Content uploaded by Christopher Lean
Author content
All content in this area was uploaded by Christopher Lean on Feb 28, 2016
Content may be subject to copyright.
1 23
Biology & Philosophy
ISSN 0169-3867
Biol Philos
DOI 10.1007/s10539-015-9511-1
The evolution of failure: explaining cancer
as an evolutionary process
Christopher Lean & Anya Plutynski
1 23
Your article is protected by copyright and all
rights are held exclusively by Springer Science
+Business Media Dordrecht. This e-offprint
is for personal use only and shall not be self-
archived in electronic repositories. If you wish
to self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com”.
The evolution of failure: explaining cancer
as an evolutionary process
Christopher Lean
1
•Anya Plutynski
2
Received: 22 November 2014 / Accepted: 12 November 2015
Springer Science+Business Media Dordrecht 2015
Abstract One of the major developments in cancer research in recent years has
been the construction of models that treat cancer as a cellular population subject to
natural selection. We expand on this idea, drawing upon multilevel selection theory.
Cancer is best understood in our view from a multilevel perspective, as both a by-
product of selection at other levels of organization, and as subject to selection (and
drift) at several levels of organization. Cancer is a by-product in two senses. First,
cancer cells co-opt signaling pathways that are otherwise adaptive at the organismic
level. Second, cancer is also a by-product of features distinctive to the metazoan
lineage: cellular plasticity and modularity. Applying the multilevel perspective in
this way permits one to explain transitions in complexity and individuality in cancer
progression. Our argument is a reply to Germain’s (2012) scepticism towards the
explanatory relevance of natural selection for cancer. The extent to which cancer
fulfills the conditions for being a paradigmatic Darwinian population depends on the
scale of analysis, and the details of the purported selective scenario. Taking a
multilevel perspective clarifies some of the complexities surrounding how to best
understand the relevance of evolutionary thinking in cancer progression.
Keywords Cancer Multilevel selection Darwinian populations Tumors
By-product of selection
‘‘Through failure we understand design.’’ Frank, 2007.
&Christopher Lean
christopher.hunter.lean@gmail.com
1
School of Philosophy, Australian National University, Canberra, Australia
2
Department of Philosophy, Washington University in St. Louis, St. Louis, MO, USA
123
Biol Philos
DOI 10.1007/s10539-015-9511-1
Author's personal copy
Cancer as an evolutionary process
At first glance, cancer seems the epitome of disorganized growth, rather than
adaptive evolutionary change. Indeed, cancer is often defined as a failure of the
otherwise functional regulation of cell birth and death. This view would seem to be
fundamentally at odds with treating cancer progression as a process of natural
selection. However, a trait that is dysfunctional at one temporal or spatial scale of
analysis may well constitute ‘‘efficient functioning’’ at another (Hausman 2012).
Indeed, this should not surprise us, given that multicellular organisms are (more or
less) hierarchically organized, with functional parts within parts, and subsystems
within systems. There is an active research program underway that takes this insight
as foundational; these researchers treat cancer progression as a process of natural
selection on cells and cell lineages (Michor et al. 2004; Merlo et al. 2006; Anderson
et al. 2006; Pepper et al. 2007; Greaves and Maley 2012). The idea that cancer
progression may be viewed as an evolutionary process is not new: Gause (1966)
compared a population of cancer cells to an evolving population of microbiota.
Nowell (1976) and Cairns (1975,1978) similarly argued that ‘cancer can be viewed
as the operation of Darwinian selection among competing populations of dividing
cells’ (Cairns 1978, 151).
1
What started as a minority view has now become an
active area of research, drawing upon mathematical modeling, computer simulation,
experimental evolution in model systems, and even the generation of phylogenies,
or branching evolution in cancer progression and metastasis, using high throughput
sequencing of tumors and metastatic lesions from deceased patients (Gerlinger et al.
2012; Navin et al. 2011; Yachida et al. 2010).
Philosophers of biology have begun to take notice of this research. Peter
Godfrey-Smith in ‘Darwinian Populations and Natural Selection’’ refers to cancer
progression as ‘‘re-Darwinization’’: a ‘‘decoupling’’ of cellular fitness from that of
the organism (Godfrey-Smith 2009). The idea of cancer being more or less
‘‘decoupled’’ in its fitness from its host is a useful theoretical starting point, as it
captures the two extremes of cancer’s relationship to the body. Cancer cells are
relatively autonomous as compared to other cells in the body, but autonomy is a
matter of degree. In some rare cases, e.g., infectious canine cancers, cancer can
become as autonomous from the host body as any virus is from its host. In the case
of Tasmanian devils, and some Canines, there are truly transmissible cancers:
Canine Transmissible Venereal Tumor CTVT and Devil Facial Tumor Disease
(DFTD). However, most cancer cells are not autonomous in this strong sense: most
cancer cells (in vivo) cannot transmit between hosts, or grow in culture. Indeed, it
took decades to identify a cancer cell line (HELA) that could grow indefinitely in
culture. What determines the degree of ‘‘autonomy’’ and fitness of cells and
collectives is not only on cell-intrinsic features, but also local conditions, such as
signals from neighboring cells, as well as structural features of the tissue
microenvironment (Bissell and Hines 2011). In vivo, cancer cells’ degree of
autonomy changes over time, dependent upon these local conditions, as well as the
stage of advancement of a cancer. While cancer cells are in competition, ultimately,
1
For a history of evolutionary perspectives on cancer, see, Morange (2012).
C. Lean, A. Plutynski
123
Author's personal copy
their survival and reproductive success also depends upon cooperative interactions
with the tumor stroma, and the cooption of organismic adaptations. Below, we argue
that cancer progression can display multilevel selection dynamics.
Recently, Germain (this journal 2012) has argued that the population of cells in a
cancer is at best a ‘‘minimal’’ case of a Darwinian population. On his view, cancer
does not acquire sufficiently complex adaptations to count as paradigmatically
Darwinian, by the lights of Godfrey-Smiths’ account of Darwinian populations
(2009). We agree to an extent, but argue that Germain’s focus is too narrowly on the
individual cell. We argue that once one takes a multi-level perspective on cancer,
cancer is meaningfully described as both process and byproduct of selection. That
is, cancer progression involves cooption of adaptations at distinct levels of
organization in the biological hierarchy. We argue that this multilevel perspective
can shed light on both the dynamics of cancer progression, and differential
vulnerability to cancer within and between species and higher clades. Indeed, taking
an evolutionary perspective may assist in cancer treatment, such as demonstrating
when and why it is better to use combinations of drugs to treat cancer (Komarova
and Wodarz 2005). Our argument has two parts. In the section titled ‘‘Multi-level
selection’’, we briefly explain and describe the multilevel perspective. In the section
titled ‘‘Applying the models’’, we apply this perspective to cancer itself. In the
section titled ‘‘Adaptation and levels’’, we explain how our view departs from
Germain.
Multi-level selection
There is an extensive cannon of interdisciplinary work on multilevel selection
(Lewontin 1970; Wilson 1975; Damuth and Heisler 1988; Michod 1997; Sober and
Wilson 1999; Okasha 2005,2006; Godfrey-Smith 2009). To understand tumor
progression, we will draw on a few key ideas in this literature. First, multilevel
selection theory starts with the relatively uncontroversial observation that popula-
tions of entities are nested within one another in the biological hierarchy:
populations of genes are nested within cells, which are nested within organisms, and
organisms are nested within populations, which are nested within species and so on
to whole lineages. Provided that these entities vary, the variation is heritable, and
makes a difference to relative fitness of variable entities, selection can act at more
than one level in this hierarchy. For, in principle, any population of entities with
heritable variation in fitness may evolve by natural selection (Lewontin 1970). In a
‘‘multilevel’’ selection situation, selection is acting on more than one level
simultaneously. Selection at one level may increase or decrease the frequency of
traits in a population, which in turn, may affect what is available to selection at a
lower or higher level of analysis.
In multilevel selection, it can become difficult to discriminate which outcomes
are due to selection at a ‘‘lower,’’ and which at ‘‘higher’’ level of analysis. Damuth
and Haisler introduced a useful distinction between multilevel selection 1 (MLS1)
and multilevel selection 2 (MLS2) to assist in discriminating between different
scenarios. As they put it, ‘‘of interest in the former case are the effects of group
The evolution of failure: explaining cancer as an…
123
Author's personal copy
membership on individual fitnesses, and in the latter the tendencies for the groups
themselves to go extinct or to found new groups (i.e., group fitnesses)’’ (Damuth and
Heisler 1988, p. 407). That is, in MLS1, fitnesses are properties of individuals, and
character values are attributed to individuals. But, group membership can have
effects on individual fitness. When this happens, one has a case of MLS1. Because
different groups have different mixes of individuals, and some mixes are more
successful than others (i.e., the individuals in some groups better survive and
reproduce than the same individuals in a different group with different proportions
of variants), we may describe outcomes of such group effects as a product of ‘‘group
selection’’ in the sense of MLS1. The outcome in such cases is simply changing
proportions of different kinds of individuals in the whole population (the
metapopulation). Damuth and Heisler explain that some have wished to call
MLS1 ‘‘group selection’’ because group level properties (such as the relative
proportion of altruists in a group) determine the relative success of these
individuals.
But there is another sense of ‘‘group’’ selection that is often confused with this
first sense: what Damuth and Heisler prefer to call multilevel selection 2, (MLS2).
Here ‘‘group selection’’ refers to change in the frequencies of different kinds of
groups, in light of ‘‘group level’’ properties. This sense of group selection is more
controversial, not least because groups are often diffuse in space and time, and their
properties also are difficult to define and measure. However, in MLS2, character
values are attributed to groups (including both aggregate and global characters), and
it is these group level characters that make a difference to group success at
propagation, or the seeding of more groups. Groups are more ‘‘fit’’ if and only if
they propagate more groups. To be clear, it is not a necessary condition of MLS2,
according to Damuth and Heisler, that group characters be ‘‘emergent,’’ in the
strong sense that there is no way to explain the properties of the group in terms of
the properties of the individuals. Average height or extent of genetic heterogeneity
in a population is, after all, a group level trait, but it can be explained in terms of the
properties of individuals that make up the group. All that MLS2 requires is that
these group level properties are what cause group success, (measured in this case by
the seeding or propagation of new groups, in the way that colonists succeed at
colonizing new islands) (Damuth and Heisler 1988). As a vivid but simple example,
consider two groups of organisms: one that is genetically uniform, and one with a
large amount of genetic variation. In the latter group, a greater extent of genetic
variation may better enable such groups to propagate and diversify in new
environmental niches. Insofar as this group level trait (extent of genetic
heterogeneity) was a cause of relative group success at propagation, this would
count as an instance of MLS2.
Multilevel selection theory has been used to explain the emergence of new levels
of organization in the biological hierarchy, sometimes called the ‘‘major transitions
in evolution.’’ (Okasha 2005,2006). The argument is as follows: the evolution of
multicellularity required the emergence of cooperative interactions between
populations of cells. Or, there were intermediate stages between functionally
integrated individuals, and mere collectives. This may have involved the cooption of
traits initially adaptive at the individual level in service of collective benefit.
C. Lean, A. Plutynski
123
Author's personal copy
Quorum sensing in bacteria may be a vivid example. Quorum sensing is a method
by which bacteria detect local cell density by diffusion and accumulation of
signaling molecules in the surrounding environment. At certain densities, the
population as a whole can engage in a coordinated response, e.g., the production of
bioluminescence, population mobility, biofilm maturation, and virulence (Williams
et al. 2007). That is, quorum sensing regulates production of ‘‘public goods,’’ or
products manufactured by an individual that can then be utilized by the individual or
its neighbors. The capacity to quorum sense likely evolved from ‘‘self-signaling’’
systems that were adaptive at the individual level. (Redfield 2002; Williams et al.
2007). There are many such cases of proto-organismic collectives in nature. Or, as
Queller and Strassman put it, there are:
…levels of organism, and each level was attained by merging formerly
separate individuals at a lower level…. Multi-cellular individuals are
cooperative groups of cells, eukaryotic cells are cooperative assemblages of
multiple prokaryotic lineages and prokaryotic cells must have emerged by
assembly of formerly independent replicators. These major transitions in
evolution construct new levels of organism out of separate individuals
(Queller and Strassman 2009, p. 3143).
How do these transitions come about? Multilevel selection theory figures in one
prominent account: During the major transitions in evolution, e.g., the transition to
multicellularity, the fitness of the collective must override the fitness interests of the
parts. That is, collectives or groups must become more successful as collectives; i.e.,
they have to outcompete other collectives in reproduction, or, in propagation of
further collectives. Cooperation and functional integration of parts can play a role in
the evolution of such a collective. Cooption, or cross-level byproducts, might also
play a role in the emergence of functional integration. As in the case of quorum
sensing, traits that originally may have enhanced individual fitness can be coopted
in service of collective fitness, or the relative success of groups at propagation.
Some have argued that what might begin as a ‘‘group selection’’ process in the sense
of MLS1 (individuals benefitting due to group membership) could evolve into a
circumstance where we have a case of MLS2 (groups being more or less successful
at propagation, due to group level properties) (Michod and Herron 2006; Okasha
2006). That is, the major transitions in evolution may have involved successive
changes in population structure and organization. Individuals gradually organized
into collectives, benefited from their membership in some collectives, and
eventually, came to succeed or fail as collectives (Okasha 2005). If particle fitness
is improved by being part of a particular collective, then we can speak of the
particles, which gain mutual advantage within the collective, as having ‘‘aligned’’
fitness. Aligned fitness allows for the emergence of complex integration and
interaction within the collective. Integration and particle interaction can lead to
collectives that are functionally organized, so that the collective fitness of the group
is distinct from that of its members. In this way, collectives with division of labor
are, as Queller and Strassman argue, akin to organisms, to a greater or lesser degree.
We will argue that something approaching this process may occur in advanced
tumors.
The evolution of failure: explaining cancer as an…
123
Author's personal copy
To summarize: in principle, selection can act within a collective—among the
parts—and, between collectives. These processes could be going on simultaneously,
or sequentially. Damuth and Heisler (1988) describe two senses of multi-level
selection, ‘‘MLS 2 v. MLS 2’’ to highlight the fact that ‘‘group selection’’ may refer
to different things, and that the collective fitness of a group is not necessarily simply
a matter of averaging over particle fitnesses (See Table 1for summary). This is the
central point of critics of genic selection (see, e.g., Sober and Wilson 1999): simply
averaging fitness of different genotypes that live in distinct groups may result in a
loss of important causal information. As Wilson demonstrated (1979), group
membership can make a difference to relative fitness; for instance, altruism can
evolve in ‘‘trait groups’’—where population structure is not freely mixing, but
where similar individuals tend to find themselves together. Cooperation is less likely
to evolve in a freely mixing population, where ‘‘cheaters’’ or ‘‘free riders’’ can arise
and spread. In trait groups, the fitness of individuals is largely contingent on the
make up of their group. Nor is the fitness of a group simply the average fitness of its
members: the two can come apart and pull in different directions. Some groups may
propagate more successfully than other groups, even if the average fitness of the
group’s members is low (for discussion, see Sterelny et al. 2014).
Applying the models
In order to see why these models may be applied to cancer, there are some basic
facts about cancer initiation and progression that need to be made clear. Many
people think that cancer is a rare and unusual event; this is a natural assumption to
make, since we are made aware of cancer on the occasion of diagnosis of invasive
disease. But, invasive cancer—that is, what clinicians define as cancer—is the
endpoint of a long process; most early stages of this process fail to advance to an
invasive cancer (Sakr et al. 1993; Mori et al. 2002; Martincorena et al. 2015). Over
the course of the lifetime of the average individual, there are many thousands of cell
Table 1 Multilevel selection
Multilevel selection
(Damuth and Heisler
1988)
MLS1 MLS2
‘‘Group selection’’
refers to
Change in frequency of
individuals in light of group
membership
Change in frequencies of different kinds of
groups
Fitnesses are
properties of
Individuals Groups
Characters are
values attributed to
Individuals (e.g. altruism) Groups (e.g., group mean, population
density, proportion of different phenotypes)
Appropriate for
investigation of
Evolution of characters of
individuals likely affected by
group membership
Changing proportions of different types of
groups; different propensities to go extinct
or to found new groups
C. Lean, A. Plutynski
123
Author's personal copy
divisions, and randomly distributed somatic mutations occur and accumulate in
normal cells. Only a very small proportion of these cells and their descendants
succeed at advancing to cancer. What molecular biologists and modelers of cancer’s
evolutionary dynamics are hoping to predict and explain is which ones will succeed
at advancing to an invasive cancer, and why. Some investigators focus on the
particular features of the cells themselves; some focus on features of whole
populations of such cells; some focus on hierarchical or spatial relations between
cells; some focus on interactions between cancer cells and their microenvironment.
Whether a cancer progresses is likely due to all of the above. Our view is that by
taking a multilevel evolutionary perspective we can integrate a variety of levels of
analysis, or show how they are mutually constraining and mutually informative.
That is, in our view, there are several different targets or levels of selection at play
in cancer initiation and progression, some more salient at different stages of
progression.
Identifying a level of selection involves identifying a population of like entities,
which have reproduction, and differential fitness due to heritable features of such
entities (Godfrey-Smith 2009). According to the multilevel selection view, there are
several distinct units or targets of selection potentially of relevance in cancer
progression. These are not mutually exclusive options, but may be co-occurring:
1. Individual selection, mutation, and drift, acting among single cells, whether
‘‘pre-cancer’’ cells, cancer cells located in a tumor, or metastatic populations of
cells. Individual cells may be more or less fit in the sense of overcoming the
variety of constraints in normal tissue that ordinarily prevent disorderly cell
growth, acquiring resources in the local tissue microenvironment, or adapting to
novel tissue microenvironments (in the case of tumor metastases).
2. ‘‘Group’’ selection in the sense of MLS1 or MLS2: Some individuals may
survive and reproduce more successfully than others, because of membership in
a group (MLS1). Alternatively relative fitness may be ascribed to groups as a
whole (MLS2). Fitness in the latter case is assessed by the ability of the tumor
to produce new tumors, due to properties of the group—these may be either
aggregate properties (e.g., population size, extent of tumor heterogeneity), or
interactive properties (functional interactions between cells or lineages in a
tumor).
We may think of the process of cancer progression as proceeding through a series
of stages, where individual selection is more salient in the beginning, and gradually,
MLS1 and MLS2 become salient (See Fig. 1). This is, arguably, not unlike the
process some have hypothesized was at work in the emergence of multicellularity,
as discussed above (see, e.g., Okasha 2006). Most models of the evolutionary
dynamics of cancer are simple individual level selection models (Nowak 2006;
Frank and Nowak 2004; Michor et al. 2004), where the fitness advantage is
attributed to a single cell, due to a gain of function or loss of function, attributable to
a single mutation. Such mutations can yield advantages such as promoting
angiogenesis, blocking tumor suppression, or inhibitors of growth (halting or
blocking apoptosis, or cell death), or promoting cell division (e.g., gain of function
The evolution of failure: explaining cancer as an…
123
Author's personal copy
mutations). Growth in any environment will be limited by space and availability of
nutrients; among incipient cancer cells, some cells will be better able to compete for
these resources, or escape mechanisms of homeostatic control. Such evolutionary
dynamics are taken to feature centrally in early stages of the development of a
‘‘proto-’’ cancer cell lineage. Models of such circumstances look not unlike
competitive exclusion models in ecology (See Wodarz and Komarova 2014 for a
review).
There could also be selection operating at the level of cell colonies, however.
Komarova and Wodarz (2004, 2014) argue that there is an optimal ‘‘strategy’’ for
chromosomal instability (CIN) or rate of loss of heterozygosity in cancer. Cancer
cells characteristically have various chromosomal abnormalities—duplications,
losses, and inversions of large parts of the genome complement of a cell. One might
think chromosomal instability is uniformly bad, but chromosomal instability in
cancer is a ‘‘two edged sword’’: too much instability can lead to destructive genomic
changes, such as the loss or destruction of whole chromosomes. In such cases, cells
simply cannot replicate and will die off. But, if timed correctly, CIN can inactivate
tumor suppressor genes, for instance, and thus promote clonal expansion. Some
lineages will thus be more successful than others in progression to metastasis due to
the specific types of traits of cells, acquired at specific times or stages in the
progression of the disease. But, some collectives, or colonies of proto-cancer cells,
may more quickly reach a cancerous state, insofar as they have the optimal
‘‘strategy’’—or, the optimal growth rate given a certain degree of instability; too
much instability is detrimental because it increases the death rate.
2
Too little slows
down progress because the rate of onset of cancerous mutations is low. Of course,
Fig. 1 Transitions in levels of selection as cancer progresses
2
A reviewer asks whether the lineage of cells in a clonal population is a ‘‘population.’’ We grant that it is
an atypical population, in that the population is all descended from a clone. But, if clonal populations of
C. Lean, A. Plutynski
123
Author's personal copy
this is an idealized model, and populations of pre-cancerous cells are not a simple
series of clonal expansions, acquiring a specific mutation at each stage, and not
‘‘adjusting’’ their rate ‘‘strategically.’’ But, Komarova and Wodarz’s point is a
theoretical one: Early acquisition of chromosomal instability carries an advantage,
which, they argue, falls off over time, as too much genetic instability overall reduces
fitness. Lineages with the optimal strategy are the ones that ‘‘succeed’’ in
progressing to cancer. Evidence that there are many ‘‘potential’’ cancers comes from
samples of otherwise healthy tissue containing cancer mutations (Martincorena
et al. 2015).
Moreover, population architecture could well predict the evolution of faster or slower
progression to a cancer phenotype; that is, where most of the tumor is maintained by a
small number of tumor stem cells, the pattern of stem cell division can affect rate of
escape from feedback regulation, and thus also the acceleration of onset invasive cancer.
Arguably, such a case would count as an instance of both simple individual selection,
and MLS2. For, the time to the evolution of escape from feedback regulation depends
upon effective population size, which is a group level property (MLS2). The ‘‘cancer
stem cell’’ hypothesis holds that some cells in a tumor have ‘‘stem-like’’ properties. Such
cells are constantly renewing and thus more likely to both seed new tumors and perhaps
also survive chemotherapy (Jordan et al. 2006). While the cancer stem cellhypothesis is
somewhat controversial (see, e.g., Gupta et al. 2009;Nguyenetal.2010), it is certainly
true that some cells in a cancer appear to be better able to seed new tumors.
Critics of the Darwinian picture have argued that models of cancer evolution treat
cancer as a simple clonal expansion process, where the population of cells is relatively
homogeneous (see, e.g., Weinberg 2014, pp. 458–468). We now know that tumors are
heterogeneous populations. However, there is no reason in principle why population
structure in a tumor cannot be incorporated into an evolutionary model (indeed, it’s
been done already, see, e.g., Wodarz and Komarova 2014). One simply reduces the
effective population size of a tumor, where the stem cells are the only constantly
renewing cells. Moreover, one could view a tumor as several co-evolving lineages,
each competing for space and resources. Such co-evolving subpopulations—or the
coevolution of metapopulations—is common in nature, and has been studied in a
variety of species where lineages are subdivided into local niches and so may evolve
either local adaptations, or engage in competition or co-evolutionary dynamics.
Weinberg himself considers whether ‘‘distinct subpopulations of cells within a tumor
act symbiotically, each supporting the growth and survival of the other, rather than
directly competing’’ (p. 466). Such coevolving lineages would count as a case of a
higher level of selection, either in the sense of MLS 1 or 2; e.g., cells can benefit from
membership in a tumor with a neighboring cells or cell lineages that produce acidic
byproducts that enable metastasis (Gatenby 2006); or, heterogeneous tumors may be
more successful at acquiring resources, resisting attacks from the immune system, or
seeding metastases, which effectively counts as an instance of MLS2. Such different
scenarios simply require that we adopt a multilevel perspective.
Footnote 2 continued
bacteria such as Lenski’s E. coli are populations, we don’t see why clonal populations in a tumor would
not be.
The evolution of failure: explaining cancer as an…
123
Author's personal copy
That is, MLS2 becomes more relevant as we consider malignancy, or invasion
and metastasis. This process requires that the carcinoma grow to a stage where it can
acquire a novel phenotype (motility), and thus reproduce: metastasis is the
production of other tumors throughout the body. But this process is not simply the
release of a single cell into the blood or lymphatic system. ‘‘Seeding’’ of metastases
is an extended process that involves a complex series of steps: the breaching of the
basement membrane, intravasion, the invasion or dissemination of metastatic cells,
and the colonization of distant organ sites. This process requires the cooperation of
tumor cells and stroma; the cooption of the very same heterotypic interactions that
are ordinarily used in service of wound healing (see, e.g., Weinberg 2014,
pp. 587–607). Most attempts at seeding, or most incipient metastases fail; the
disseminated cells more often than not are destroyed in circulation. The ability of
micrometastasis to colonize remote sites is not simply a given, in other words, but
requires many attempts, akin to the colonization of remote islands by Darwin’s
finches. Some metastatic populations are more effective at colonization than others,
and once a metastasis is established, these colonies, in turn, can be more or less
successful at disseminating secondary metastases. The secondary wave of metas-
tasis is sometimes called a ‘‘metastatic shower’’ because they send off many
‘‘showers’’ of potential colonists (cf. Weinberg 2014, p. 654).
The success of metastases is not due exclusively to properties of metastatic cells.
Metastasis involves the cooperation of the stromal tissue; activation of the EMT (the
epithelial mesenchymal transition) in carcinoma requires signals from the tissue
microenvironment; and the tumor both recruits stromal cells (macrophages, mast
cells, fibroblasts) and coopts signals ordinarily associated with wound healing to
break down surrounding tissue and metastasize. That is, the entire process of
metastasis is not achieved by single cells in isolation, but the activity of both the
tumor and tumor-associated cells in stimulating angiogenesis, breaking down
surrounding tissue, and enabling intravasation. In other words, while it is single cells
that eventually establish metastases, the process of metastasis is enabled by and
indeed functionally requires the interaction of many cells in the tumor and
surrounding tissue. We might think about metastasis as rather like ‘‘r’’ selection
rather than ‘‘k’’ selection; given that the environment is unstable, and most
metastases fail, tumors more successful at sending out large populations of multiple
metastases are more likely to be successful.
3
More heterogeneous primary tumors –
in the sense of both more genetically diverse, and epigenetically plastic and modular
– may be able to generate more and more successful metastases (Weigelt et al. 2003;
Klein et al. 2002).
Evolutionary dynamics also apply outside the primary tumor. Metastatic
populations continue to evolve in their new environments, and may be more or
less successful at colonizing, and generating successful secondary colonists (or,
secondary metastases), in light of the properties inherited from the primary tumor
(Psaila and Lyden 2009). The various properties of the primary tumor can be
3
A ‘r’ selection population is one with a high growth rate, seeding many individuals. ‘R’ selection is
often associated with populations which gain from dispersing over hostile environments. In contrast ‘k’
selection population have low growth rates due to the population living in an environment near carrying
capacity. See Macarthur and Wilson (1967).
C. Lean, A. Plutynski
123
Author's personal copy
maintained in secondary tumors. Evidence for fidelity in reproduction include
micro-assays of breast cancer showing that gene expression is maintained across
metastatic sites (Weigelt et al. 2003). There is in fact evidence that the ability to
colonize itself evolves (see Klein et al. 2002). Differential success is due in part to
group level properties of the primary tumor, namely, genetic heterogeneity and
functional diversity. Different cell lines in a tumor acquire differential capacities
that combined lead to more successful reproduction of secondary metastases at
different sites. All in all, provided that the fitness-increasing properties of the group
are reproduced in the novel site, we have a case of MLS2. If multiple metastases
arise from the primary tumor, and the tumor is heterogeneous, then it is possible for
the heterogeneity of the primary tumor to be found likewise in the metastatic sites.
Indeed, we urge further work on this empirically, but we believe that the evidence
already suggests that looking outside of cell-intrinsic properties to understand the
evolutionary dynamics of metastasis is essential.
4
Thus, depending on the level of analysis and the empirical details of any
particular case, cancer progression could be viewed as an individual level selection
process, or a group level selection process, in the sense of MLS1 or MLS2. That is, a
cancer cell could be competing with, or co-evolving with other cell lineages, or
incipient cancers or proto-cancer cell populations might be competing with other
proto-cancers for space, and resources. Or, differential reproduction via differential
success at invasion and metastases could be due to complex properties of the tumor
as a whole, including functional differentiation within a tumor, and cooption of
organismic adaptations in service of invasion and metastasis. That is, if invasion and
metastasis requires the acquisition of several capacities, such as the capacity to
coopt certain signaling systems in the microenvironment, different cell lineages
within a tumor might acquire these different capacities over time, and together
cooperatively produce the cancer phenotype. Komrova and Wodarz make the point
quite simply:
Assume that tumor progression requires the presence of three gene products,
call them A, B and C. A cell can evolve to accumulate all three mutations, thus
4
A reviewer comments: ‘‘There is no denying that the success of metastasis depends on many evolved
features and interactions (including cells other than the colonizing cells) and, as the author points out, that
features such as heterogeneity are tumor-level property…However, in order to show that tumours have
complex adaptations, it is insufficient to show that tumours have developed complex traits, or that these
promote progression, and one must also show that these traits were selected at this level, rather than for
the benefit of single cells.’’ This is an excellent comment and central concern worth addressing. In our
view, a trait that promotes relative survival and reproductive success of an entity at some level need not
be (initially) selected for survival and reproductive success at that particular level. As long as it
(currently) promotes survival and reproductive success of entities at that level, the trait is selected for at
that level. We believe that we have shown this to be the case, and, the following references provide
further elaboration of examples (Gatenby et al. 2007; Gavert and Ben-Ze’ev 2010; Egeblad et al.
2010; Mueller and Fuesnig 2004). Ruling out selection on a given level by arguing that a trait was not
initially selected for to benefit that level would rule out all selection on all coopted traits, which would
(potentially) rule out a good deal of selection even at the level of individual cells or organisms. See, e.g.,
Lloyd (2012) for discussion.
The evolution of failure: explaining cancer as an…
123
Author's personal copy
achieving the malignant phenotype. On the other hand, the three gene products
can arise independently in different cells, and they can share the gene product
as a public good. In this setting, the malignant phenotype arises as an emergent
property from a collection of cooperating cells rather than an individual cell
(Wodarz and Komarova 2014, p. 460–1).
Put simply, on the MLS2 view, collectives are more or less successful in light of
group level properties: heterogeneity, or division of labor in the process of invasion
and metastasis. In the latter case, the collective of cell lineages in the tumor in some
sense has ‘‘aligned’’ fitnesses. Reproductive ‘‘success’’ in such cases is measured in
terms of invasion and propagation of metastases (recall: most incipient cancers fail
to progress); or, metastases could be more or less successful at generating secondary
metastases. In MLS2, we are interested in the way group level traits affect group
level fitness. Thus, cell lineages (including cancer and non-cancer stromal cells)
might interact as a collective (both with other cancer cells and with the tumor
stroma) thus increasing the collective’s ability to survive and reproduce, relative to
other collectives (which might include other precancerous lesions, or other
metastases, at seeding further ‘‘secondary’’ metastases). If better functional
integration in a tumor results in more successful metastasis, then the collective
fitness of the whole is distinct from the aggregative fitness of the individual cells.
There is now ample evidence that tumors are complex heterogeneous populations of
cells, and that signaling between these cells plays an essential role in tumor
progression (Egeblad et al. 2010).
Cancer cells also recruit neighboring cells in order to participate in tumor
development, both a structural support, and in building a blood supply (Mueller and
Fuesnig 2004). Arguably, this is a form of niche construction in which the cancer
cells and normally functioning cellular populations both evolve and co-evolve with
their surrounding environments (Odling-Smee et al. 2003). That is, cancer
progression requires the cooperation of the tissue microenvironment; the behavior
of a tumor is not due exclusively to the behavior of cancer (or, neoplastic) cells
alone. Carcinoma cells release growth factors that recruit macrophages, neutrophils,
and lymphocytes, which in turn orchestrate an inflammatory response that involves
the release of signals that stimulate the proliferation of epithelial cells and the
process of angiogenesis in the tumor stroma. Epithelial cells release PDGF, for
which fibroblasts & myofibroblasts have receptors; these in turn release IGF-1,
which benefits growth and survival of nearby cancer cells. Stromal and epithelial
cells collaborate in construction of the extracellular matrix (ECM). Stromal cells are
recruited by tumor cells to aid in construction of tumor colonies. Metastatic cells
can spread throughout the body because they can acquire a ‘‘disguise’’ of platelets
that enable them to evade detection (Egeblad et al. 2010). In all these cases, cancer
cells are exploiting preexisting biological programs that are normally used by
tissues for other purposes, in particular, wound healing. Tumor cells are activating a
complex program that is already available to cancer cells, insofar as they evolved
from healthy cells with the very same capacities. Wound healing uses a number of
growth factors to increase permeability of the blood vessels near the wound, attract
fibrogen to create blood clots, and attract fibroblasts, convert them into
C. Lean, A. Plutynski
123
Author's personal copy
myofibroblast, which in turn release matrix metalloproteinases (MMPs) which play
a role in building up the extracellular matrix, or remodeling the ECM, carving out
spaces for new cells. Recruiting fibroblasts to the tumor allows the creation of a
fibrin matrix, which serves as scaffolding in the initial formation of a tumor. It’s
important to see that this is not in tension with natural selection; selection and niche
construction could be going on simultaneously. Those types more effective at
shaping their environment to suit their ends will outcompete their peers.
The cooperation of the tumor cells and stromal cells is not simply a cooption of
existing program of cells, however. The cancer cells and non-cancer cells may
coevolve; that is, cancer-associated stromal cells become more adept over time at
assisting in the survival of cancer cells within the tumor. Experimental evidence for
the importance of the stromal tissue in tumor progression comes from a variety of
sources, but we will discuss just one powerful case study. CAFs are cancer-
associated fibroblasts, very similar to fibroblasts in normal cellular tissue, which
play a role in building up scaffolding in injured tissue. An experiment to test the role
of CAFs mixed normal human prostate cancer cells in combination with different
kinds of fibroblastic cells, and then implanted them in host immunocompromised
Nude mice. When cancer cells were mixed with CAFs, they grew tumors 500 times
as large than when mixed with fibroblasts from normal prostate glands prior to
injection into hosts (Olumi et al. 1999). Weinberg argues that the cumulative
evidence of the role of CAFs in tumor formation suggests that ‘‘stromal cells in
advanced carcinomas, having coexisted with epithelial cancer cells for many years,
may change their genotype and acquire traits that genetically normal stromal cells
cannot achieve. This suggests that stromal cells co-evolve with their neoplastic
neighbors during long periods of tumor development.’’ (Weinberg 603) In other
words, a tumor (which is composed not only of cancer cells, but also of stromal
tissue) is a complex tissue with co-evolving cell lineages, composed not only of
cancer cells, but also non-cancer cells, which together cooperate in the progression
of a tumor to metastasis. Indeed, Egeblad et al. (2010) claim that the highly
integrative functional organization of some tumors suggests that they should be
considered as akin to organs.
It’s important to note, as Fisher pointed out, natural selection is not evolution
(Fisher 1930). While there are multiple levels of selection in operation that
potentially play an explanatory role in cancer, we also grant that there must be a role
for drift. There is no doubt that the population size of cells in a tumor as a whole is
sufficiently large for selection to overcome drift in the tumor as a whole; the volume
of a cubic centimeter within a tumor cell mass contains about 10
9
cells. However,
tumors—to say the least—have a complicated population structure. It is quite likely
that whether a cancer cell initially lives or dies may have to do as much with
accidents of location as particular fitness-enhancing features intrinsic to the cell.
The relative importance of selection versus drift will vary case by case; for instance,
there are likely many isolated sublines of cancer cells within a tumor that may or
may not survive, given lack of access to a blood supply, or architectural features of
the tissue microenvironment, rather than features intrinsic to the cancer cells or
populations of such cells (Crespi and Summers 2006). The threshold of .2 mm is the
maximum required for oxygen to effectively diffuse between tissues; so, tumors will
The evolution of failure: explaining cancer as an…
123
Author's personal copy
not grow if greater than .2 mm distance from blood vessels. This mimics the way
the fitness of normally functioning somatic cells is determined by their local cellular
signaling environment rather than their own intrinsic features. Further, where a
cancer is initiated may affect how slowly or quickly it progresses. Drift may well
explain a good deal of variation we find in whether and why tumors progress to
metastasis. And, if stem cell architecture affects tumor growth, such that only
certain subpopulations are driving growth, drift will play a significant role.
The relevant grain of analyses and thus level of selection which explains any
given phenomena of interest will be in part dictated by the questions we ask. If we
are interested primarily in which cells come to predominate in a tumor, then a
selection process within the population of cancer cells itself is a better target.
However, if we are interested in using the body’s own resources to fight cancer (e.g.,
via the development of immune therapies), perhaps we may be interested in viewing
the tumor and stromal tissue as a whole (and its collective fitness enhancing
properties) as a subject of analysis. Thus, there is a pragmatic dimension to the
question: where is selection acting? But, there is also an empirical fact of the matter
that constrains whether and to what extent selection (or, perhaps drift) is genuinely
responsible for a given outcome of interest.
In sum, this multi-level perspective provides a representation of cancer
progression that is both predictive and explanatory. The accumulation of mutations
(and epigenetic changes) in neoplastic progression involve heritable variation in
fitness of cancer cells and lineages, e.g., timing and acquisition of particular
mutation types, chromosomal instability, hypoxia, acid resistance, transition to
mesenchymal phenotype. Cancer progression involves several transitions in
individuality and levels of selection, and, is a unique case of multi-level selection:
it evolves at the level of populations of cells, competition among cell lineages, via
cooption, a kind of cross-level exaptation, and as a group, or higher level
‘‘individual.’’
Adaptation and levels
Recently it has been claimed that somatic cell evolution by natural selection is not
particularly relevant to understanding cancer (Germain 2012). Germain claims that
while cancer meets the minimal requirements for natural selection, as described in
Godfrey-Smith’s Darwinian Populations, the main cases described as adaptations
via a natural selection are not ‘‘complex’’ adaptations. Given that natural selection
cannot explain the formation of such adaptations, he claims that it lacks explanatory
power.
A central presupposition in Germain’s argument is that he holds that for any
feature to be an adaptation it must be the product of a gradual accumulation of
iterative changes to the phenotypes. Each case Germain discusses is the product of a
single large change. Senescence and apoptosis avoidance are products of mutations
to TP53; immune evasion is gained through the loss of Human Leukocyte Antigen
(HLA); and resistance to therapy is described as the loss of the expression of a
targeting intermediate for the attachment of drugs or the activation molecular means
C. Lean, A. Plutynski
123
Author's personal copy
for resisting drug uptake. In all these cases, Germain argues that the adaptation in
question is due to a single mutation, not the product of cumulative change, and
therefore not meaningfully described as complex adaptations.
Whether or not a trait is due to a single mutation of large effect or many
mutations of small effect is irrelevant to whether it is in fact an adaptation. As
Turner (1977) showed, mimicry in butterflies (a significant adaptive trait) can in
some cases be due to major mutations with large effects on fitness, followed by
mutations of small effect. But, there is a larger point to be considered here. Whether
a trait is an adaptation, and for what, will depend on the grain of analysis or entity
that the phenotype is predicated of. If selection only occurs in populations of cancer
cells, and adaptations only to the single cell, affecting only its immediate
descendants, then a Darwinian perspective would indeed not be very informative.
However, if we ‘‘scale up’’ or ‘‘zoom out’’ on the population over a longer time
scale, it becomes less implausible to see these adaptations as cumulative changes. If
what we are interested in is the production of the cancerous phenotype as described
by something similar to ‘‘The Hallmarks of Cancer’’ then a natural selection
framework is very informative
5
. While some of the genes described by Germain are
definitely important for a cancer cell’s survival, they only can be treated as part of a
‘‘cumulative’’ adaptation insofar as collectively they contribute to a tumor’s growth,
persistence and metastasis.
Many of the cancer adaptations only appear to make sense at the tumor level, or
at the level of collective fitness, where the collective includes the stromal tissue.
There is evidence of cooperative coevolution between both cancer and stromal cells,
each contributing to cycles of signals that promote growth, reshape the extracellular
matrix, and up regulation of growth factors to produce angiogenesis, as well as
promote invasion and metastasis. There is an incredible wealth of such positive
feedback loops between carcinoma cells and what are called ‘‘Tumor-associated
Macrophages,’’ ‘‘Cancer-Associated Fibroblasts,’’ and other cells in the reactive
stroma. Acquisition of malignancy is not, in other words, a product of a series of
mutations alone, but of the collaboration of both mutant alleles in cancer cell
genomes and the signals that these cells send and receive from the tumor
microenvironment (Karamysheva 2008). If we analyse a particular mutation of an
individual cell that up regulates any of these growth factors we would not be able to
say there was adaptation. However, the impact of multiple mutations can be
cumulative and amplified by positive feedback, yielding, e.g., the angiogenic
switch. That is, angiogenesis is product of the aggregation of multiple cell line
mutations, and interaction between carcinoma and stroma. This is a stepwise
accumulation of changes in the population that produces an adaptive phenotype for
the tumor as a whole.
The key is to see this as a multilevel selective process, with selection operating at
distinct levels simultaneously. There are enough features in common between the
production of the cancerous phenotype and the production of other phenotypes in
5
In Hanahan and Weinberg (2000) they claim cancer is a population of cells with the following features
(1) Self stimulation of growth (2) resist inhibitory signals (3) apoptosis avoidance (4) angiogenesis (5)
unlimited replication (6) metastasis.
The evolution of failure: explaining cancer as an…
123
Author's personal copy
populations to apply natural selection style models to cancer dynamics. This is not
to say that there are not important differences. Cancer is a by-product of the
adaptation and evolution of complex organisms. Thus, many (most) adaptations of
complex tumors are themselves byproducts of our evolutionary history.
Why are metazoans so susceptible to cancer?
We have described cancer as process in which somatic cell evolve to have features
of a Darwinian population. Cancer progression involves transitions in the levels of
selection similar to the major transitions of evolution. Of course, major transitions
take hundreds of thousands of years, but cancer progression from the first progenitor
cells usually takes decades, though it can occur much more quickly. How is it that
metazoans are particularly vulnerable to cancer? We have argued that this is in part
due to cancer cells’ cooption of organismic adaptations; here we claim additionally
that cancer is a byproduct not only of organismic adaptations, but features
distinctive to the metazoan clade (though, see, e.g., Aktipis et al. 2015 for a
discussion of ‘‘cancer like’’ phenomena in plants).
The fact that many parts of our body (from organs, to cells, to organelles, to
features of the genome themselves) are modular and thus quasi-independent, or can
evolve or change over time without affecting the fitness of other parts, has made it
possible for complex adaptations like the vertebrate limb to evolve. It also makes a
particular type of defection possible. The very features that enable evolvability
(quasi-independence, plasticity), also permit a tumor to arise or remain undetected.
Thus, that cancer happens is in part a product of our being a highly evolvable
lineage; in other words, it is a product of our fitness from a clade level perspective.
The plasticity that enables metazoan cells to pass from epithelial to mesenchymal
and then from mesenchymal to epithelial phenotypes, is what makes metastasis
possible in carcinoma. While plants can suffer from uncontrolled cellular growths,
the lack of an EMT, and thus cellular motility in plants prevents the possibility of
metastatic tumors.
There is a history of selection for a number of traits that are recruited or deployed
to promote tumor growth. Many traits that contribute to organismic fitness (or,
fitness at the level of the whole organism) also are ‘‘coopted’’ by cancer cells
(Crespi and Summers 2006). For instance, cellular plasticity, extensive proliferative
capacity, mobility, and availability of lymphatic and vascular channels for cell
migration all contribute to embryogenesis, wound healing, tissue renewal, and
placental implantation. They also contribute to cancer initiation and progression. In
other words, cancer is both product and byproduct of a multi-level selection process:
the process of selection among cells within whole organisms and, a by-product of
selection between individual organisms.
The way cancer uses local micro-environment can be understood in reference to
the adaptations the normally functioning cell lineages have, in light of selection at
the level of the organism as a whole. Metastatic cells acquire their mesenchymal
phenotype because they both send and receive signals to surrounding tissue that
permit the breakdown of the extracellular matrix, and re-activate the very same
C. Lean, A. Plutynski
123
Author's personal copy
‘‘protocols’’ involved in the EMT (epithelial-mesenchymal transition) in develop-
ment and wound healing. EMT is the process by which epithelial cells lose their
polarity and cell–cell adhesion allowing them to be mobile. Cancer also coopts the
signaling networks it is a part of; and, it ‘‘re-ignites’’ the very same internal
programs that make possible wound healing and development, over the course of its
natural history. In other words, subversion in cancer is only made possible by taking
advantage of those very mechanisms by which subversion is (generally speaking)
prevented: those very same signaling routes that ordinarily foster or further
cooperative activity.
The transitions cancer goes through as it goes metastatic mirror the major
transitions of evolution. The same features that make multicellularity possible
operate in cancer progression to metastasis: the development of shared fitness
interests; genetic bottlenecks (with both stem cells, and metastasis); division of
labor (variable fitness of cancer cells in different tissue microenvironments); export
of fitness (to the tumor as a whole); and rarely, perhaps, functional integration. Just
as the multilevel selection framework can help explain the major transitions it can
also help explain the major transitions in cancer phenotype.
Conclusions
We have argued that cancer is a multi-level evolutionary process. The accumulation
of mutations (and epigenetic changes) in neoplastic progression can be understood
as an evolutionary process; both individual cells and lineages in a cancer may be
more or less successful in progressing to metastasis, depending upon features like
the timing and acquisition of particular mutation types, or chromosomal instability.
Cancer also coopts signaling systems that evolved in service of higher levels of
organization, systems that ordinarily regulate wound healing, development and
immune response. At advanced stages, the combined tumor and stroma can be
viewed as an integrated entity, i.e., a collective with organization and fitness as a
whole (MLS2 as opposed to MLS1). Cancer is an exemplary case multi-level
selection process: it evolves at the level of populations of cells, competition among
cell lineages, via cooption or a kind of cross-level expatiation, and as a group.
References
Aktipis CA, Boddy AM, Jansen G, Hibner U, Hochberg ME, Maley CC, Wilkinson GS (2015) Cancer
across the tree of life: cooperation and cheating in multicellularity. Phil Trans R Soc B
370(1673):20140219
Anderson A, Weaver A, Cummings PT, Quaranta V (2006) Tumor morphology and phenotypic evolution
driven by selective pressure from the microenvironment. Cell 127(5):905–915
Bissell M, Hines WC (2011) Why don’t we get more cancer? A proposed role of the microenvironment in
restraining cancer progression. Nat Med 17(3):320–329
Cairns J (1975) Mutation selection and the natural history of cancer. Nature 255:197–200
Cairns J (1978) Cancer science and society. Freeman and Co, San Francisco
Crespi B, Summers K (2006) Evolutionary biology of cancer. Trends Ecol Evol 20(10):545–551
The evolution of failure: explaining cancer as an…
123
Author's personal copy
Damuth J, Heisler IL (1988) Alternative formulations of multilevel selection. Biol Phil 3(4):407–430
Egeblad M, Nakasone ES, Werb Z (2010) ‘‘Tumors as organs: complex tissues that interface with the
entire organism’’. Dev Cell 18:884–901
Fisher R (1930) The genetical theory of natural selection. Clarendon, Oxford
Frank SA (2007) Dynamics of cancer: incidence, inheritance and evolution. Princeton, New Jersey
Frank SA, Nowak MA (2004) Problems of somatic mutation and cancer. BioEssays 26(3):291–299
Gatenby RA et al (2006) Acid-mediated tumor invasion: a multidisciplinary study. Cancer Res
66:5216–5223
Gatenby RA et al (2007) Cellular adaptations to hypoxia and acidosis during somatic evolution of breast
cancer. Br J Cancer 97(5):646–653
Gause GF (1966) Aspects of antibiotic research. Chem Ind 36:1506–1513
Gavert N, Ben-Ze’ev A (2010) Coordinating changes in cell adhesion and phenotype during EMT-like
processes in cancer. F1000 Biol Rep 8(2):86
Gerlinger M, Rowan AJ, Horswell S, Larkin J, Endesfelder D, Gronroos E, Swanton C (2012) Intratumor
heterogeneity and branched evolution revealed by multiregion sequencing. New Eng J Med
366(10):883–892
Germain PL (2012) Cancer cells and adaptive explanations. Biol Phil 27(6):785–810
Godfrey-Smith P (2009) Darwinian populations and natural selection, vol 22. Oxford University Press,
Oxford
Greaves M, Maley CC (2012) Clonal evolution in cancer. Nature 481(7381):306–313
Gupta PB, Chaffer CL, Weinberg RA (2009) Cancer stem cells: mirage or reality? Nat Med
15(9):1010–1012
Hanahan D, Weinberg RA (2000) The hallmarks of cancer. Cell 100(1):57–70
Hausman DM (2012) Health, naturalism, and functional efficiency. Phil Sci 79(4):519–541
Jordan CT, Guzman ML, Noble M (2006) Cancer stem cells. New Eng J Med 355(12):1253–1261
Karamysheva AF (2008) Mechanisms of angiogenesis. Biochemistry (Moscow) 73(7):751–762
Klein CA, Blankenstein TJ, Schmidt-Kittler O, Petronio M, Polzer B, Stoecklein NH, Riethmu
¨ller G
(2002) Genetic heterogeneity of single disseminated tumour cells in minimal residual cancer. Lancet
360(9334):683–689
Komarova NL, Wodarz D (2004) The optimal rate of chromosome loss for the inactivation of tumor
suppressor genes in cancer. Proc Natl Acad Sci USA 101(18):7017–7021
Komarova NL, Wodarz D (2005) Drug resistance in cancer: principles of emergence and prevention. Proc
Natl Acad Sci 102(7):9714–9719
Lewontin RC (1970) The units of selection. Ann Rev Ecol Syst 1:1–18
Lloyd E (2012) Units and levels of selection. In: Edward NZ (ed) The stanford encyclopedia of
philosophy. URL =\http://plato.stanford.edu/archives/win2012/entries/selection-units/[
MacArthur RH, Wilson EO (1967) The theory of island biogeography. Princeton University Press, New
Jersey
Martincorena I et al (2015) High burden and pervasive positive selection of somatic mutations in normal
human skin. Science 348(6237):880–886
Merlo LMF, Pepper JW, Reid BJ, Maley CC (2006) Cancer as an evolutionary and ecological process.
Nat Rev Cancer 6(12):924–935
Michod RE (1997) Evolution of the individual. Am Nat 150(S1):S5–S21
Michod RE, Herron MD (2006) Cooperation and conflict during evolutionary transitions in individuality.
J Evol Biol 19:1406–1409
Michor F, Iwasa Y, Nowak MA (2004) Dynamics of cancer progression. Nat Rev Cancer 4(3):197–205
Morange M (2012) What history tells us XXVIII. What is really new in the current evolutionary theory of
cancer? J Biosci 37(4):609–612
Mori H et al (2002) Chromosome translocations and covert leukemic clones are generated during normal
fetal development. Proc Natl Acad Sci 99(12):8242–8247
Mueller M, Fuesnig NE (2004) ‘‘Friends or foes – bipolar effects of the tumor stroma in cancer. Nat Rev
Cancer 4:839–849
Navin N, Kendall J, Troge J, Andrews P, Rodgers L, McIndoo J, Wigler M (2011) Tumour evolution
inferred by single-cell sequencing. Nature 472(7341):90–94
Nguyen LV, Vanner R, Dirks P, Eaves CJ (2010) Cancer stem cells: an evolving concept. Nat Rev Cancer
12(2):220–228
Nowak MA (2006) Evolutionary dynamics: exploring the equations of life. Harvard University Press,
Massachusetts
C. Lean, A. Plutynski
123
Author's personal copy
Nowell PC (1976) The clonal evolution of tumor cell populations. Science 194:23–28
Odling-Smee FJ, Laland K, Feldman M (2003) Niche construction: a neglected process in evolution.
Princeton University Press, New Jersey
Okasha S (2005) Multi-level selection and the major transitions in evolution. Phil Sci Proc 72:1013–1025
Okasha S (2006) Evolution and the levels of selection. Oxford University Press, Oxford
Olumi AF, Grossfeld GD, Hayward SW, Carroll PR, Tlsty TD, Cunha GR (1999) Carcinoma-associated
fibroblasts direct tumor progression of initiated human prostatic epithelium. Cancer Res
59(19):5002–5011
Pepper JW, Sprouffske K, Maley CC (2007) Animal cell differentiation patterns suppress somatic
evolution. PLoS Comput Biol 3(12):e250
Psaila B, Lyden D (2009) The metastatic niche: adapting the foreign soil. Nat Rev Cancer 9(4):285–293
Queller D, Strassman J (2009) Beyond society: the evolution of organismality. Phil Trans R Soc B
364:3143–3155
Redfield RJ (2002) Is quorum sensing a side effect of diffusion sensing? Trends Microbiol 10:365–370
Sakr WA et al (1993) The frequency of carcinoma and intraepithelial neoplasia of the prostate in young
male patients. J Urol 150(2 Pt 1):379–385
Sober E, Wilson DS (1999) Unto others: the evolution and psychology of unselfish behaviour. Harvard
University Press, Massachusetts
Sterelny KR, Joyce B, Calcott B, Fraser B (2014) Cooperation and its evolution. MIT Press,
Massachusetts
Turner JR (1977) Butterfly mimicry: the genetical evolution of an adaptation. Evol Biol 10:163–206
Weigelt B, Glas AM, Wessels LF, Witteveen AT, Peterse JL, van’t Veer LJ (2003) Gene expression
profiles of primary breast tumors maintained in distant metastases. Proc Natl Acad Sci
100(26):15901–15905
Weinberg RA (2014) The biology of cancer, 2nd (edn.). Garland Science, NY
West S, Diggle SP, Buckling A, Gardner A, Griffin AS (2007) The social lives of microbes. Ann Rev
Ecol Evol Syst 38:53–77
Williams P, Winzer K, Chan W, C
´amara M (2007) Look who’s talking: communication and quorum
sensing in the bacterial world. Philos Trans R Soc Lond, Ser B 362(1483):1119–1134
Wilson DS (1975) A theory of group selection. Proc Natl Acad Sci 72(1):143–146
Wodarz D, Komarova NL (2014) Dynamics of cancer: mathematical foundations of oncology. World
Scientific Publishing Co, Singapore
Yachida S et al (2010) Distant metastasis occurs late during the genetic evolution of pancreatic cancer.
Nature 467:1114
The evolution of failure: explaining cancer as an…
123
Author's personal copy
A preview of this full-text is provided by Springer Nature.
Content available from Biology & Philosophy
This content is subject to copyright. Terms and conditions apply.
