. et al., 2020) or STARmap (Wang et al., 2018), these trajectory-based assays alone are not sufficient to capture the intricacy of such dynamic systems.

These high-resolution data approaches provide an unprecedented level of detail and are indeed revolutionizing the study of CNS development. They demonstrate (again) that complexity is build up during embryonic development, and suggest that once “crude cell fates” are established, the final cell identities are refined upon time with cells transitioning through different cell states. In other epithelial systems, stem cells constitute a heterogeneous compartment in which cells transit reversibly between different states of competence (Blanpain and Simons, 2013). The big leap forward would be to combine cell lineage, developmental cell trajectories, and molecular mechanisms, to comprehend how neuronal diversity arises. Most importantly perhaps, if dynamic cellular features are predictable at a population rather than single-cell level, understanding the emergent properties of cell populations instead of by the detailed account of their individual components should be considered to address the emergence of functional circuits during embryogenesis.

A long-standing question in developmental neurobiology is to determine how cell diversity arises from groups of “equivalent” progenitor cells. Understanding how cell fate choices are made is crucial to comprehend the spatiotemporal dynamics of tissue and organ formation and to predict cell behaviors. Waddington’s analogy of cells represented as bowls rolling down valleys in a downhill landscape is one of the most well-known examples of cell fate decision-making in a dynamic system (Waddington, 1957). The path to two different states or fates is illustrated as branching valleys and it represents fate choices in a cell’s endeavor, not mitotic events. This model, referred as well as the epigenetic landscape, proposes that a cell’s potential for development, meaning the down-path it takes, is marked by groups of genes or biochemical interactions, already speculating about the GRN control of cell fate decision (for reviews see Enver et al., 2009; Fagan, 2012). Waddington’s landscape model is general, qualitative, and although portraits the concept of cell competence, it does not provide the 3D-positional input within the tissue, or it considers intermediate cell states between one decision and the next. It implies bowls (cells) that go down chosen valleys (fate decision) cannot go back to the previous position once they are committed. It does not resolve if the decision of choosing a developmental path (to go through the different valleys) is stochastic or already determined. Since Waddington’s theory, several genetic fate mapping studies in combination with mathematical models have tried to tackle this issue in specific structures (for review see Zechner et al., 2020). While there is no debate around the vast heterogeneity harbored in the CNS, there is not a single view in the generation and acquisition of such cell type variety. While some lines of research suggest it must correlate with progenitor subtype diversity –implicating fate is determined before differentiation–, other studies suggest that the brain may harbor multipotent progenitors whose decision of generating different fates is merely stochastic.

One of the best studied cases is the brain cortex (see section “The two paradigm models for tissue growth in the central nervous system: the spinal cord and the brain cortex”). In the recent years the technological advances in single-cell transcriptomes led to revisit the knowledge on cortical projection neuron heterogeneity and its ontogeny (Briscoe and Marín, 2020). A genetic fate mapping study on a subset of RGC demonstrates that the murine cerebral cortex contains RGC sub-lineages with distinct fate potentials. Using in vivo genetic fate mapping and in vitro clonal analysis, they identify a Cux2-positive RGC lineage intrinsically specified to generate only upper-layer neurons, independently of niche and birthdate. Interestingly, when forced to exit earlier the cell cycle, the outcome is also specific for upper layer neurons, indicating these RGC progenitors already specified to generate upper layer neurons regardless of birthdate were intrinsically programmed to generate neurons predominantly later than their lower layer counterparts (Franco et al., 2012). Thus, this study indicates that molecular fate specification ensures proper birth order, rather than vice versa. In this same line, a MADM clonal analysis on RGC progenitors revealed that all progenitors give rise to eight to nine neurons in a reproducible way, and that after this stereotyped neuronal generation, glial cells are produced. Such behavior was interpreted as a deterministic neural fate acquisition pattern (Gao et al., 2014). However, Guo et al. (2013) demonstrated the existence of multipotent neocortical progenitors. Genetic fate mapping of Cux2-positive RGC shows that they sequentially generate both deep- and upper-layer projection neuron subtypes and glia. More recently, clonal analysis studies combined with mathematical models favored the existence of a stochastic acquisition of neuronal fates, challenging again the deterministic view. They developed mathematical models that could emulate their biological observations in mice cortex using MADM-induced clonal analysis, and the best fitting models suggested that indeed neurogenic fate decisions could be stochastic (Llorca et al., 2019). They proposed that the heterogeneity of neuronal fates in the cortex might be explained by the existence of two distinct progenitor cell populations, which would randomly generate the translaminar cell diversity across the cortical plate. These observations raise further questions such as (i) can the two progenitor subtypes be molecularly identified, (ii) do the stochastic events occur within progenitors or in the progeny, and (iii) what is the relative final contribution of stochastic and deterministic processes (Klingler and Jabaudon, 2020).

The stochasticity in neurogenic fate decisions has also been shown in other CNS structures such as the adult zebrafish telencephalon (Than-Trong et al., 2020) and both zebrafish and Drosophila’s retina (Bell et al., 2007; He et al., 2012). Intriguingly, both cell fate decision mechanisms operate in the retina. Retinal progenitor cells have the same competence; however, extrinsic and intrinsic cues induce cell fate in a reproducible deterministic manner during embryonic development (for review see Cepko, 2014). Stochastic cell fate decisions are most abundant at early stages of retinal neurogenesis (He et al., 2012). By contrast, more deterministic division patterns become prominent late in embryonic development, probably because they often arise from a committed precursor, which gives rise to later born neurons. For instance, in zebrafish the assignment of Müller glial fate, a retina-specific RGC, has been shown to follow a deterministic pattern instead (Rulands et al., 2018). This is a clear illustration of how two distinct neural fates can be acquired following a different pattern within the same structure. Overall, these data suggest that generation of cell diversity cannot be explained by one model solely but instead might be better represented by a synergistic cooperation of stochastic and deterministic features.