Health and Fitness at the Single-Cell Level

The Oxford Dictionary defines health as “The state of being free from illness or disease.” But in the context of a cell in a multicellular organism, this definition lacks precision; cells perform functions in the body, and we hesitate to describe a cell that does not perform its requisite duties as “healthy.” The word “health” is derived from the Old English hælth, which derives from a Germanic term meaning “whole,” which comes closer to the intuitive understanding of health.

The focus of this article is health and fitness at the level of a single, generic cell in a multicellular organism, predominantly in mammals. Multicellular organisms undoubtedly evolved from single-celled organisms, and in considering health, the level at which health operates can have contrasting outcomes. As a simple example, a cancer cell that is “healthy” (see below for a working definition) can ultimately kill the individual. Indeed, all cancer therapy is based on making transformed cells unhealthy.

Thinking about health is largely based on the above definition (the absence of illness or disease), and most efforts to promote health have historically focused on the elimination of disease. However, an emerging view (1) proposes that health is an active process that can be directly enhanced through the manipulation of systems that have evolved to enable proper function. Herein, I discuss how to apply such thinking to a single cell and consider how this may inform thinking about disease processes at the organismal level.

Fitness is often conflated with health, although they are often distinct. Fitness is a formal term indicating the likelihood that, and extent to which, an individual will contribute genetic material to the next generation. We understand that an animal that is incapable of doing so, and hence has reduced fitness by this definition (“reduced,” because in some circumstances it may nurture-related offspring) can nevertheless be healthy.

Single cells in a multicellular organism have, for the most part, identical genomes. Nevertheless, cells of a given type necessarily compete for resources (e.g., growth and survival factors, nutrients, structo-spatial components of a niche or extracellular matrix), and the cells that compete successfully are those that have the greatest potential to survive and proliferate (Fig. 1A). That is, they are the most fit. Cancer cells are successful competitors, and despite their impact on organismal health, they are more fit than those of the tissue from which they derive. Bacteria and viruses that infect cells can increase the competitive advantage of their “hosts,” conferring fitness upon them (often to the detriment of their host organism), a situation that emerges from antagonistic host–microbe coevolution (the so-called Red Queen Effect; refs. 2, 3). In contrast, a cell that is terminally differentiated, and thus cannot proliferate, has a fitness of zero (unless it can dedifferentiate and proliferate) at the level of the single cell. However, such cells can be extremely important for fitness at the organismal level.

Health is much more difficult to define at the single-cell level. In multicellular organisms, the active process of health can be determined by monitoring an array of variables, and its maintenance is a function of homeostatic regulatory systems (1). Using a similar strategy, we can determine the health of a single cell by monitoring different cellular parameters (e.g., metabolite levels; osmolarity; DNA, plasma membrane, and organellar integrity; and energy). Thus, a “healthy” cell is one for which such parameters are in the normal range for the state of the cell (e.g., quiescent or proliferating). As with organisms, maintaining this normal range in the face of perturbation is a function of homeostatic regulatory systems. At the single-cell level, we call these “stress responses” (discussed further, below). Although the “function” of a cell in the context of the body can certainly be important for “organismal” health, here I avoid asserting a requirement for such tissue and systemic function in describing the health of a single cell. Thus, a pancreatic islet cell that survives with its normal parameters intact but fails to produce insulin may nevertheless be “healthy” by this definition. This can similarly apply to a “healthy” cancer cell (as mentioned above), although not all cancer cells are healthy by this definition. Admittedly, the interplay of health at the single-cell, tissue, and organismal levels deserves a more complete analysis than is provided here.

The features of health and fitness at the different levels of selection (cell versus tissue versus organismal) are key to our understanding of the evolution of multicellularity (4). This provides a framework for consideration of cellular health and how it is maintained in the context of the whole organism. Although a multilevel analysis would be enlightening, herein, I restrict the discussion to single cells of a given type because of the complexity of such interactions.

Most fully functioning, healthy cells exist in a physiologically controlled environment that limits challenges to that healthy state. There are, however, exceptions to this: those cells at barriers to the outside world (e.g., skin, lung, and gut) can be damaged by the environment, and cells that travel to different tissues as part of their normal function (e.g., immune cells) must respond to changes in the different tissue environments. Of course, any cell can undergo cellular stress when the organism is stressed by changes in the outside environment (e.g., nutrient or oxygen deprivation, and infection).

Another process that affects cellular health is competition between cells. As with competition among individuals, this can result in cellular stress. During such competition, the cells that will become the “losers” produce signals that help make the “winners” win, whereas the cells that will become the winners signal the losers to lose, and often cause them to die (5, 6). These features of competition between cells likely have emerged as a consequence of multicellularity and the contrasting selective forces operating at the distinct levels of cells and organisms (4). Despite our realization that cell competition exists, relatively little is known about this process in mammals, and most of the information we have comes from studies in Drosophila. In general, cells attain winner status if they have elevated levels of Myc (in flies and mammals), Wingless (flies), or Stat (flies; ref. 7). In contrast, cells that engage p53 (mammals; refs. 8, 9) or Notch signaling (mammals; refs. 10, 11) attain loser status. In effect, the latter represent genes that promote cellular cooperation (Fig. 1A).

Cell competition occurs during immune responses and is best exemplified in the context of antibody affinity maturation (12). As antigen levels decline, B cells bearing the highest affinity antigen receptors effectively compete (win) against cells with lower affinity receptors, resulting in the production of higher affinity antibodies over time. Although this effect is well known, the principles of cellular competition gleaned from other systems have not been applied to antibody-affinity maturation and B-cell competition. That is, we do not know how the “losers” affect the “winners,” and vice versa.

During cell competition, the ability of a winner cell to induce cell death in loser cells has best been studied in Drosophila. This appears to depend on factors, such as the ligand Spaetzle, that engage Toll receptors and NF-κB responses, driving the expression of apoptosis proteins (in flies, these are Hid and Reaper) in the loser cells (13). In contrast, the activation of NF-κB in mammalian cells is usually associated with inhibition of apoptosis, although this is by no means universal (14).

Perturbations to normal cellular functions elicit responses in the cell. There are three types of general cellular responses to such challenges, as well as specific responses in a particular cell type. These are homeostatic control mechanisms, cellular salvage and repair processes, and the responses to intracellular infection. These function in the context of single cells, but are shaped by principles established in the evolution of multicellular organisms; most, but not all, cells are “dispensable” and can be sacrificed if the challenges to cellular health can affect the health of the organism as a whole. Such sacrifice takes the forms of cell death and senescence, which protect the organism from oncogenic transformation and infection, events at the single-cell level that can profoundly affect the whole organism.

Herein, I do not detail the diverse responses to stress, each of which would itself require a lengthy review, or focus on cells of the immune system. Rather, I consider these responses at the level of single, generic cells, in the context of whether and how they affect their own outcome, as well as that of identical, neighboring cells, and in doing so, potentially affect the outcome of competition among cells. This discussion includes chronicling how responses to cellular stress affect a cell's health and fitness, and highlighting how diverse responses to stress trigger the production of signals from the stressed cell that can affect other cells (cytokines and chemokines) and how the intrinsic response of the stressed cell (repair or death) affects the production of such signals. This discussion has implications for our thinking about the more complex variables that exist among multiple cell types, for example, in the tumor microenvironment.

All cellular stresses elicit responses that are often specific to the stress and can be specific to the cell type. These include metabolic stress (nutrient or oxygen deficiency), endoplasmic reticulum (ER) stress, mitochondrial stress, osmotic stress, accumulation of unfolded proteins, and damage to the DNA, plasma membrane, or organelles. Another important source of cellular stress is intracellular infection.

In many cases of extreme stress, the associated damage is so extensive that the cell dies by necrosis, as cellular integrity is directly lost. However, in most instances, there are two possible outcomes: the cell survives and repairs or the cell dies by a regulated cell death mechanism. In some forms of stress (e.g., DNA damage), there is a third alternative: the cell may undergo senescence.

Most regulated cell death in animals is by the process of apoptosis, which is also probably the best understood cell death process. Unlike necrosis, cells that die by apoptosis do not lose plasma membrane integrity before being cleared by phagocytes such as macrophages. Apoptotic cells effectively “package” themselves for phagocytic clearance, cutting the chromatin, reducing size, and displaying “eat me” signals on the surface (15). These events are caused by the action of caspase proteases, which are activated by the engagement of specific molecular pathways.

A broad variety of stresses engage signaling that converges on the so-called integrated stress response (ISR; ref. 16). The key event of the ISR is the reprogramming of translation, executed by phosphorylation of its central player, eIF2α. As part of the translation initiation complex, eIF2α is required for 5′Cap-dependent protein translation. The phosphorylation of eIF2α transforms it into a competitive inhibitor of eIF2B, which leads to an attenuation of global protein translation and promotes the translation of selected mRNAs, including ATF4. This rapid translational switch is the immediate adaptive change that opens a time window for the cell to resolve the stress or, if it cannot do so, undergo cell death. Ultimately, it is the expression of genes induced by ATF4 in combination with its diverse cofactors that is the main effector of the ISR (17).

The phosphorylation of eIF2a is mediated by any of four eIF2a-kinases: PKR-like ER kinase (PERK), general control nonderepressible 2 (GCN2), heme-regulated eIF2a kinase (HRI), and double-stranded RNA-dependent protein kinase (PKR), each of which is activated by dimerization in response to specific stresses. ER stress engages PERK, amino-acid and heme deprivation engage GCN2 and HRI, respectively, and viral infection induces the activation of PKR. All four can also be induced in response to reactive oxygen species (ROS; Fig. 1B).

In every case, there appears to be a decision to either alleviate the stress or undergo apoptosis. Although it is widely assumed that this decision is one of degree, it is not this simple: at an LD₅₀ dose of a stress-inducing agent, 50% of genetically identical cells in a cell line will die, despite all cells receiving the same insult. The remaining cells engage the reparative arm of the stress response, and this is required for their survival under these conditions. In general, the ISR promotes cell survival by limiting protein synthesis, and by the induction of transcription of genes that promote the folding and degradation of misfolded or aggregated proteins, amino-acid transport, and metabolic adaptation, and protection from oxidative stress (16). Additional repair processes are engaged in response to specific stressors via other mechanisms. This idea, that the LD₅₀ response can provide deep insights into mechanisms of health, has been insightfully explored at the organismal level in the elucidation of healthy host–microbe interactions (1, 18). Indeed, much of the thinking herein was inspired by this pioneering work.

In the cells that die under conditions that engage the ISR, one mechanism involves the ATF4-mediated induction of the expression of two genes encoding the proteins, PUMA and NOXA (19, 20). These so-called BH3-only proteins of the Bcl-2 family function to neutralize the antiapoptotic Bcl-2 proteins, allowing apoptosis to proceed. Stress-induced apoptosis also involves other factors, but this raises an issue: ATF4 induces various protective responses to cellular stress, but it can also cause apoptosis; what determines the outcome on a cell-by-cell basis? It is frequently asserted that cells sense the level of stress, and that this dictates the repair or die decision. But if this is mediated by the same stress response, is the determining variable simply the level of response, e.g., the level of ATF4?

Evidence from another stress-induced response dichotomy suggests a more complex answer to this question. The transcription factor and tumor suppressor p53 is expressed at the mRNA but not the protein level in unstressed cells, owing to the ability of the p53 target MDM2 to ubiquitinylate p53 for degradation. DNA damage, persistent oncogene activation, and un-complexed ribosomal proteins all trigger the stabilization of p53 by disrupting the p53–MDM2 interaction via different mechanisms (21). As a result, p53 induces a variety of responses, including homeostatic repair responses and apoptosis (22). The induction of apoptosis by p53, like that of ATF4, is largely dependent on p53-induced expression of PUMA and NOXA (23), and a repair or die dichotomy is often seen as “if the stress is too great, and repair is insufficient, the cell defaults to cell death.” However, the decision is more complex. Studies using systems in which endogenous p53 is tagged with fluorescent proteins showed that it was the kinetic pattern of p53 stabilization, rather than the levels obtained, that predicted the outcome of the stress response: cells in which p53 protein levels either spiked and returned to baseline, or showed a regular oscillation in p53 levels, resolved the stress and survived. In contrast, cells in which p53 levels continued to rise underwent apoptosis (24). This oscillation is a result of the ability of p53 to induce the expression of MDM2, which in turn induces the degradation of p53 (Fig. 1C). Indeed, the frequency of the p53 oscillation is determined by specific p53 response sequences in the MDM2 promoter, which differ between rodents and primates (25). It is also known that modifications of p53, such as acetylation, affect its ability to engage proapoptotic gene expression (26), and the relationships between the kinetics of p53 stability and such modifications are unknown.

Oscillation of the ISR has been predicted in a computational model (27) involving the function of GADD34, the product of an ATF4 target gene, which targets phospho-eIF2a for dephosphorylation. Oscillation of protein translation following viral infection-induced stress has also been observed experimentally (28). ATF4 is subject to ubiquitinylation by SCF^TrCP and rapid proteasomal degradation, leading to its very short half-life. This enables a fine-tuned regulation and increases the impact of upstream regulators. Whether ATF4 levels oscillate in response to cellular stress is not known. Modifications of ATF4, including methylation and phosphorylation, appear to dictate its ability to engage the expression of PUMA and NOXA and subsequent apoptosis (29, 30). Whether there is any relationship between protein modification of ATF4 and oscillation of the ISR is unknown, as is the role of such a relationship in the decision to survive or die. However, oscillating kinetics of proteins with short half-lives and negative feedback loops often reflect the interactional dynamics of intricate networks (31), which might determine the cell's fitness. As such, the diverse ATF4-interacting partners determine the transcriptional preferences of ATF4 and the resultant cell fates (Fig. 1C; refs. 17, 32).

From the perspective of cell fitness, senescent cells are similar to “dead,” in that they have no fitness. Senescence is induced by DNA damage and the decision to repair or senesce is observed to be a consequence of the extent and persistence of the damage (33). Although senescence is one potential outcome of p53 activation, via the induction of the cell-cycle inhibitor p21, senescence can also occur in a p53-independent manner, via induction of p16INK4. The decision to senesce or die is in part cell type dependent, but either can occur in the same cell type, and how this decision is made is not clear (34). The common assertion that p21 and p16INK4 inhibit apoptosis has little mechanistic basis; no molecules in the core apoptotic pathway (e.g., BCL2 family proteins) functionally interact with p21 or p16INK4. Although cell-cycle components can modify BCL2 proteins (e.g., by phosphorylation; ref. 35), effects of such modifications are relatively minor (35, 36) and cannot account for the “choice.”

Although much is known about the life–death decisions in response to cellular stress, and while mathematical models of these decisions (37, 38) are consistent with predicted outcomes, the application of new technologies to single-cell analysis will be useful in gaining a better understanding of how the response to cellular stress relates to cellular health and fitness. For example, we can envision using markers of cell fate to deconvolute single-cell RNA sequencing to gain a better understanding of the distinct trajectories that determine if a cell will repair and survive, die, or senesce. To date, no such single-cell analysis of cell competition has been reported, and whether winner versus loser trajectories will reflect life versus death trajectories is unknown.

The abscopal effect, originally described as the ability of irradiation of a local tumor to effect regression of a distal tumor, was noted in 1953 (39). This is now largely ascribed to the induction of antitumor immunity (40); however, cells in culture can also exhibit abscopal effects. Any stress that engages DNA damage responses can elicit changes in cocultured, untreated cells (41).

Although technically the abscopal response is defined as an effect of tumor irradiation on a nonirradiated tumor in vivo, here I use the term “abscopal” to refer to any effect of a stressed cell on a neighboring, otherwise unstressed cell (others might prefer to think of it as a “bystander” effect). Although various mechanisms have been suggested to play roles in such intercellular effects, e.g., release of damage-associated molecular patterns (DAMP), exosomes, transfer via GAP junctions, reactive oxygen and nitrogen species (41), I restrict the discussion herein to the release of cytokines and chemokines as key mechanisms of the abscopal effect.

Virtually any cellular stress can trigger the production of inflammatory cytokines and/or chemokines from stressed cells (42). These stressors include DNA damage (41), ER stress (43), plasma membrane damage (44), osmotic stress (43), oxidative stress (43), heat shock (43), and intracellular infection.

These inflammatory responses resemble those induced by the engagement of pattern recognition receptors, such as Toll-like receptors (TLR), and the senescence-associated secretory pattern (SASP), both of which are dependent, at least in part, on activation of the transcription factor, NF-κB (45). Indeed, the SASP is a feature of senescent cells, and both the inflammatory cytokine response and senescence occur as a consequence of DNA damage, with the extent of each related to the duration and severity of the DNA damage (46, 47). Senescence resulting from DNA damage is dependent on mitochondria (48), and mitochondrial ROS can activate NF-κB (49), possibly providing a link between mitochondrial metabolism and the SASP. Mitochondrial dysfunction associated with aging can induce senescence, although via a mechanism that is distinct from that of DNA damage and with a different secretory profile (50).

In the case of DNA damage, activation of the ATM kinase (which phosphorylates and stabilizes p53) induces signaling that leads to activation of NF-κB (51, 52), and this may be responsible for inflammatory cytokine expression following DNA damage. However, stabilization of p53 can also activate NF-κB (53), although it is not clear how this occurs.

The ISR plays roles in stress-induced inflammatory responses, although, as with the mechanisms of cell death, there are divergent pathways responsible for the resultant inflammatory response that are specific to the stress. ATF4 can directly drive the expression of IL6 (54), and promotes TLR-induced inflammatory cytokine production (Fig. 1D; ref. 55). In addition, ATF4 induces the expression of the TRAIL receptor TNFR10b (56). In ER stress, unligated TRAIL receptor can signal NF-κB activation and inflammatory cytokine production (57), and can promote apoptosis via its death receptor signaling properties (58). ER stress is frequently a consequence of a wide variety of cellular insults, and this TRAIL receptor–dependent mechanism is also induced by microtubule toxins (Fig. 1D; ref. 57). However, the role of the TRAIL receptor in inflammatory signaling has not been examined in other forms of cellular stress.

Responses to cellular stress have not been integrated into our emerging understanding of cellular fitness. Under conditions of cell competition, it is likely that cell stress occurs, but how do the different responses to stress affect winner versus loser status? If a cell dies, its ability to synthesize and secrete chemokines and cytokines is curtailed, whereas its release of DAMPs and other preformed mediators, including cytokines of the IL1 family, can ensue. Cells that die are rapidly engulfed by phagocytes in the process of efferocytosis, and subsequently the phagocyte produces pro- or anti-inflammatory cytokines, dependent on the mode of cell death (59). Although the discussion herein focused on responses from individual cells, this phenomenon adds considerable complexity to the calculus of cell competition in a physiologic setting.

As in life versus death decisions, single-cell variation in the inflammatory response to cellular stress has not been investigated, and it remains possible that winner versus loser status in the context of cell competition is paralleled by differences in this response. It is not clear, however, what we should expect any such correlations to be. For example, T cells actively catabolize the cytokines that drive their expansion (60), and this effect appears to be dependent on the expression of c-Myc, a major driver of cellular fitness. The conclusion that c-Myc drives cytokine catabolism is based on the observation that in T cells in which c-Myc has been acutely deleted, activation-induced cytokine gene expression is unaffected, whereas protein levels of the cytokines in culture are markedly elevated (61). Perhaps this is a general phenomenon: winners in cell competition actively deplete growth factors, depriving the losers of the fitness that such factors provide.

The concepts of health and fitness at the single-cell level are only the beginning of our understanding of the integration of cellular homeostasis, cell death, and the ensuing effects on the population of cells into the physiology of health, and ultimately, the fitness of an individual. This discussion only touched on these rich issues, and there is much to learn in the coming decade as this integration proceeds.

D.R. Green reports personal fees from Ventus Therapeutics and Inzen Therapeutics outside the submitted work.

The author is grateful to Halime Kalkivan, Beiyun Liu, and Piyush Sharma for making the figures. D.R. Green is supported by grants from the U.S. National Institutes of Health (AI40646, AI44828, AI123322, and CA231620) and American Lebanese Syrian Associated Charities (ALSAC).