Cancer progression is a series of evolutionary changes. Those changes include enhanced cellular proliferation, reduced cellular death by abrogating normal apoptotic mechanisms, greater invasiveness by increased expression of proteases, and improved colonizing ability to achieve metastasis [1]. In response to drug treatment, cancer cells often evolve resistance and continue to spread.

Each evolutionary step typically proceeds by acquisition of genetic or heritable epigenetic changes in cells. When does the novel genetic change arise in an evolutionary step? By the current view, rare genetic variants arise before widespread phenotypic change. The idea is that a novel phenotype always comes from a novel genotype [2],[3].

For example, rare resistant genetic variants may exist before drug treatment [4]. The drug selects those rare resistant variants, killing the other cancer cells. In progression, a genetic mutation may abrogate apoptosis, allowing that genetic clone to expand. Genotype leads to phenotype leads to evolutionary change.

But does genotype always come before phenotype in an evolutionary response [5]? Consider the alternative in which phenotype comes first, before any genetic or heritable epigenetic change. In initial drug treatment, cancer cells with the same genotype may vary phenotypically with regard to resistance. Nongenetic phenotypic variation arises by stochastic fluctuations in cellular state or by cells responding physiologically to the changed environment.

Some of the phenotypic variants may be resistant, although not genetically or heritably different from the susceptible cells. In the absence of further treatment, the surviving cells would eventually produce the same range of phenotypes as before treatment. No evolutionary change has occurred. With repeated treatment, the novel selective pressure of the drug treatment may eventually select a new genetic variant among those initially surviving cells. At that point, evolutionary change occurs. Nongenetic phenotypic variability eventually leads to acquisition of a genetic variant and evolutionary change.

Phenotypic variability may also come before genetic variability during progression and metastasis. For example, in metastatic colonization, a subset of phenotypically variable cells among a population of genetically similar cells may survive initially. Among those survivors, the novel selective pressure of the new environment may eventually favor a new genetic variant, leading to evolutionary change.

It is certainly possible that nonheritable phenotypic variants come before genetic variants in cancer evolution. But does it actually happen that way? And if so, does it matter whether phenotypic or genetic variants come first in evolutionary progression and drug resistance? How does the particular ordering influence one's understanding of progression and the approaches one might use in treatment?

Overview

In the past few years, many studies have directly measured the nonheritable phenotypic variability in populations of cells [6],[7]. Several articles have argued that nonheritable cellular variability may significantly influence the evolution of drug resistance or other key steps in cancer progression [8]–[11]. However, mainstream cancer research continues to emphasize the primary role of genetic variants or heritable epigenetic variants in initiating the evolutionary changes of cancer progression and drug resistance. The current literature on cellular variability, although interesting, has yet to make a convincing case for the fundamental role of nonheritable cellular variation in cancer.

We review some of the recent observations on cellular variability. We then extend that prior work in two ways.

First, we use fundamental concepts of evolutionary theory to show how nonheritable cellular variability likely plays a key role in the evolutionary steps of cancer progression and drug resistance. Nonheritable variability accelerates evolutionary rate particularly strongly when populations experience intense competition or face novel and extreme challenges for adaptation [5]. That intensity of competition and extremeness of environmental challenge characterize the evolutionary steps in cancer progression and drug resistance.

Second, we predict that cancer cells will often evolve to express greater nonheritable variability, because the evolutionary changes of carcinogenesis, metastasis, and drug resistance are more likely to occur in cellular populations that express enhanced variability.

Cellular Variability

Nonheritable variability takes two forms. Stochastic cellular variability arises from random fluctuations in the numbers or functions of proteins. Phenotypic plasticity arises from the response of cells to the environment. Either form of variability may allow a cell to express a novel phenotype without genetic or heritable epigenetic change. The initial expression of phenotypic novelty accelerates subsequent heritable evolutionary changes. In this section, we briefly describe examples of cellular variability. In later sections, we turn to more detailed discussion of the evolutionary consequences.

Sigal et al. [12] measured stochastic variability in protein levels in human cells. They followed the dynamics of 20 proteins in individual cells over several cellular generations. They corrected for variations between cells in the stage of the cell cycle. After correction, the protein levels varied between cells, with most standard deviations between 15% to 30% of mean levels. High protein levels in a particular cell tended to decay over a few cellular generations. This reversion to the mean shows that cellular variability does not arise from intrinsic differences between cells. Instead, random fluctuations in protein levels between cells create heterogeneity in the population.

Many earlier studies of bacteria demonstrated stochastic variation in protein levels and phenotypes [7]. Those earlier studies, combined with the study by Sigal et al. [12] and other experiments [6],[13]–[16], demonstrate the significant levels of protein and phenotypic variability that arise from random fluctuations of cellular state. Those random fluctuations generate nonheritable cellular variability.

Alternatively, cells with a common heritable genome can generate different phenotypes by their response to the environment [17]. Changes in external stresses and in signals from other cells greatly alter cellular physiology and sometimes push cells to different developmental states [18],[19]. In a changed developmental state, a cell may take on the characteristics of different tissues or of a stem cell like proliferative capacity. In those altered states, cells often change in their ability to tolerate stress and to respond to signals for proliferation or apoptosis. These types of cellular plasticity potentially generate diverse and sometimes novel phenotypes without underlying genetic or heritable epigenetic change. The new nonheritable phenotypes can initiate evolutionary change [20], including resistance to drugs.

Resistance of Cancer Cells to Chemotherapy

In the common theories of chemotherapy, resistance arises from rare mutant cells present before the start of treatment [4],[21]. Such preexisting genetic variation for resistance is expected in large cellular populations when single or double site mutations confer resistance.

With a combination of drugs applied simultaneously, the probability is very low that any cell contains all of the genetic variants necessary to protect against all of the treatments. As in current HIV treatment strategies [22],[23], combination chemotherapy minimizes the evolution of resistance particularly well when multiple mutations are needed for the initial expression of resistance. However, three recent studies of cancer find that, in particular cases, the origin of resistance begins with nonheritable cellular variation instead of preexisting genetic variants (Box 1). Each study emphasized a different mechanism that could generate nonheritable cellular variability: protein fluctuation, cellular signaling state, or histone-mediated alterations in cellular proliferation. Although these studies provide an intriguing suggestion of broader issues in the evolution of drug resistance, the role of nonheritable variability has yet to be integrated into widely understood conceptual or practical approaches to drug treatment design [8]–[10].

These studies are important, because initial drug treatments may be driving tumors to genetically based stable resistance by first selecting nonheritable phenotypes generated by cellular or tissue variability. If so, we may need to rethink approaches to treatment design. Before turning to aspects of treatment, we first explain how nonheritable variability accelerates the evolution of heritably based resistance.

Nonheritable Variability and Evolutionary Theory

To move forward, we need a clear conceptual framework. How should the recent results on nonheritable variability in drug resistance be interpreted in relation to genetic variability? What is the broader significance in the context of other evolutionary steps in carcinogenesis? How should new experiments be designed in light of the potential role of nonheritable variability?

These questions are timely, because new technologies provide the tools to measure both genetic variability and nonheritable aspects of cellular variability. Those refined tools offer great opportunity, but we need clear principles to exploit that technology and to gain a deeper understanding of carcinogenesis and drug resistance. Because the problems concern evolutionary change of cells and tissues, the principles arise from evolutionary theory.

The idea that nonheritable phenotypic variability accelerates evolutionary rate has a long history. In cancer research, Rubin [20] clearly described observations and concepts in which nonheritable phenotypic variability initiates key steps in carcinogenesis. Rubin did not connect his ideas to classical evolutionary theory, instead roughly sketching out the logic as a novel view of cancer evolution. Later authors have repeated this argument for drug resistance or, more generally, for the evolutionary steps in the development of cancer [8]–[11]. Those later articles sometimes mention the literature from evolutionary theory, but do not develop that connection in a way that leads to specific predictions or novel insights.

Within the evolutionary literature, many recent reviews discuss the relation between nonheritable variation and evolutionary rate. The extensive history goes back to Baldwin [24] and includes several subsequent theoretical refinements [5],[25]–[33]. For our purposes, we can start by thinking of evolutionary adaptation as analogous to the problem of searching for an improved outcome in a complex space of alternatives [34]. With regard to cancer, “improved outcome” means modified cells or tissues that grow beyond normal restraints and may be regarded as outcompeting normal cells and tissues.

The key issue concerns how new traits arise [5]. How does the evolving system of cancer tissues find novel phenotypes that improve competitive success and cause the spread of the tumor? The search process for evolutionary novelty can be divided into short and long components that differ by the degree of phenotypic change required for cells and tissues to respond to environmental challenges [32].

“Short-range search” occurs when a cellular clone expresses a variety of phenotypes that differ by relatively small changes, such as altering ligand binding or modulating the pace of cell cycle progression. “Long-range search” occurs when cells derived from a clone express phenotypes that differ by relatively large changes, such as altered developmental state, tissue type, or other variant expression unlikely to occur as a stochastic phenotypic fluctuation. Short-range and long-range search play the role of discovery in the evolutionary search for novel traits. In addition to those two types of discovery process, “local adaptation” by genetic variation and selection refines the match between existing traits and the environment.

The following sections discuss each of these three components of evolutionary variability and adaptation. Box 2 shows how each component arises in the adaptive vertebrate immune system. Vertebrate immunity provides the best-known biological example of how the three aspects of variability and adaptation combine to determine the overall nature of evolutionary change. By studying vertebrate immunity, one may gain insight into the different processes of variability and adaptation in cancer evolution.

Synergism between Different Adaptive Processes

Understanding cancer progression and drug resistance requires understanding the evolutionary processes that change cellular populations and tissues. We have emphasized three types of evolutionary process that differ by the underlying cause of phenotypic variability and by the amount of phenotypic change required for cells and tissues to respond to environmental challenges. These three processes are not mutually exclusive. In fact, the changes required to achieve drug resistance or the steps in carcinogenesis may often arise by synergism between the different adaptive processes.

Consider the problem of drug resistance. In several studies, empirical evidence suggests that preexisting genetic variants provide the phenotypic variability selected by novel drug-specific environments [4],[21]. At the same time, we described three experiments in which nonheritable phenotypic fluctuations provide the phenotypic variability that allows initial survival (Box 1). Many bacterial studies also show that initial survival in the presence of drugs depends on nonheritable phenotypic variability [37]–[39].

How can we reconcile the evidence for preexisting genetic variation with the observations on nonheritable phenotypic variability? Part of the answer is that, to some extent, one finds what one is looking for. Experimental designs that seek preexisting genetic variability will often find it, and experimental designs that seek nonheritable phenotypic variability will often find it.

Focusing on preexisting genetic variability versus nonheritable phenotypic variation may be a misleading contrast [5]. Instead, we think that a natural synergism occurs between the different types of variability [32]. Nonheritable physiological plasticity often provides a type of long-range search, in which extreme or novel environments induce novel phenotypes. Preexisting genetic variants can also generate novel phenotypes in long-range search when the genetic changes induce major phenotypic shifts. By contrast, nonheritable stochastic fluctuations often provide a type of short-range search. The complex process of evolving a phenotype that can survive an extreme or novel challenge may sometimes require both a long-range component to create a phenotype somewhere near a viable solution and a short-range component to provide additional phenotypic variability that closely matches the particular environmental challenge (Box 2).

Our main point is that genetic and nonheritable phenotypic variation are not alternatives. Rather, to understand the complex evolutionary processes in drug resistance and in the stages of carcinogenesis, one needs to consider the synergisms between different aspects of evolutionary search and adaptation.

Open Problems

We described three experiments in which nonheritable phenotypic variability played a key role in drug resistance (Box 1). Those experiments, along with the theory we have developed, demonstrate that nonheritable variability can be important. But how important and widespread are the consequences of nonheritable variability? Study of the following open problems may help to answer that question.

Conclusions

The current cancer literature is dominated by the view that genetic or stably heritable epigenetic variants initiate the key evolutionary changes in carcinogenesis and drug resistance. Against that view, our theme is that nonheritable cellular variability often plays a key role in initiating the major evolutionary steps of carcinogenesis and drug resistance.

The main point is that tumorigenesis and drug treatment typically impose novel or extreme environmental challenges to cells and tissues. The rate of evolutionary change in response to such challenges is greatly enhanced by nonheritable phenotypic variability, often acting synergistically with genetic variability. A clear understanding of cancer and its treatment requires closer attention to these fundamental evolutionary processes.

Our view has two important implications for drug treatment design. First, the mechanisms of tumor resistance may combine genetic mutations and nonheritable phenotypic fluctuations. If so, then combinations of drugs should target both the likely genetic resistance mutations and the likely nonheritable resistance mechanisms, rather than solely targeting genetic mutations.

For example, a drug may target a particular cellular protein. Resistance may arise by a mutation in the targeted cellular protein or by nonspecific stress responses that compensate for the loss of function of the targeted protein. Drug combinations for both the targeted protein and the nonspecific stress response may be required. Alternatively, initial resistance to a drug that targets a particular protein may arise by nonheritable variability in rate of cellular proliferation. A drug combination that targeted the particular protein and also reduced fluctuations in cell cycle state may be effective.

The second implication for treatment arises when drugs drive cellular populations to a resistant state via a nonheritable intermediate state. For example, those cells initially resistant to a treatment may, by purely nonheritable phenotypic fluctuation, express a greater stress response than those cells that die. With continued treatment, some of those initially resistant cells may subsequently acquire a heritable change that upregulates the stress response. Those mutated cells will tend to increase, causing the heritable fixation of a generalized resistance mechanism. Subsequent alternative drugs may then perform poorly. In this case, initial drug combinations should include stress response inhibitors.

In these two examples, we have mentioned stress response as a form of nonheritable resistance. We chose that response because it is easy to understand how resistance to stress may enhance resistance to drugs. However, our point is more general. Any physiological mechanism that promotes nonheritable phenotypic variability in resistance tends to enhance evolutionary rate, including the rate at which stably heritable drug resistance emerges. Closer attention to the broad range of mechanisms that generate nonheritable phenotypic variation is likely to improve our understanding of resistance and the design of combination treatments.

Acknowledgments

Robin Bush, Tim Fessenden, Carlo Maley, and Robert Rosner provided helpful comments.