Order by chance: origins and benefits of stochasticity in immune cell fate control

Abstract

To protect against diverse challenges, the immune system must continuously generate an arsenal of specialized cell types, each of which can mount a myriad of effector responses upon detection of potential threats. To do so, it must generate multiple differentiated cell populations with defined sizes and proportions, often from rare starting precursor cells. Here, we discuss the emerging view that inherently probabilistic mechanisms, involving rare, rate-limiting regulatory events in single cells, control fate decisions and population sizes and fractions during immune development and function. We first review growing evidence that key fate control points are gated by stochastic signaling and gene regulatory events that occur infrequently over decision-making timescales, such that initially homogeneous cells can adopt variable outcomes in response to uniform signals. We next discuss how such stochastic control can provide functional capabilities that are harder to achieve with deterministic control strategies, and may be central to robust immune system function.

Central to the immune system’s wide-ranging defense capabilities is the close control of cell fate decisions in space and time. To respond against pathogens with a wide range of life cycles and molecular profiles, the immune system must maintain an arsenal of distinct innate and adaptive cell types at steady-state, at the right numbers and proportions. Upon pathogen detection, these cells must activate and acquire a spectrum of effector functions, with the timing and frequency closely tailored to the nature and magnitude of the threat. We have made exciting progress defining the molecular circuits controlling immune development and effector responses; however, it is still unclear how this molecular circuitry enables robust, yet tunable control of population size and proportion necessary for optimal system-level responses.

Unlike most other cell types, immune cells and progenitors are highly motile and broadly distributed in the body. As a result, they must depend less on pre-defined spatial cues and constraints to control population sizes and fractions, but instead rely on bottom-up self-organization. In the fly embryo, for instance, developing cells generate different fates at defined proportions by reading out levels of external signals in their vicinity [1,2], a process that unfolds across a largely fixed spatial domain (Figure 1A). In contrast, hematopoietic progenitors and activating T and B cells generate diverse progeny in the presence of apparently uniform signals [3–6], and do so while dramatically expanding in number, often by multiple orders of magnitude (Figure 1B). Such autonomous, bottom-up generation of heterogeneous cell populations imposes distinct functional constraints on the molecular circuits controlling fate decisions in immune cells.

Figure 1: Bottom-up cell fate control in the immune system.

Embryos (left) and other systems rely on pre-defined spatial cues and constraints to determine the sizes and fractions of differentiated cell populations. In contrast, the immune system (right) uses autonomous bottom-up mechanisms to regulate cell fate decisions in dividing precursors for population control.

Here, we discuss the emerging view that inherently probabilistic events in cells – stemming from the stochastic nature of biochemical reactions – direct cell fates and responses in the immune system and underlie proper population control [7,8]. Since early pioneering work in hematopoiesis [3,9], it has been observed at many developmental and effector response decision points that clonally related cells can show highly heterogeneous outcomes [6,10–12], even in a uniform environment [13,14]. This heterogeneity has long been suspected to reflect stochastic control at the single cell level; however, it had been unclear how such stochasticity arises on a molecular level. Here, we discuss evidence that key gene regulatory and signaling events in cell differentiation are controlled in an inherently probabilistic manner. Next, we will discuss how such stochasticity enables robust control of population sizes and fractions. We do not review asymmetric cell division, a conceptually distinct mechanism for generating clonal heterogeneity, whereby cell fate determinants are partitioned unequally but deterministically between two daughter cells during cytokinesis [14,15]. We also do not review the influences of cell-cell communication and lymphoid tissue organization for cell fate control [16,17]. For these topics, we refer the reader to these excellent reviews [18–20].

Origins of stochasticity

All biochemical processes are inherently stochastic. Because thermal fluctuations randomize the positions and conformations of biological macromolecules, the timing of a ligand-receptor interaction or a polymerase release event, for instance, can at best be described by a probability density function over time. To generate predictable responses from stochastic events, cells must then rely on the law of large numbers: by averaging over high copy numbers of molecules, or over time on individual molecules, multiple rapid events can generate coherent activity over a longer timescale. These copy-number or temporal averaging regimes predominate in many cellular processes, providing order and predictability in cell function. However, there is growing evidence that key control points in the cell fate decision-making circuitry may be gated by biochemical events that rely on low copies of molecules, and occur infrequently over decision-making timescales. The rarity of these biochemical events can lead to cell-to-cell variability in response dynamics and consequent fate outcomes (Figure 2A).

Figure 2: Rare, stochastic events controlling cell fate decisions.

(A) Regulatory systems consisting of many component molecules (left) or fast reaction rates relative to cellular response timescales (middle) generate deterministic, homogenous responses at the single-cell level. In contrast, regulatory systems consisting of slow events involving low copy-numbers of component molecules (right) generate stochastic, heterogeneous responses at the single-cell level. Both deterministic and stochastic regulatory mechanisms allow for predictable population-level responses. The rate of the regulatory event, represented as the transition from gray to green circles, and the concentration of component molecules dictate the response dynamics (generation of blue squares) as shown for two hypothetical cells. Only stochastic regulation can give rise to a bimodal population in response to homogenous signals. (B) Energetically unfavorable nucleation (signal transduction) or dissolution (epigenetic regulation) of condensates involving multivalent interactions between regulatory factors can give rise to rare, stochastic events over long time scales.

Gene activation and silencing in response to signals has long been suspected to be controlled by rate-limiting molecular events at gene loci [21–24]. Several steps in transcription are fast relative to decision-making timescales of days, and are thus subject to temporal averaging [25,26]. Transcription factors bind and unbind within seconds [27], whereas bursty transcription, involving polymerase release from a promoter, occurs over tens of minutes [28]. However, in diverse eukaryotic systems, there is growing evidence that epigenetic changes – involving stable and heritable changes in chromatin and gene expression states at individual gene loci – may involve rare, rate-limiting events with long time constants (hours to days) that can span multiple cell generations [29,30]. These rate-limiting epigenetic events can produce slow, stochastic transitions between stable expressing and non-expressing cell states; thus, if utilized to control the activation of genes that inform cell fate, these events could generate heterogeneous fate outcomes in response to homogenous signals.

If a stochastic epigenetic event produces switch-like activation of a gene locus, it would proceed independently for multiple copies of the gene in the same cell. Therefore, simultaneous monitoring of the two gene copies in single cell lineages using distinguishable reporters can reveal the existence and kinetics of such events [31,32]. Such dual-allelic reporter approaches initially revealed evidence for widespread stochastic epigenetic control in the induction of many cytokines [33–36] but have recently demonstrated that this phenomenon can function more broadly in the control of lineage-specifying genes [32]. In many cases, identified monoallelic states have been demonstrated to reflect clonally stable states at gene loci as opposed to transient fluctuations between expressing and non-expressing states [39], indicating heritable epigenetic regulation rather than transcriptional bursting as a mechanism.

To gain insights into the timescales and molecular basis of these rare epigenetic events, Mariani and co-workers combined mathematical modeling with experimentation to elucidate the kinetics of the chromatin opening step underlying stochastic IL-4 activation during CD4 T cell differentiation [35]. They found that bimodal IL-4 expression was explained by a combination of two processes – a slow, stochastic chromatin opening step that renders the locus accessible to the transcriptional machinery, and a separate process of active transcription, occurring only at an opened locus. Both the probability of locus opening and the rate of transcription were dependent on levels of the transcription factor NFAT, and, importantly, the time constant for chromatin opening was long relative to the time window for NFAT activity, resulting in only a fraction of cells activating IL-4 upon stimulation. If either the chromatin opening rate were much faster or the duration of transcription factor binding much longer, temporal averaging would prevail, and the ability of the system to generate heterogeneity would be lost (Figure 2A).

Recent studies have provided evidence that these slow and stochastic epigenetic events can function more broadly at genes encoding lineage-specifying transcription factors and thereby generate and control heterogeneity in lineage decisions during immune cell differentiation. Ng and co-workers re-applied a dual-allelic reporter approach to analyze the activation Bcl11b, a transcription factor that drives T cell lineage commitment from hematopoietic progenitors [32]. By monitoring individual T cell progenitor clones over multiple days, the authors found that each Bcl11b allele is activated independently and stochastically within and between cells, with long multi-day differences in timing between the activation of each allele. Such activation produces progenitors in transient, but heritable monoallelic states, where progenitors maintain active and inactive alleles across multiple days and cell generations before switching to a permanent biallelic state. This result provides direct evidence that rare stochastic epigenetic events, operating at individual transcription factor loci, can dictate the timing and outcome of cell fate decisions, and thereby generate heterogeneity in immune cell fate specification. Stochastic epigenetic events may also gate the activation of other lineage-specifying genes; indeed, other immune fate-specifying transcription factors, such as Pax5 and Gata3 [37,38], may also exhibit stable monoallelic states prior to activation. However, the true prevalence of such stochastic epigenetic control in immune cell differentiation has been difficult to assess due to the brief time window during which monoallelic states are observable.

Before fate-determining transcriptional programs are activated, environmental signals sensed at the cell surface must be propagated to the nucleus. Signal propagation can be highly variable in single cells [40,41], and some of this heterogeneity may stem from stochasticity in the expression of early feedback genes [42] or from pre-existing heterogeneity in the levels of signaling components [43–45]. However, signaling initiation may itself be gated by rare, inherently probabilistic events proximal to receptor engagement [46,47] that lead to responses in only a fraction of cells. Numerous studies demonstrating an all-or-none activation response of T cell receptor (TCR) antigen recognition provide evidence for this view [48–55]. Signal strength, as determined by antigen affinity and concentration, dictates the fraction of cells that respond, with a stronger signal leading to a larger fraction of responding cells. Recent evidence suggests that these differences may reflect the existence of rare stochastic signaling steps that generate an inherently low pathway activation probability upon antigen exposure [49–51]. To gain insights into these stochastic steps, Lin and co-workers used single-molecule imaging to simultaneously monitor individual ligand-receptor binding events and second messenger activation in the same T cell [51]. They found that only a small fraction of receptor-binding interactions can initiate downstream signaling pathway activation. Activation coincided with rare persistent, long-lasting ligand-receptor binding events that frequently occurred in close spatial proximity. These observations are consistent with initiation of signaling by discrete stochastic regulatory events at the membrane [10]. In other immune cell types, signal strength can also dictate the fraction of responding cells in an all-or-none manner [56], and while the origins of this heterogeneity remain unclear, it is possible that it reflects the occurrence of stochastic receptor-proximal events in other signaling cascades.

The molecular basis of the stochastic regulatory events in gene and signaling pathway activation remain unclear. Recent work has suggested that initiation of TCR signaling [57,58] and innate immune signaling [59], as well as the maintenance of both repressed and active transcriptional states [60,61], involves higher-order assemblies of protein subunits held together by weak, multivalent interactions [62–64]. Such higher-order assemblies could either represent defined macromolecular structures [65], though recent studies indicate that they could also represent liquid condensates formed by phase separation [66], an attractive mechanism for achieving cooperativity and specificity in activity. In either case, formation or elimination of such higher-order assemblies would require energetically unfavorable nucleation steps that occur slowly, thus explaining the long timescales over which events occur (Figure 2B) [62]. Importantly, while nucleation is inherently stochastic, its energy barriers are tightly controlled, thus providing a mechanism for precisely tuning the likelihood of fate-determining events in response to environmental inputs. For example, the rate-limiting epigenetic events described above are modulated by upstream transcription factors and/or chromatin-modifying enzymes [32,35]. In the case of T cell receptor signaling, modulating receptor-proximal phosphorylation kinetics can tune the activation threshold for T cells by introducing slow, rate-limiting biochemical events [67,68]. Activation threshold-tuning mechanisms also exist in B cells [69], but the stochastic nature of B cell receptor signaling has not been directly shown.

Functions of stochasticity

At both the levels of gene activation and signaling, stochastic control enables the generation of heterogeneous cell types and responses from homogenous signals, and therefore provides an economical means for cells to diversify: rather than hard-wiring deterministic responses to specific combinations of inputs, a strategy that becomes overly complex when many cell states are required and input signals are uncertain, systems can leverage fractional response dynamics inherent to stochastic regulation [70,71]. This strategy is particularly useful for immune cells, which must diversify to a spectrum of varied states and adapt to dynamic immune challenges [31,72]. While beneficial for generating diversity, stochastic control may also lead to aberrant activation in response to self-antigens and autoimmunity, particularly when probabilities are perturbed by genetic risk variants [72,73]. Such risks may necessitate careful tuning of stochastic rates for optimal immunity.

Stochastic epigenetic switches could activate or silence independently at different gene loci in the same cell; this independence could allow for the modular combination of distinct gene programs, and thus enable the generation of mixed cell states with unique functions (Figure 3A). For example, independent, probabilistic activation of Ly49 genes in NK cells generates varied combinations of class I MHC receptors [74,75]. At the population level, these mixed receptor states enable sensitivity to virally infected or tumor target cells that downregulate any one of multiple MHC alleles. Similarly, independent, stochastic regulation of IL-4 and IFN-ɣ in CD4 T cells enables expression of both Th1- and Th2-associated cytokines in response to mixed Th1 and Th2 polarizing signals [76]. Cells can ‘hedge their bets’ by assuming a mixed response phenotype until the input signals more clearly drive one response over the other. Co-expression of master transcription factors in immune cells may also reflect independent stochastic regulation [77]. Foxp3+ expressing T regulatory (Treg) cells, for example, can acquire the Th17 transcription factor RORɣt at sites of intestinal inflammation, resulting in an epigenetically and transcriptomically hybrid Treg - Th17 cell state that optimally suppresses gut-specific immune responses [78,79]. Treg cells can also drive Th1-specific immunosuppression by stably expressing T-bet in Th1 polarizing conditions [80]. These observations of environment-specific mixed states are unified by recent evidence that CD4 T cells respond additively to varied combinations of cytokine combinations, such that the mean response is a summation of responses to single inputs [81]. These findings support a model for independent, tunable regulation of key gene programs in differentiating immune cells. We propose that the resultant mixed functional states arise from stochastic regulatory events acting on individual alleles of fate-specifying transcription factors. Studies that monitor gene activation and silencing with allelic resolution and that assess population fraction control are needed to address this possibility.

Figure 3: Functions of stochasticity.

(A) The independence of stochastically regulated gene programs enables generation of mixed functional states in response to mixed signals, in contrast to deterministically regulated binary states. These mixed states may range from cells co-expressing independently regulated receptors or cytokines to epigenetically and transcriptomically hybrid cells expressing multiple fate-specifying transcription factors. (B) The heritable nature of epigenetic regulation allows precursor cells to expand in cell number prior to differentiation when the activation rate of a fate-commitment gene is slower than that of cell division.

Epigenetic gene repression is uniquely heritable through cell division, and stochastic activation events can unfold on timescales that span many cell generations [29]. Therefore, stochastic epigenetic control of cell fate-specifying genes could enable precursors to expand in number prior to differentiation (Figure 3B). For example, the epigenetic switch we have described at Bcl11b delays T cell lineage-commitment for many cell divisions, and could thus contribute to the hundred-fold expansion of progenitors during early T cell development [37,76]. Similar low probability epigenetic switches may underlie the stochastic differentiation events that regulate the output of other types of immune cells. It has been well-established that hematopoietic progenitor clones expand to heterogeneous sizes as a result of undergoing a variable number of cell divisions prior to differentiation into mature immune cells [3]. However, the average number of cell divisions completed differs among precursors of different lineages and thus gives rise to highly predictable immune cell numbers and relative proportions [82]. Similarly, the burst size of individual B and T cell clones following stimulation is controlled by stochastic decisions among daughter cells to quiesce or apoptose [5,6,83]. While these stochastic cell fate decisions are ubiquitous throughout immune cell development and function, the underlying mechanisms remain poorly understood. The heritable and cell-division-independent properties of epigenetic switches make them uniquely suited for flexibly regulating immune cell output during hematopoiesis and adaptive immune responses.

Conclusions

Stochastic control mechanisms, with their versatility and ability to guide ‘bottom-up’ differentiation from rare precursors, offer capabilities in population size and fraction control that may be hard to replicate with deterministic mechanisms. Thus, they may be more common in the immune system than is currently appreciated. Moving forward, it will be important to establish the prevalence, mechanisms, and functional significance of stochastic control in the immune system. To see whether rare, stochastic regulatory events also gate other immune fate decisions, we will need to track these decisions over time in single cells. In particular, the approach of tracking multiple instances of a key regulatory event in the same cell, that has been more widely utilized in studies of gene regulation [84,85], but has more recently been employed to analyze signaling pathway activation [51], will allow us to precisely disentangle intrinsic and extrinsic sources of heterogeneity. To elucidate the molecular mechanisms of stochastic control, it will be important to define the underlying components and the consequences of their disruption on regulation dynamics at the single-cell level. As stochastic control often generates dynamic behaviors that are not intuitively predictable, it will be useful to combine such work with mathematical models to better understand the basis of stochasticity and how it can be controlled. Finally, to test functional significance, we will need to evaluate effects of disrupting stochastic control on immune system performance, using both animal models and human studies. While complete knockout of essential factors typically yields severe pleiotropic effects, effects of stochastic control could be probed through specific manipulation of regulatory elements, such as cis-regulatory elements that control activation timing, or amino acid sequences affecting the rate of signaling protein phosphorylation [50,68,86]. For instance, a mutation of an IL2RA enhancer element was recently demonstrated to delay the activation of the IL-2 receptor in CD4 T cells, thus decreasing the fraction of IL-2Ra-dependent T regulatory cells that arise soon after activation [87]. Indeed, mutations in this enhancer element are associated with autoimmune disease in humans, linking altered activation probabilities and population sizes to immune dysfunction. Similar types of studies, coupled with a deeper investigation of the underlying physical basis, will help us understand the origins of immune dysfunction in disease, and lay groundwork for rational engineering of the immune system for therapeutic benefit.