Tracing the Dynamics of Stem Cell Fate

LESSONS FROM GERMLINE STEM CELL MAINTENANCE: FROM INVARIANT TO POPULATION ASYMMETRIC RENEWAL

Historically, studies of germline maintenance in the Drosophila melanogaster have provided fundamental insights into the regulation of stem cell self-renewal (Fuller and Spradling 2007; Lehmann 2012). In Drosophila testes, sperm production relies on the activity of germline stem cells (GSCs) that lie anchored by adherens junctions to specialized somatic hub cells (Fig. 1A). By providing a polarized source of “fate determinants” (including the ligands of the Unpaired family [Upd] and Decapentaplegic [Dpp]), the hub functions as a closed niche, maintaining GSC competence. During division, the anchoring of the mother centrosome to the adherens junction orients the spindle, positioning daughter cells away from the hub, and facilitating their entry into a differentiation pathway that begins in the gonialblast state (Fig. 1B). Through the combination of force-mediated spindle orientation and the localization of fate determinants, the niche provides a basis to regulate asymmetric GSC fate (Venkei and Yamashita 2018). But how generic and robust is such a mechanism of invariant asymmetric stem cell self-renewal, in which each and every stem cell division results in an asymmetric fate outcome?

In mouse, sperm production relies on the proliferation of spermatogonial stem cells (SSCs) that “roam” freely and actively among their differentiating progeny on the basement membrane of the seminiferous tubules in the testis (see Yoshida 2019). In homeostasis, SSCs self-renew, giving rise to differentiating progeny that transit through some 10 to 11 rounds of incomplete division before entering into meiosis, when they translocate across tight junctions separating the basal layer from the adluminal compartment (Fig. 1C). Further progression is then directed through distinct stages of maturation by periodic cycles of retinoic acid (RA) signaling that propagate as “phase” waves—like a “Mexican wave”—along the length of the tubules (de Rooij and Russell 2000). But where does stem cell function reside? Is self-renewal potential limited, as in Drosophila, to a defined subpopulation of singly isolated (A_single) spermatogonia that undergo rounds of complete division, giving rise to A_single cells committed to differentiation, the “A_s model” (Huckins 1971; Oakberg 1971)? Or is renewal potential distributed more widely to include syncytial pairs (A_pair) or “aligned” chains (A_{aligned-4, etc.})?

To assign stem cell identity, emphasis is often placed on finding signature patterns of gene expression. However, markers may overlap between functionally distinct populations, whereas expression levels may fluctuate or adjust in response to environmental cues (Graf and Stadtfeld 2008; Morrison and Spradling 2008; Tang 2012; Donati and Watt 2015). As a result, it is often instructive to focus on a functional definition of stem cell identity based on fate behavior, calling for a “dynamic” measure (LeBlond 1965). Using tritiated thymidine incorporation as a primitive clonal mark, early tracing studies by LeBlond and colleagues provided key insights into the functional identity of stem cells in cycling epithelial tissues (Clermont and LeBlond 1953). However, it was not until the advent of genetic lineage-tracing methods that the fate of targeted cell populations could be assessed in vivo over the long term. In this approach, the activation of one or many fluorescent reporter genes following the administration of an inducing agent confers a hereditary mark allowing the fate of individual clones to be traced in undisturbed tissue over a defined time course (Kretzschmar and Watt 2012).

But how can such “static” lineage-tracing measures be used to recover information on individual cell fate decisions? For a given clone, the history of fate decisions following cell division may be ambiguous, with multiple combinations and permutations of duplicative, asymmetric, and terminal divisions leading to the same outcome. Yet, from features of the statistical distribution of clone size and cell composition, and its evolution over time, quantitative information on the “rules” of cell fate can be recovered by—often straightforward—mathematical reasoning. In particular, if SSC fate were characterized by rounds of invariant asymmetric division, as predominates in Drosophila testis, the distribution of clone sizes would simply become fixed at the “unit size” of cells supported by an individual SSC. However, if fate asymmetry was enforced only at the level of the population, so that chance SSC loss through differentiation is compensated by duplication of others, a steady increase in the average size of surviving clones would be compensated by a proportionate reduction in clone density. Such processes, involving the balanced stochastic expansion, contraction, and loss of clones, constitute a dynamics known as “neutral drift” (Klein and Simons 2011).

Clone dynamics based on population asymmetric self-renewal finds a signature in the convergence of the resulting size distribution toward statistical “scaling” behavior in which the probability C_n(t) of finding a clone at a time t postinduction with a cell number n larger than some multiple of the average n(t) becomes fixed,

Notably, from the form of the “scaling function,” f(x), insight into the pattern of self-renewal can be inferred. In particular, when the balance between stem cell loss and replacement is correlated locally in space, the scaling function takes the form f(x) = exp[ − πx²/4] in the “quasi” one-dimensional geometry of the seminiferous tubules, whereas the average clone size grows as $\langle n\left(t\right)\rangle \approx \sqrt{\lambda t}$, where λ denotes the effective average stem cell loss/replacement rate.

Applied to the mouse testis, Yoshida and colleagues used such a quantitative lineage-tracing strategy to trace the long-term fate of cells marked by the expression of Neurogenin 3 (Ngn3), a gene enriched in undifferentiated spermatogonia “primed” for differentiation (Nakagawa et al. 2007). Consistently, following an initial large-scale loss of differentiation-primed clones, only a minority of clones were found to persist over the longer term, representing those rooted in the SSC population (Fig. 1D). Consistent with populational asymmetric fate, convergence of clone sizes onto the hallmark scaling clone size dependence (Fig. 1E–H) showed that SSC maintenance involves frequent and stochastic stem cell loss compensated by the duplication of neighbors along the seminiferous tubules (Klein et al. 2010).

Although this finding provided insight into the functional fate behavior of SSCs, it did not reveal their morphological identity (viz. A_single, A_pair, etc.) nor the underlying “mechanism” of fate balance. Indeed, the resolution of such fate “rules” posed a conundrum: With SSCs separated by numerous differentiating progenies on the basement membrane of the seminiferous tubules, how does SSC duplication and differentiation become coordinated locally to maintain density homeostasis? This is the challenge faced by all renewing tissues supported by an open or “facultative” niche in which stem cells lie dispersed among their differentiating progeny (Morrison and Spradling 2008).

Recently, insight into the basis of stochastic SSC renewal has been obtained. In a mechanism that generalizes the closed niche–based regulation of GSC fate in Drosophila testis, Yoshida and colleagues have argued that SSC maintenance relies on a self-organizing feedback mechanism involving competition for “fate determinants” (Kitadate et al. 2019). By correlating the reception and consumption (internalization) of secreted factors (including fibroblast growth factor [FGF] family members), released by (lymphatic endothelial) niche cells, with the inhibition of differentiation licensing factors (including retinoic acid receptor [RAR]-γ), SSCs are able to sense their local density and adjust their fate bias in response (Fig. 2A). When the local SSC density levels are low, FGF levels are high and SSCs are biased for renewal; when the density is high, FGF levels are low and SSCs become licensed (primed) toward differentiation and loss.

Once again, evidence for the density-dependent feedback mechanism was found through the development of a minimal modeling–based scheme in which temporal changes in the local stem cell density, s(t), depend on the concentration of fate determinant, c(t), through the kinetic equation

where λ denotes the SSC division rate, and the Hill-type function h(x) = x^m/(1 + x^m) interpolates smoothly between unity (viz. cell duplication) at high concentration, c ≫ c₀, and zero (differentiation) at low concentration, c ≪ c₀. At the same time, the concentration of fate determinant varies as

where μ denotes the effective production rate by (endothelial) niche cells, k is the degradation rate, and k^′ is the consumption rate by SSCs, moderated by the factor 2h(c/c₀), which accounts for the limiting effect of receptor concentration. In steady state, the model predicts an SSC density, $s_{\ast }=\left(\mu -k{c}_0\right)/k^{\mathrm{\prime }}$, that depends linearly on the production rate of fate determinant (Fig. 2B), whereas displacement from steady state predicts an oscillatory return to homeostasis (Fig. 2C,D). Consistent with this, the analysis of genetic mouse mutants confirmed that SSC density varied linearly with the allele fraction of FGFs, whereas the recovery of SSC density following large-scale chemical ablation followed the predicted oscillatory dynamics (Fig. 2E,F; Kitadate et al. 2019).

As a basis to regulate stem cell fate, such a “Malthusian-like” feedback mechanism of density regulation, reminiscent of “quorum sensing” in bacterial populations (Miller and Bassler 2001), has many advantages. As well as providing a basis to regulate stem cell density both in homeostasis and in response to injury, it can adjust straightforwardly to heterogeneity or spatial variation in the concentration of niche factors, interpolating smoothly between an open and closed niche organization (Jörg et al. 2019). Although the model relies on the secretion and reception of local signaling factors, the feedback mechanism is easily generalized to accommodate mechanical cues based on cell crowding (Shraiman 2005; Hannezo et al. 2016; Yamaguchi et al. 2017; Yamaguchi and Kawaguchi 2019) or signals from differentiating progeny (Rodriguez-Brenes et al. 2013). Indeed, parallel studies have shown that a reciprocal feedback mechanism, in which the fate of each cell population is correlated with the reception of fate determinants secreted by the other, provides a robust and stable mechanism to regulate the proportion of tissue components (Zhou et al. 2018; Varahan et al. 2019).

Note that the competition mechanism bears similarities with the “chalone hypothesis” introduced in the 1960s as a framework to explain the potential regulation of proliferative activity in epithelial tissues (Bullough 1962). In this model, secreted factors, chalones, from differentiating cells were thought to regulate cell-cycle progression, so that the production rate of differentiated cells can be matched to their demand. Such a “mitogen” competition-like mechanism, which might function in concert with competition for niche factors, acts through proliferative activity and not through affecting changes in fate outcome. Intriguingly, support for such a mechanism has been found in the regulation of the hair follicle, acting to ensure the “episodic” regulation of the hair cycle (Plikus et al. 2009).

A further feature, implicit in the feedback mechanism, is the manifestation of fate priming: When deprived of access to fate determinants by cell crowding, the up-regulation of differentiation licensing factors leaves SSCs primed for differentiation, becoming committed only when exposed to the periodic RA signal. If SSCs regain access to fate determinants before commitment, they may readjust their fate bias toward renewal, a capability evidenced by the massively increased persistence of Ngn3-targeted clones following SSC loss (Nakagawa et al. 2007). Such behavior undermines the traditional concept of an invariant equipotent stem cell state defined by a discrete gene expression signature. Instead, SSCs are characterized by variable and changing survival potential, responding reversibly to changes in local niche factor concentration, reflected in a heterogenous and dynamic gene expression pattern (see Yoshida 2019). But are invertebrates so different? Indeed, when traced over the longer term, a similar pattern of “cell state flexibility” is found in both the male (Sheng and Matunis 2011) and the female (Kronen et al. 2014) Drosophila germline, suggesting that such behavior may be a conserved feature of stem cell regulation in gonads and potentially other tissue types (see below).

Finally, the feedback mechanism also provides a basis to explain the phenomenon of “selfish selection” in human spermatogenesis (Maher et al. 2014). By promoting a competitive advantage, activating mutations in Fgf receptor genes or Ras family confers a survival advantage on SSCs, driving nonneutral mutant clone expansion (Jörg et al. 2019) leading to progressive field transformation, similar to the process of field cancerization (Curtius et al. 2018), a vulnerability evidenced by the increased prevalence of certain congenital diseases in the offspring of aging fathers.

NEUTRAL DRIFT DYNAMICS AND CELL STATE FLEXIBILITY IN THE INTESTINAL CRYPT

We turn now to consider a second canonical stem cell system: the mammalian small intestine. The small intestine is composed of glands, crypts of Lieberkühn, that invaginate into the stroma and villi that protrude into the gut lumen (Peterson and Artis 2014). The large intestine (colon) has a similar organization but lacks villi. The small intestine is lined by a columnar epithelium comprising intestinal stem cells (ISCs), lineage-restricted progenitors, and differentiated absorptive and secretory cells (Grün et al. 2015). In homeostasis, ISCs positioned at the crypt base self-renew, giving rise to progenitors and their progeny that actively move in migration streams along the axis of the crypt and onto the villus (Krndija et al. 2019), where eventually they shed (Fig. 3A).

The identity and fate behavior of ISCs has been long in debate (Gehart and Clevers 2019): Early studies by Cheng and LeBlond (1974) placed emphasis on slender crypt-base columnar cells (CBCs), whereas, later, attention switched to “labeling-retaining” cells positioned near the +4 cell position along the crypt axis (Potten et al. 1974). However, it was not until the advent of genetic labeling techniques that the dynamics of renewing cells could be traced over the long term (Barker et al. 2007). Using a lineage-tracing strategy based on a leucine-rich repeat-containing G-protein-coupled receptor 5 (Lgr5) promoter, the monoclonal conversion of individual crypts confirmed that epithelial cell types were maintained by multipotent ISCs. But what is their multiplicity and pattern of renewal?

Once again, using a static lineage-tracing approach, evidence for the fate behavior of ISCs was sought in the statistical distribution of clone sizes, indexed by the circumferential width of labeled “ribbons” of cells that migrate along the axis of the crypt (Fig. 3B). As in the mouse testis, convergence of the clone size distribution onto the same hallmark statistical scaling behavior supported a model in which stochastic ISC loss through differentiation was compensated by the duplication of neighbors (Fig. 3C,D), leading to neutral drift of clones around the crypt base until clones are lost or the crypt becomes monoclonally fixed (Williams et al. 1992; Lopez-Garcia et al. 2010; Snippert et al. 2010). But what is the multiplicity of ISCs and their rate of loss and replacement, λ? From a fit to the average clone size dependence, it was possible to obtain only an estimate of the ratio of N²/λ, where N denotes an “effective” ISC number. However, later, these parameters were disentangled in a clever strategy based on continuous clone induction using measurements of the ratio of partially labeled to monoclonal crypts (Kozar et al. 2013). Based on this analysis, estimates of the effective stem cell number—both in mouse and human tissue (Nicholson et al. 2018; Stamp et al. 2018)—ranged from ∼N = 4 to 6, depending on the region of the intestine, a figure notably smaller than the number of Lgr5⁺ CBCs, taken by many to be a proxy for ISC identity.

But what is the meaning of the effective ISC number, and what is the basis of the stochastic renewal program? Although the minimal one-dimensional model, depicted in Figure 3C, predicts the medium- to long-term clone dynamics of ISCs, it represents only a caricature of a more complex cellular organization at the crypt base. To gain deeper insight into the cellular dynamics and fate, intravital live imaging was used to reconstruct in time-lapse lineages of individual Lgr5-targeted clones at the crypt base over a several-day time course (Ritsma et al. 2014). Extending the one-dimensional model, quantitative statistical analysis of the lineage data showed that the emergence of statistical scaling behavior in the medium term masks a more refined process in which ISCs positioned at the crypt base experience a bias toward renewal (i.e., are more likely to persist over the long term), whereas those positioned near the border of the niche (defined by the range of Lgr5 expression) are primed, but not yet committed, for differentiation (i.e., are more likely to be lost through differentiation over the long term).

Taken together, this behavior echoes that resolved in mouse testis. ISC competence is mediated by access to “fate determinants,” including Wnt, Notch, and other signaling factors derived from secretory (Paneth) cells and stromal components (Tan and Barker 2014; Santos et al. 2018). As ISCs divide, others become displaced, losing access to niche factors and leaving them poised (licensed) for differentiation, with commitment occurring in response to secondary signals. Consistently, manipulation of the levels of R-spondin, an activator of canonical Wnt signaling, showed that the size of the functional ISC pool (and range of Lgr5 expression) adjusts rapidly to changes in the niche (Yan et al. 2017b). When viewed from this perspective, a large number of intestinal cells—characterized by a dynamic and heterogeneous expression pattern linked to niche location—maintain self-renewal potential, whereas only a limited, effective number (abstracted from the mapping of the clone dynamics to the minimal one-dimensional neutral drift model) function to renew over the intermediate term, and only one, by chance, “succeeds” in the long term.

Notably, such a renewal mechanism provides a framework to understand how mutations that stimulate and/or recruit local niche factors, or activate signaling receptors, can promote nonneutral cell competition and mutant clone expansion, driving accelerated drift to monoclonality. Consistently, analysis of lineage-tracing data following the activation of Kras or deletion of adenomatous polyposis coli (APC) shows that the dynamics of mutant clones can be captured quantitatively through a minimal refinement of the one-dimensional model (Vermeulen et al. 2013; Snippert et al. 2014; Huels et al. 2018).

As well as showing evidence for reversible fate priming under normal homeostatic conditions, lineage-tracing studies based on injury models show that, echoing the behavior in the germline, intestinal progenitors normally committed to differentiation into either secretary or absorptive lineages are able to reacquire long-term self-renewal potential in response to injury, pointing to cell state flexibility (van Es et al. 2012; Buczacki et al. 2013; Roche et al. 2015; Tetteh et al. 2016; Jadhav et al. 2017; Yan et al. 2017a; Tomic et al. 2018). Although the mechanism of damage-induced “cellular reprogramming” remains in debate, one study based on single-cell profiling analysis proposes the role of a “revival” stem cell, a distinct quiescent cell that repopulates the adult stem cell compartment in response to injury (Ayyaz et al. 2019), whereas other studies based on injury models (Yui et al. 2018) and organoid assays (Serra et al. 2019) suggest the reacquisition of a developmental-like state. Yet, the manifestation of cell state flexibility, emerging as a conserved feature of multiple epithelial tissue types (Donati and Watt 2015; Clevers and Watt 2018; Ge and Fuchs 2018; Tai et al. 2019), further undermines a rigid classification of stem cell identity.

Despite the success of the neutral drift theory as a model of clonal dynamics in the intestinal crypt, puzzling features still remain. Among these is the observation of a seemingly low rate of clonal drift in the human colonic crypt, with clone fixation times measured in years (Nicholson et al. 2018; Stamp et al. 2018). Whether this behavior reflects a preponderance of asymmetric ISC divisions or betrays the existence of a lineage hierarchy characterized by a quiescent ISC subpopulation at its apex remains an intriguing and important open question.

As with the intestine, the stomach is compartmentalized into anatomically distinct regions (Mills and Shivdasani 2011) comprising the proximal forestomach, which forms a squamous epithelium, the glandular corpus, and the distal glandular pylorus, which lies adjacent to the duodenum. The glandular compartments are organized into units comprising a gland base, a neck, and an isthmus domain connecting to a pit compartment that opens out onto the gastric surface epithelium. Once again, quantitative clonal lineage-tracing studies show that maintenance of the pyloric glands mimics the organization of the intestinal crypt, with turnover of the epithelium supported by the neutral competition of Lgr5⁺ stem cells at the gland base (Leushacke et al. 2016). However, maintenance of the corpus epithelium is distinct.

Recently, tracing studies based on targeted reporters show that the base and isthmus regions of the corpus epithelium are maintained by distinct stem cell pools. The base region is replenished by slow-cycling Lgr5/Troy⁺ targeted chief cells, which function as a reserve population capable of regenerating the entire gland in response to damage (Stange et al. 2013; Leushacke et al. 2017). By contrast, the upper region is maintained by a pool of rapidly cycling isthmus-localized stem cells that give rise to lineage-restricted progeny that segregate bilaterally along the axis of the crypt, giving rise to distinct differentiated cell types. In a generalization of the intestinal crypt dynamics, quantitative lineage-tracing analysis reveals an isthmus stem cell renewal program involving “punctuated” neutral drift in which long-lived parietal cell barriers restrict the lateral expansion of clones around the gland circumference (Han et al. 2019), a behavior resonant with the slow dynamics seen in the human colonic crypt. Once again, analysis of single-cell expression profiling data suggests that, in common with ISCs, isthmus stem cells are not characterized by a single gene expression signature, but are heterogeneous, with survival potential linked to the exposure to, as yet uncharacterized, localized niche factors.

MAINTENANCE OF SQUAMOUS EPITHELIA

Staying with the theme of epithelial tissues, we now turn to skin. In mammals, the interfollicular skin epidermis (IFE) comprises a squamous stratified tissue interspersed with hair follicles, sebaceous glands, and sweat glands (Rognoni and Watt 2018). The esophagus comprises a similar structure but lacks most of the skin appendages. In both cases, the constant turnover of tissue is supported by stem cells confined to the basal layer (Fig. 4A). As some cells divide, others detach, entering into a differentiation program that leads to the formation of dead flattened squames that eventually slough from the surface.

Despite decades of investigation, the basis of epidermal homeostasis remains controversial (Rognoni and Watt 2018). Based on tritiated thymidine incorporation, Marques-Pereira and LeBlond proposed that maintenance of the rodent esophageal epithelium involved a stochastic program in which chance cell differentiation delamination is compensated by division (Marques-Pereira and Leblond 1965). Later, in epidermis, emphasis was placed on a hierarchical model in which IFE is compartmentalized into a mosaic of “epidermal proliferative units,” each supported by a stem cell and its transit-amplifying (TA) cell progeny, the latter having limited proliferative potential (Mackenzie 1970; Potten 1974). However, it was only through the advent of genetic labeling that the fate of renewing cells could be traced over the long term.

Once again, applied to mouse tail epidermis, quantitative analysis of genetic lineage-tracing data revealed progressive clonal loss allied with a linear-like increase in the average size of surviving clones (Clayton et al. 2007). At the same time, convergence of the clone size distribution onto statistical scaling behavior, C_n(t) = f(n/〈n(t)〉), with a scaling function f(x) = exp[ − x], suggested a pattern of maintenance involving the balanced stochastic loss and replacement of a single renewing “progenitor” compartment. Later, similar behavior was reported in mouse ear (Doupé et al. 2010), paw (Lim et al. 2013), oral mucosa (Jones et al. 2019), and esophagus (Fig. 4B–E; Doupé et al. 2012), pointing to a conserved pattern of self-renewal in squamous epithelia. However, in contrast to the one-dimensional dynamics of the seminiferous tubules and intestinal crypt, in the two-dimensional arrangement of the skin epidermis, it is not possible to determine whether cell duplication is correlated locally with differentiation delamination, as both translate to the same scaling dependence (Klein and Simons 2011). Elegant short-term intravital live imaging by Greco and colleagues later argued in favor of a program in which cell delamination promotes the division of neighbors, echoing the model of Marques-Pereira and LeBlond (Rompolas et al. 2013; Mesa et al. 2018).

Despite the appeal of this simple “one-progenitor cell” paradigm—a manifestation of “voter model” dynamics in statistical physics (Klein and Simons 2011)—it remains controversial. First, culture assays based on primary human keratinocytes provide compelling evidence in support of a proliferative hierarchy, with some cells showing long-term colony-forming capacity (holoclones)—used therapeutically in transplantation—whereas others show only limited proliferative potential (paraclones) (Barrandon and Green 1987; see De Rosa et al. 2019). Moreover, tracing studies based on targeted promoters (Mascre et al. 2012) provide evidence for proliferative heterogeneity in mouse tail IFE, findings supported by the identification of two distinct IFE differentiation programs in the tail (Gomez et al. 2013) and the differential tumor-initiating capacity of targeted populations (Sánchez-Danés et al. 2016).

Further work is required to resolve the question of proliferative heterogeneity in squamous epithelia. As with germline and intestinal maintenance, the ubiquity of long-term scaling behavior of clone size exposes a strength and limitation of the modeling-based method. Because multiple cell-based models of renewal translate in the long term to the same statistical scaling dynamics, fine-scale structure of the renewing population and its progeny are difficult to resolve. For example, if differentiation occurred via an intermediate subpopulation of TA-like basal cells, the long-term dynamics would be characterized by the same exponential scaling behavior. Further progress will require emphasis on the mechanisms that underpin stochastic renewal programs. Indeed, one possible explanation for the conflicting reports in skin is that the basal cell layer may constitute a facultative niche environment in which proliferative activity is regulated by local mechanical cues, but fate outcome is informed by access to signaling factors from dermal components and/or neighboring epithelial cells (Miroshnikova et al. 2018; Wickström and Niessen 2018; Mobasseri et al. 2019). By correlating the inhibition of differentiation licensing factors with reception of fate determinants, the regulation of density homeostasis may mirror the dynamics of SSCs. Variations in the local concentration or diversity of niche components, which correlate with changes in gene expression of epithelial cells, could explain the observed heterogeneity in the long-term survival potential—and tumor-initiating capacity—of targeted basal populations. Such factors could also explain variations in proliferative activity (Jones et al. 1995) and/or prime the in vitro clonogenic capacity of keratinocytes in human IFE.

Similar patterns of stochastic self-renewal have been defined from lineage-tracing studies of other “extended” epithelial tissues, from the pseudostratified columnar epithelium of mouse trachea (Watson et al. 2015) to (the structurally similar) Drosophila midgut (De Navascués et al. 2012). Indeed, in the context of trachea, differentiated (club) cells have been shown to undergo damage-induced cellular reprogramming following stem cell loss (Tata et al. 2013), similar to the cell state flexibility reported in the intestinal crypt, whereas reciprocal feedback mechanisms have been proposed as a mechanism to balance the size of the epithelial cell compartments (Pardo-Saganta et al. 2015).

HEMATOPOIESIS

A survey of the mechanisms of stem cell self-renewal would be incomplete without a discussion of blood. The blood system is composed of two major lineages, the myeloid and lymphoid (Boisset and Robin 2012). Using bone marrow transplantation to rescue lethally irradiated mice, early pioneering work by Till and McCulloch showed that these lineages are derived from a single multipotent stem cell (McCulloch and Till 1960; Siminovitch et al. 1963). Since then, transplantation assays have been widely used to study hematopoietic stem cell (HSC) behavior and lineage relationships. Combined with in vitro colony-forming assays (Iscove et al. 1970), these studies established a classic hierarchical organization in which slow-cycling HSCs, with long-term clonogenic capacity, give rise to more rapidly cycling HSCs with limited proliferative potential. These cells then give rise to cycling multipotent progenitors (MPPs) that, in turn, differentiate into the myeloid and lymphoid sublineages. But, to what extent do culture and transplantation assays reflect the dynamics of renewing cell in unperturbed tissue (a key question pertinent to other tissue types)?

Recently, advances in lineage-tracing technology have begun to allow the fate of HSCs and their progeny to be traced in undisturbed mouse tissue (Sun et al. 2014; Pei et al. 2017; Rodriguez-Fraticelli et al. 2018). Challenging the prevailing hierarchical model, these studies suggest that hematopoiesis relies largely on the renewal activity of MPPs, whereas HSCs appear to make only a minimal contribution during normal homeostasis. Indeed, without preconditioning, the most primitive HSCs (as assessed by transplantation potential) rarely contribute even during regeneration after targeted stem and progenitor cell depletion (Schoedel et al. 2016). As well as challenging the function of HSCs, these findings also question how actively proliferating MPPs are able to renew in the facultative environment of the bone marrow niche. One possibility is that, once again, in common with germline, density regulation of actively migrating MPPs may be controlled by competition for fate determinants, whereas slow-cycling HSCs contribute only infrequently to the MPP pool, protecting genomic integrity over the long term. Indeed, preliminary evidence for such behavior is revealed in the nonmonotonic pattern of recovery of MPP density following fluorouracil (5-FU) treatment of mice (Schoedel et al. 2016), a hallmark of density-dependent feedback (Jörg et al. 2019). Further support for such a model is provided by long-term clonal lineage tracing using lentiviral labeling of HSCs transplanted into rhesus macaques (Goyal et al. 2015).

In a model in which asymmetric HSC divisions at rate μ contribute infrequently and stochastically to a population of MPPs that renew through competition for fate determinants, the clone size distribution is predicted to converge to a negative binomial form, formally a “birth–death process with immigration” (Goyal et al. 2015; Simons 2016),

where λ denotes the effective MPP loss/replacement rate and n₀ denotes the average output of a clonally labeled MPP. Notably, serial measurements of clone size over a 4- to 12-yr time course provide evidence for the convergence of clone size to a stationary form with statistical size distribution that matches well with that predicted by the model (Goyal et al. 2015).

BEYOND HOMEOSTASIS

So far, our focus has been on the dynamics of adult tissues in which stem cells must self-renew. By contrast, the proliferative potential and potency of progenitors may change during development, to ensure that tissues are specified with the correct size, pattern, and composition. These contrasting demands suggest that the programs that regulate the changing balance between cell duplication and differentiation of progenitors may be different from those that control adult stem cell fate. These differences are emphasized in the context of mouse cortical development, in which quantitative analysis of clonal fate data shows that radial–glial progenitors follow an unfolding and “deterministic”-like fate program, transferring sequentially within and between “competence states” (Gao et al. 2014). Such behavior echoes that found in invertebrates, where temporal changes in neural progenitor fate are mediated by cascades of transcription factor expression (Holguera and Desplan 2018).

However, in later-stage developmental processes, the recursive output of differentiating cell types may involve the “renewal” activity of precursors, invoking fate programs similar to those seen in adult. The situation is exemplified by the development of ductal tissues, such as the mammary gland epithelium. In mouse, the mammary gland initiates embryonically as a placode-like structure along the ventral epidermis. At birth, epithelial cells form a rudimentary tree-like structure that invades into an adipocyte-rich stroma known as the fat pad. Then, during puberty, tip-localized mammary stem cells (MaSCs), restricted to the myoepithelial basal and luminal sublineages, act cooperatively to drive serial rounds of ductal elongation and bifurcation, leaving behind a ramified ductal network. But what regulates the fate behavior of MaSCs, coordinating the large-scale patterning of tissue?

Using a combination of lineage tracing and whole gland reconstruction, statistical modeling–based analyses have linked network growth to a “self-organizing” process based on a minimal set of local rules (Hannezo et al. 2017; Scheele et al. 2017). Within this framework, tips—known as terminal end buds (TEBs)—constitute a crypt-like niche environment, supporting the renewal activity of MaSCs driving ductal growth. By correlating ductal termination (signaled by collective cell-cycle exit) with exposure to factors secreted from maturing ducts, the network develops as a “branching-annihilating random walk.” Within this framework, active TEBs self-organize into a (soliton-like) pulse at the edge of the growing network. Density-dependent feedback drives the system to a critical state in which ductal bifurcation is balanced by termination. Remarkably, evidence for the same self-organizing mechanism is found in the development of other ductal tissue types including mouse pancreas (Sznurkowska et al. 2018) and kidney (Hannezo et al. 2017). Although these studies define the basis for the large-scale organization and patterning of tissue, they do not identify the mechanisms that ensure MaSC renewal. However, given the stochastic nature of individual fate decisions, niche-based competition mechanisms are likely to be relevant.

OUTLOOK

Over the past decade, understanding of the cellular mechanisms that govern tissue maintenance and repair have advanced. At the same time, perspectives on the identity and functional fate behavior of stem cells have been revised. At the turn of the century, tissue stem cells were widely thought to be discrete, individually long-lived, and defined by signature expression of molecular markers. Twenty years later, functional lineage-tracing studies have shown that stem cell potential is not “invariant” but, through exposure to local niche factors and environmental cues, cells may switch reversibly between states biased for renewal or primed for differentiation. At the same time, progenitors normally fated for cell differentiation and loss may reacquire long-term self-renewal potential following injury (see Rajagopal 2019). Through the quantitative statistical analysis of long-term static lineage tracing and short-term intravital live-imaging data, conserved “principles” of stem cell fate have emerged. Together, these studies emphasize that self-renewal potential is not invested in individual cells but rests within a “community” in which interactions with the environment allow stem cells to sense their density and adjust continuously their fate bias in response (Fig. 5). Although these “rules” of cell fate do not, in themselves, disclose the underlying molecular mechanisms, they provide a quantitative platform to frame targeted questions into the programs that regulate cell fate decision-making and the basis of cell commitment.