Note to POBAM 2016 participants: Please do not circulate this material without permission. This paper is an early draft of what (I hope) will be a chapter in my PhD dissertation (expected defense: 2018).

Genealogy and Evolvability Celso A. A. Neto†

Abstract Recent debates about evolvability have taken insufficient account of genealogy. In this paper, I demonstrate the influence of genealogical features on evolvability via a discussion of laboratory work on Escherichia coli bacteria. Recognizing genealogy’s effect on evolvability allows us to overcome two limitations in recent debates. First, development has been overemphasized: we have failed to recognize other factors influencing evolvability. Second, those debates fail to appreciate different roles of lineages in evolvability. Overcoming these limitations broadens our understanding of evolvability's epistemic roles.

†

PhD Candidate at the Institute of Philosophy, Leibniz University Hannover. Address: Im Moore 21, 30167, Hannover, Germany. E-mail:[email protected]

1

1. Introduction Actual evolutionary histories suggests something important about their likelihoods. Given what we know about the evolutionary groups that gave rise to them, some evolutionary patterns and outcomes are likely. Such evolutionary groups seem to have the disposition or propensity to evolve those patterns, reaching certain areas of the morphospace more easily than others1. Biologists have called this propensity to occupy morphospace ‘evolvability’(Dawkins 2003). This concept broadly refers to the disposition of populations to evolve. Despite agreement on such a broad definition, biologists disagree on the nature of evolvability. For example, when discussing the causes of this disposition, most biologists focus on developmental processes and mechanisms within organisms (Wagner & Altenberg 1996; Kirschner & Gehard 1998) 2. This focus has been recently counterbalanced by some authors who emphasize the influence of environment and population-level structures (Sterelny 2007). The recognition of these influences broadens our understanding of evolvability and its explanatory power in evolutionary biology (Brown 2014). For this reason, it is worth analyzing other types of influence over evolvability. In this paper, I discuss a type of cause of evolvability that has been neglected in 1 2

A morphospace or “morphological space” can be understood as a multi-dimensional representation of all possible morphologies of organisms. Maybe “causes” is not the best term to denote the underlying features of a disposition, if these features and the disposition are synchronic. Here I use “causes” to denote “dependency relations” among such features and dispositions. Hence, by saying that there are different causes of evolvability I am merely saying that evolvability (a disposition defined as objective probability, as indicated in section 2) depend on different aspects (traits, properties, mechanisms, etc.) of organisms and populations. Some features are developmental in nature, but others are not.

2

the literature so far, namely genealogical features, i.e., features directly responsible for the formation, organization and maintenance of ancestor-descendant relationships in space and time. As I will argue, the recognition of such causes has significant implications for our understanding of evolvability, illuminating its role as both explanans and explanandum in evolutionary biology. In what follows, I introduce ‘evolvability’ via a discussion of Rachael Brown's work (2014) (section 2). Then, I present some experiments on the evolution of Escherichia coli bacteria (Woods et al. 2011) (section 3). These experiments show how the evolvability of E. coli is affected by ‘path dependence’, which I take to be a genealogical feature. I will explain how this has implications for two limitations in the current debates. The first limitation corresponds to a focus on developmental causes of evolvability, whereas many other causes remain to be explored. The discussion of genealogical features broadens our knowledge about the ways in which evolvability is causally grounded (section 4). The second limitation is that as genealogical entities lineages were recently taken as mere phylogenetic manifestations (effects) of evolvability (Brigandt 2007). Nonetheless, the experiments with E. coli suggest that the importance of lineages goes beyond its role as phylogenetic manifestations of evolvability (section 5).

3

2. What Evolvability Really Is

The characterization of evolvability as the disposition of populations to evolve is very broad. It is important to refine this characterization to understand it’s explanatory roles in scientific practice. With this aim in mind, Rachael Brown has recently proposed the following analysis of evolvability:

E: Pr x, b (F t)

Here evolvability is represented as the objective probability (Pr) of a particular trait, set of traits or trend (F), to evolve up to a particular future time (t), given the internal features of a population (x) in a particular environment (b). These variables are supposed to make clear what is at stake when biologists discuss evolvability. First, evolvability is always relative to a particular evolutionary outcome. There is no such a thing as “the disposition to evolve”, but only “a disposition to evolve in a certain way”. Second, evolvability is always relative to a temporal span. For example, the probability of a certain population of bivalves to evolve eyes after 10,000 generations is considerably different from its probability to evolve eyes after 200,000 generations. Third, the variable x encompasses not only causal features of organisms (e.g., developmental ones) but also causal features of the population they belong to (e.g., population-structure). Finally, environmental factors also affect evolvability.

4

Brown's representation is interesting in various ways. First of all, this representation makes clear what sort of explanation evolvability can provide. Evolvability contributes to the explanation of a certain evolutionary outcome by referring to its likelihood (or objective probability) given a set of parameters or “initial conditions”. In particular, the parameters x and b are placeholders for causal or underlying features which are present in a population at a certain time. These features ground the likelihood of that population evolving a certain F during a time span t. Hence, evolvability explains a future evolutionary outcome by considering the disposition that a population has (in a present or past time-slice) to evolve that outcome. For my current purposes, Brown's representation serves as a basis from which we can identify and analyze the limitations of current debates about evolvability. For instance, I noted before that many authors emphasize certain types of causes which can figure in x, such as developmental and populational features, whereas others point out the relevance of b – environmental factors (Love 2003, 1022). An interesting question is whether and how other types of causes affect evolvability. Another question is why the parameter t receives so little attention in the literature. Brown seems aware of these questions. For instance, she recognizes that the relation between time and evolvability deserves a closer look (2014, 18). In the next sections, I discuss a type of cause of evolvability which involve time span. I call such causes “genealogical features” since they have to do with the formation, organization and maintenance of ancestor-descendant relationships across

5

generations. These relationships matter to evolvability because they are responsible for transmitting variations, carrying the organisms through different regions of morphospace. Nonetheless, my point here is that such relationships also matter for another (mostly unrecognized) reason, namely: they have features which affect evolvability, similarly to how features of populations (i.e., properties at the populationlevel, such as population-structure) affect evolvability. In other terms, both the spatial dimension (population-structure) and the temporal dimension (genealogical-structure) of populations affect evolvability. Here I discuss only the latter, showing its impact on evolvability and its recent debates. I start by describing a case-study.

3. A long-term experiment with Escherichia coli bacteria

Over the last decades, Richard Lenski's research group at Michigan State University has been carrying out a long-term experiment with Escherichia coli bacteria. Bacteria are allowed to replicate in a controlled environment, and frozen samples are extracted and stored (effectively kept in stasis) every 500 generations. This allows them to examine long-term evolutionary patterns in the populations, and directly compare different generations. A recent study considered the influence of evolvability in the evolution of E. coli, focusing on the disposition of those bacteria to increase their fitness across generations (Woods et al 2011, 1433)3. In a first experimental round, the analysis 3

The Woods et al. study defines evolvability operationally as the disposition of populations to generate higher-fitness descendants. Nonetheless, it is important to realize that evolvability is not necessarily associated with an increase in fitness.

6

of samples at 500 generations showed that two different groups of E. coli in a common environment evolved a slightly different set of mutations. The first group had a fitness deficit of 6.3% in comparison to the second one and, based on these values, the first group was expected to be extinct after another 350 generations. Surprisingly, samples at 1,000 and 1,500 generations showed a radical change: the first group took over the common environment and led the second one to extinction. To understand the observed change in the first experimental round, biologists designed a second experiment (Woods et al. 2011, 1434). They took samples of both groups at 500 generations, isolating twenty populations of each group. The bacteria replicated during 883 generations under the same environmental conditions as the first round4. As a result, the descendants of the first group overcame their initial fitness deficit and evolved to a higher fitness than the descendants of the second group by 2.1% on average. This result was obtained by isolating random members from each of the forty populations at the 883-generation endpoint and comparing their fitness. Such a comparison has showed that the great majority of members of the first group have a higher fitness than the members of the second group. Given this robust pattern, Woods et al. claim that the radical change observed in the first experimental round was not due to chance5. According to them, the first group outcompeted the second one after 1,000 generations because it was more evolvable: The first group is more likely to increase in 4 5

Shortly after 883 generations it was still possible to identify the two groups of bacteria in the same environment, but the competition between them led quickly to the extinction of the second one. This is a reason why the scientists have chosen the time span of exactly 883 generations. According to the hypothesis of change, the relative success of the first group over the second group was a result of lucky mutations. If chance were the appropriate explanation in the first experiment, we should not expect the success of the first group over the second one só many times.

7

fitness in the long-term than the second one. The next step in the research was to understand the underlying biochemical causes leading to a higher evolvability in the first group than in the second one (Woods et al. 2011, 1434-1435). That is, to attempt to identify the features of the populations in virtue of which their evolvabilities diverge. Genetic sequencing and parsimony methods allowed scientists to track mutations in each group, determining their sequences across generations. These sequences revealed information about biochemical causes of evolvability. The comparison of fitness between members of all forty populations in the second experiment showed that mutations in the gene spoT are present only in members of the first group. The reason for this is that such mutations have proven to be highly beneficial to the first group, but they are neutral or less beneficial to the second one. Hence, mutations in spoT were not retained by the “survivors” of the second group at the 883-generation endpoint. Woods et al. suggest that, since those mutations were also present in the first experimental round, they were important to the success of the first group and the extinction of the second. The difference in the interaction between spoT and each group is ultimately due to the biochemical process of negative epistasis. Negative epistasis occurs when genes interact in a way that the expected effect of a gene is undermined by other genes (Phillips 2008). As Woods et al. suggest, this process occurred in the second group, but not in the first one. The interaction between new mutations in spoT and the particular genetic composition of the second group was not fruitful for the increase of its fitness

8

(Woods et al. 2011, 1435). In other words, the usual boost in fitness provided by spoT was overshadowed by the genetic context already present in the second group of E. coli. Hence, the authors conclude that the difference in evolvability between the first and second group is – at least in part – due to a difference in how their respective genetic compositions interact with new mutations in spoT. In the next section, I discuss this conclusion and show how it teaches us a general lesson about the relation between genealogy and evolvability.

4. How Genealogy affects Evolvability

The experiments described above provide us with an example of how genealogy affects evolution. Consider how Woods et al. conclude their paper:

In essence, the ELs [second group] followed a trajectory in the fitness landscape that allowed more rapid improvement early on, but which shut the door on at least one important avenue for further improvement. By contrast, the EWs [first group] followed a path that did not preclude this option, giving them a better than otherwise expected chance of overtaking the Els (2011, p.1435).

Despite the fact that the first and second group exhibited many mutations in common, they evolved very distinct fitness trajectories, i.e., distinct rates of change in

9

fitness6. A reason for this difference is the distinctive mutations each group had in the course of evolution, such as mutations in the gene spoT. Another reason is the very sequence of mutations in each group across generations. Mutations arose in a different order in each group, and thus interacted with different genetic contexts in each group. The sequence of interactions in each group had a relevant role in their evolutionary fate, contributing ultimately to their proliferation (first group) or extinction (second group). This is an example of so-called path-dependence (Dejardins 2011). Path-dependence is a genealogical pattern occurring whenever an outcome depends not just on the initial conditions of a system, but also on other aspects of the path leading to the outcome. In other terms, the precise sequence of changes in the system affects the outcome. As a consequence, the probability of this outcome is a function of the initial conditions and the path realized (Desjardins 2011, 727). This type of causal dependence is a common pattern in the evolutionary change of populations and speciation, having significant consequences for our understanding of contingency and historicity in evolution (Ereshefsky 2014, 717-720). Path-dependence is a genealogical feature, i.e., a characteristic of ancestordescendant relationships through generations (and time). Insofar as populations evolve through generations, they can exhibit this feature. As the above quotation from Woods et al. suggests, path-dependence was an important causal factor in the evolution of both 6

What is meant by “fitness trajectories” here it is not only the increase of fitness over time, but also the acceleration or deceleration of this increase over time. “Fitness landscape” refers to the multidimensional representation of all possible fitness trajectories of a population. It is no coincidence that fitness landscape and “morphospace” are analogously defined (see footnote 1). Once the experiment discusses the evolvability of fitness, the notion of “morphospace” gets translated into “fitness landscape”.

10

groups. Without this factor, the sequence of mutations (and interactions) would not be relevant for the outcome, only the mutations themselves would be. However, due to path-dependence, different sequences of (the same or different) mutations lead to different evolutionary outcomes. Hence, path-dependence illustrates how genealogy matters to evolution. The quotation from Woods et al. also suggests that genealogy matters to evolvability. To understand this point, firstly it is important to notice the characterization of evolvability in those experiments of E. coli bacteria. Woods et al. claim that in generation 500 of the first experimental round the first and second groups have a few distinctive mutations. The mutations in the first group gave more “options” to increasing its fitness at a reasonable rate in early and later generations. Hence, the first group had the disposition to take a fitness trajectory which lead it to achieve a certain fitness value and, therefore, outcompete the second group. In contrast, mutations in the second group gave the disposition to increase in fitness fast and early on, and then gradually decelerate its increasing rate of fitness(Woods et al. 2011, 1434). This type of fitness trajectory does not allow the second group to outcompete the first group. So described, the evolvability of each group is not simply a function of how their few distinctive mutations favor a certain evolutionary outcome (i.e., a certain fitness value to be achieved). Evolvability is a function of how those mutations favor a certain evolutionary outcome by favoring certain fitness trajectories leading to this outcome. How do initial mutations favor fitness trajectories? As the experiments with E.

11

coli show, they favor certain fitness trajectories by affecting how beneficial subsequent mutations will be. For example, depending on the initial mutations in each group (at 500-generation) downstream mutations in spoT can be more or less beneficial to the population. This way of favoring fitness trajectories is revealing because it presupposes path-dependence. To some extent, it is presupposed that the benefits of mutations in spoT depend on early mutations, i.e., these benefits depend on the path leading to their respective mutations. In sum, mutations affect evolvability in a way which presupposes path-dependence7. Path-dependence should be taken to be an influence on the evolvability of the E. coli bacteria. This influence is not explicit, such as the influence of distinct initial mutations. Rather, path-dependence works as a background condition to evolvability, underlying all possible fitness trajectories to be favored by those mutations 8. This is to say that path-dependence frames – at least part of – the fitness landscape. More precisely, path-dependence frames the fitness landscape because every change or move in a fitness trajectory presupposes path-dependence. To have a better grasp of how path-dependence affects evolvability, it is useful to turn back to Brown's characterization of evolvability and analyze how it could apply to the experiments on E. coli. Suppose we want to analyze only the second experimental round. We want to understand how, in the course of 883 generations, the second group 7

8

There is another dimension of path-dependence going on here. Woods et al. noticed that none of the twenty populations belonging to the second group (in the second experimental round) evolved mutations in spoT. The reason for this is that early mutations in other genes do not favor the rise of mutations in spoT later. Here I make reference to the traditional distinction between background conditions and causes, but I remain neutral about the metaphysical status of this distinction.

12

could not achieve a certain fitness value (evolutionary outcome) available to the first one. Given that the evolvability of the first group (E1) is higher than the evolvability of the second (E2), we have:

E1: Pr x1, b (F t) > E2: Pr x2, b (F t)

Firstly, to represent evolvability it is necessary to fix a particular span of time (t). Since we analyze the second experimental round, we can delimit the period of 883 generations as that time span. For representational purposes only, suppose this fitness value is 1.5 relative to the fitness of the ancestor bacteria 9. The parameter F corresponds to this fitness value. Since both groups were replicated in the same environmental conditions, the parameter b is the same for them. For this reason, the difference in evolvability between these groups must be a function of their internal features x1 and x2 at the generation 0 from 883 (this is equivalent to generation 500 in the first experimental round). Hence, we have:

E1: Pr x1, b (1.5 883) > E2: Pr x2, b (1.5 883)

As I described before, the two groups of E. coli bacteria differ in having slightly distinct mutations at the beginning of the second experiment. The first group had 9

F must correspond to the fitness value actually achieved by the first group after 883 generations. Since Woods et al. do not calculate that value, I stipulate one for representational purposes. The authors limit themselves to say that F is 2.1% higher than the fitness value of the second group after 883 generations.

13

mutations in the genes TopA and rbs, whereas the second group had mutations in TopA1 and FadR. These initial mutations are part of x1 and x2 respectively. Hence, everything else being equal, these mutations seem to be the only reason why the first and second groups favor different fitness trajectories and, therefore, different evolutionary outcomes10. But is this true? Yes, but only if path-dependence is presupposed. Here is why. The initial mutations present in x1 favor certain fitness trajectories leading to a high fitness because mutations in TopA favors beneficial impacts of mutations in spoT, whereas these beneficial impacts of mutations are disfavored by TopA1. Hence, the way in which initial mutations favor certain trajectories, making the difference for the evolvability of E. coli bacteria, is by affecting how beneficial the subsequent mutations will be11. As I argued above, this way of favoring trajectories or “making a difference” just makes sense if we presuppose path-dependence. In sum, we have to presuppose dependence relations among mutations in order to assume that mutations affect evolvability by favoring the rise of subsequent mutations. Since path-dependence is a common pattern in the evolution, it should be assumed that its influence on evolvability of different groups is widespread. This can be a reason to abstract path-dependence away from representations (models) of evolvability12. Path-dependence can be treated as a background condition to evolvability 10 Another way to say this is by claiming that those initial mutations are the difference makers (Woodward 2003, Waters 2007). Nevertheless, I remain neutral about theories of causation here. 11 The relation between TopA1 and spoT is one instance of what was called negative epistasis (section 3). Mutations in spoT have a neutral or less beneficial impact on the fitness of bacteria that already have mutations in TopA1. See also footnote 7. 12 In her paper, Brown illustrates a robust-process explanation, from which evolvability is an instance (2014, 9). One way to incorporate path-dependence in evolvability is to say that each arrow in her illustration of robust-process explanation implies path-dependence.

14

of many groups and, as such, do not need to be explicitly represented. This is why my position is not in conflict with Brown's characterization of evolvability. Nonetheless, using Brown's characterization of evolvability we can highlight a certain type of causal influence on evolvability which is not discussed in literature. The fact that pathdependence can be seen as a background condition in representations of evolvability does not imply that path-dependence does not have causal relevance. Moving beyond the experiments of E. coli, it is important to recognize that many other genealogical features affect evolvability. It is not obvious that such features also work as background conditions or if they should be explicitly accommodated in Brown's equation as part of x. For instance, Brown discusses briefly how evolvability is affected by generation time. This feature concerns the time span between two consecutive generations and, therefore, concerns a feature of ancestor-descendant relationships in time (Brown 2014, 18). Generation time is thus a genealogical feature. Since generations can be short or long in time, a population can have more or less mutations and achieve a certain evolutionary outcome faster or slower than others. As a result, this population is more or less evolvable. There are many other genealogical features in need of discussion. For instance, processes such as lateral gene transfer (LGT) and hybridization clearly affect evolvability. They are not recognized as genealogical features, but they should be. LGT and hybridization promote exchange and connections between different ancestordescendant relationships and can alter the very structure of those relationships. Hence,

15

they can change the disposition of populations to evolve in certain directions. For instance, Parsons et al. (2011) show how hybridization affects the evolution of Cichlid fish in Lake Malawi in Tanzania. Such fish were able to evolve rapidly due to the genetic recombination and transgressive segregation made possible by hybridization. In particular, hybridization was advantageous because it had occurred in long-term evolution, having an impact on the genealogical patterns of those fish. Genealogy is not simply a product of evolvability. As I indicated in this section, the existence of genealogical features influences evolvability and should be taken into account by the literature. It is philosophically relevant to highlight this particular class of features. First, it is a type of feature that is barely recognized as important for evolvability. Maybe this lack of recognition is related to the usual focus of biologists on developmental causes, which I take to be a limitation of the debate about evolvability. Such a limitation needs to be overcome if we want to understand the variety of causal factors acting on evolvability. Second, seeing how genealogical effects on evolvability can lead us to recognize a more profound connection between genealogy, time and evolvability. Evolvability is not simply a matter of how easily features of populations and their organisms allow them to reach certain evolutionary outcomes; but it is also a matter of how these populations and their organisms are connected through generations and time. The way these connections are set up affects evolvability, as much as spatial organizations within and among populations do. In the next section, I show how this discussion ties into the notion of “lineages”, highlighting some roles this notion plays in

16

biological practices.

5. Lineages in Phylogeny and Evolvability

In the context of philosophical debates about homology, Ingo Brigandt claims that evolvability is manifested through phylogeny – the evolutionary history of biological groups (2007, 15). His claim is based on the idea that evolvability affects phylogeny by affecting the evolution of populations through time 13. Hence, the evolutionary patterns (outcomes) represented in phylogeny are – at least in part – the result of evolvability. For example, evolvability contributes to the divergent evolution of limb ratios between primates and monkeys (Young et al. 2011; Brown 2014). The phylogeny of primates and monkeys shows that divergence in limb ratios. For this reason, it manifests the effects (or, to use the famous ancient jargon, it manifests the actualizations) of evolvability. The relation between evolvability and phylogeny seems to be clearer when we recognize that both rely on the notion of lineages. Phylogeny is the reconstruction of evolutionary history and, therefore, the representation of lineages – sequences of biological entities (e.g., genes, organisms, species) connected through ancestry (Velasco 2013). Evolvability is a disposition whose manifestation occurs through lineages 14. After 13 It is a matter of debate whether causal notions – such as “to affect” or “to influence” – should be applied to dispositions such as “evolvability”. I do not discuss this point here, but adopt the causal vocabulary for the sake of simplicity. My argument is not compromised by adopting such a vocabulary or not. 14 Calcott uses the notion of lineages to discuss a type of explanation in the mechanism debate in biology (2009). Here I discuss this notion in a different context, which is closer to phylogenetics.

17

all, there is no evolvability if populations cannot persist across generations, i.e., if populations cannot form lineages. Hence, evolvability should be more appropriately defined as the disposition of populational lineages15. Lineages-talk helps us to recognize the connection between evolvability and phylogeny beyond the considerations made by Brigandt (2007). Populational lineages occur at different hierarchical levels, such as genes, organisms, species, and genera. For this reason, they can compose or be composed of lineages at other hierarchical levels (Haber 2012). The upshot is that evolvability can affect different portions of phylogeny and express itself in different levels. For instance, evolvability can be responsible for differences among wide-range lineages, such as monkeys and primates, as well as differences among lineages within a single species, such as populations of E. coli. Moreover, evolvability can contribute to genealogical discordance among lineages at different levels since those lineages can have different dispositions at the same time. The above considerations suggest that the relation between evolvability and phylogeny is much more subtle than previously recognized16. The importance of lineages to evolvability goes far beyond their manifesting the

15 This point is only tacitly made by most authors, since they treat populations as entities persisting through different generations (e.g., Love 2003, Pigliucci 2008; Brown 2014). Such authors define evolvability as dispositions of populations, but they could express themselves more clearly by defining evolvability as a disposition of populational lineages. 16 In this respect, one last consideration is worth noticing. Lineages-talk establishes a common ground for evolvability and phylogeny, but this should not imply that lineages need to be individuated in the very same way in both contexts. Phylogenetic studies treat lineages as synonym of “monophyletic groups”, i.e., groups composed of a common ancestor and all its descendants (Ereshefsky 2001). It is far from obvious that researchers of evolvability have to commit themselves to such a way to establish lineage boundaries. In fact, it is not clear whether these researchers have to adopt one and the same criteria for lineages individuation.

18

effects of evolvability in phylogeny. As I claim in the remaining of this section, lineages can play different and important roles in studying evolvability. This claim is substantiated if we turn back to the experiments with E. coli bacteria. In those experiments, tracing (tracking) bacterial lineages allowed researchers to identify a phenomenon to be explained: a bacterial group was able to increase its fitness in the long-term evolution, out-competing a second bacterial group that had a higher fitness in the short-term17. After the identification of this phenomenon, two alternative explanations were considered, namely: chance or evolvability. The tracing of new bacterial lineages allowed researchers to follow the evolutionary dynamic of E. coli in more detail, tracking their sequences of changes (Woods et al 2011, 1433). This method was the basis for the evaluation of the alternative explanations. In this regard, the researchers concluded that an explanation based on evolvability has proven more appropriate than an explanation based on chance for the phenomena in question. The method of tracing lineages was also important to the measure of increase in fitness of E. coli bacteria and to the study of the underlying causes of evolvability (Woods et al 2011, 1434). The tracing of lineages seems to offer a way to mitigate a problem concerning evolvability. This is the problem of testing evolvability hypotheses. As some authors point out, research on evolvability suffer from a lack of quantitative and testable

17 Tracing lineages is a method often used in developmental biology and consists in tracking ancestordescent relationships across generations of biological entities. For instance, this method allows scientists to identify genealogical patterns in the proliferation of cells during different embryological stages. For more details, see Kreztschmar 2012.

19

methods (Siniegowski & Murphy 2006). Without such methods researchers are hardly able to make predictions about evolvability and test it against alternative hypotheses. Given this problem, the relevance of the experiments with E. coli bacteria is clear. These experiments provide a case where evolvability hypotheses can be tested, predictions can be made, and evolvability of fitness can even be quantified. The method of tracing lineages is at the heart of those experiments and grounds the test of the evolvability hypothesis18. At this point, it is possible to raise a concern about my use of experiments with E. coli. These experiments are part of a very specific research program and, for this reason, it can be argued that they are not representative of how lineages figure in research on evolvability. In particular, tracing lineages seems only a suitable method for studies of experimental evolution, when it is possible to observe the evolution of lineages under controlled settings. This is a problem because most of research about evolvability does not deal with experimental evolution. In sum, the concern here is how generalizable the lessons extracted from those experiments can be. The above concern is reasonable, but it is not enough to prevent us from recognizing the importance of lineages to evolvability. Even though the specific method of tracing lineages maybe restricted to research in experimental evolution, the roles of lineages I pointed out are not. First, lineages can be relevant to the identification of

18 Both Brown (2014) and Brigandt (2015) raise the difficulty of distinguishing evolvability from natural selection in practice. Nevertheless, I suspect that tracing lineages can be a way to disentangle issues like these. This methodology allows us to directly manipulate or observe changes in lineages, assessing the effects of evolvability under stable environments (Siniegowski & Murphy 2006, 834).

20

different phenomena related to evolvability. Second, they can be relevant to the test of alternative explanatory hypotheses concerning such phenomena. I contend that the experiments with E. coli are representative in virtue of capturing these two roles. In fact, I assume that many researchers of evolvability attribute – explicitly or not – such roles to lineages19. In the previous section, I argued that genealogical features influence evolvability. The recognition of this influence sheds new light on the importance of lineages for the study of evolvability. As I noticed, there is a certain type of cause of evolvability which concerns the very formation and structure of ancestor-descent relationships. These causes – genealogical features – capture intergenerational aspects of populations, i.e., they are about population-level lineages. Hence, by identifying and discussing such causes researchers are tacitly making reference to lineages. Furthermore, the identification and discussion of genealogical features can benefit from methodologies

(such as

tracing lineages) targeting

lineages

(i.e., targeting

intergenerational relations) instead of populations at specifics time-slices (e.g., focusing only on developmental and genetic components of a population at a time). If philosophers and biologists aim to understand the different features affecting evolvability, they are impelled to focus on lineages to uncover those features concerning ancestor-descendant relationships. For instance, given the capacity to replicate E. coli bacteria in controlled conditions, the experiments discussed in this paper focus on lineages to research evolvability. For this reason, these experiments present a good 19 In this sense, my aim here is not to vindicate such roles, but to make them explicit.

21

opportunity to recognize the influence of genealogical features in evolvability. The focus on the internal features of populations at a time, in particular the focus on developmental and genetic factors within its organisms, was considered a limitation of debates about evolvability in the previous section. As I have argued, such a focus overshadows the importance of genealogical features (and time) in evolvability. The overshadowing of genealogical features is connected to the overshadowing of the importance lineages to evolvability. After all, once we do not recognize the role of genealogy in evolvability, it is hard to realize the roles lineages (products of genealogical relations) play in evolvability. One way to downplay these roles is to characterize lineages merely as phylogenetic manifestations of evolvability. I take this characterization to be another limitation of debates about evolvability. As I have argued, lineages play different roles in researches about that topic. The concept of lineages has relevant uses beyond the realm of phylogenetics. I will conclude by saying what all my considerations about genealogical features and lineages mean for the epistemic roles of evolvability.

6. Conclusion

A look at the philosophical literature concerning evolvability indicates two main epistemic roles played by this concept (Brigandt 2007; Sterelny 2007; Pigliucci 2008). On the one hand, evolvability is an explanans of evolution and evolutionary history

22

(e.g., patterns in phylogeny) insofar it indicates the objective probability of evolutionary outcomes given certain “initial conditions”. More precisely, evolvability measures the extent to which internal features of populations and their environments make a certain evolutionary outcome likely to appear. On the other hand, evolvability is an explanandum in evolutionary biology. Since evolvability depends (or supervenes) on internal features of the population and environmental conditions, it can be said that it is explained by these elements. The recognition of the influence of genealogy on evolvability has implications for both epistemic roles of evolvability. First, by identifying genealogical features, I am broadening the range of elements which can affect and, therefore, explain evolvability. My point here is that ancestor-descendant relationship should be taken into consideration in many situations where we want to understand the underlying causes (whether seen as “true causes” or background conditions) of evolvability. Second, once genealogical features are taken into consideration, they can affect the range of phenomena explained by evolvability. This is to say that the explanatory capacity we attribute to evolvability changes with our knowledge of the features which affect evolvability. The discussion of genealogy led us to the discussion about certain genealogical products – lineages. The way lineages are articulated in phylogeny has to be explained by evolutionary biologists and evolvability is one of the many conceptual resources these scientists rely on. Hence, evolutionary history (as represented by phylogeny)

23

seems to be – at least in part – explained by evolvability. This suggestion connects nicely with the idea that lineages manifest the effects of evolvability – they “carry” the actual evolutionary outcomes that were made likely due to evolvability. After all, the explanation goes from evolvability to lineages because the evolution of the latter is affected by the former. The problem here is that lineages are not simply a target for evolvability explanations. Lineages can play relevant roles in the studies of evolvability, as the methodology of “tracing lineages” in E. coli exemplifies. For instance, paying attention to lineages seems adequate if we want to identify the genealogical features which affect evolvability. Hence, lineages are conceptual resources that biologists – tacitly or not – use and can use in order to increase their knowledge about evolvability. The study of lineages can lead biologists to better understand the causes of evolvability and, as a result, to better understand the phenomena that evolvability is able to explain.

Acknowledgments

I thanks Thomas Reydon, Marc Ereshefsky, Ken Waters, Adrian Currie, Jay Odenbaugh, Alison McConwell, Sheldon Chow and audiences at both Hannover Universität and University of Calgary for comments and suggestions.

Bibliography Brigandt, Ingo. 2007. “Typology Now: Homology and Developmental Constraints

24

Explain Evolvability”. Biology and Philosophy 22:709-725.

Brigandt, Ingo. 2015. “From Developmental Constraint to Evolvability: How Concepts Figure in Explanation and Disciplinary Identity”. In Conceptual Change in Biology: Scientific and Philosophical Perspectives on Evolution and Development, ed. Love Alan, 305-325. New York: Springer

Brown, Rachael. 2014. “What Evolvability Really Is”. British Journal for the Philosophy of Science. 65: 549-572.

Calcott, Brett. 2009. Lineages Explanation. British Journal for the Philosophy of Science. 60:51-78.

Dawkins, Richard. 2003. “The evolution of evolvability”. In On growth, form and computers, ed. Kumar Sanjeev and Bentley Peter, 239-255. London: Elsevier.

Desjardins, Eric. 2011. “Historicity and Experimental Evolution”. Biology and Philosophy 26:339-364. De Queiroz, Kevin. 1999. “The general lineage concept of species and the defining properties of the species category”. In: Species: New Interdisciplinary Essays, ed. Wilson, Robert, 49-91. Cambridge: MIT press.

25

Ereshefsky, Marc. 2001. The Poverty of the Linnaean Hierarchy: A Philosophical Study of Biological Taxonomy, Cambridge : Cambridge University Press. Ereshefsky, Marc. 2014. “Consilience, Historicity, and the Species Problem.” In Evolutionary Biology: Conceptual, Ethical and Religious Issues, ed. Thompson and Walsh. Cambridge University Press. Haber, M.H., 2012. “Multilevel lineages and multidimensional trees: The levels of lineage and phylogeny reconstruction”. Philosophy of Science, 79:609-623. Kirschner, M., & Gerhart, J. 1998. “Evolvability”. Proceedings of the National Academy of Sciences 95: 8420-8427. Kretzschmar, K., & Watt, F. M. 2012. “Lineage tracing”. Cell 148: 33-45. Love, Alan 2003. “Evolvability, Dispositions, and Intrinsicality”. Philosophy of Science 70:1015-1027

Parsons, K. J., Son, Y. H., & Albertson, R. C. 2011. “Hybridization promotes evolvability in African cichlids: connections between transgressive segregation and phenotypic integration”. Evolutionary Biology 38: 306-315.

Phillips, P. C. 2008. “Epistasis—the essential role of gene interactions in the structure and evolution of genetic systems”. Nature Reviews Genetics 9: 855-867.

26

Pigliucci, Massimo 2008. “Is Evolvability Evolvable?”. Nature Reviews Genetics 9:7582 Sniegowski, P. D., & Murphy, H. A. 2006. “Evolvability”. Current Biology 16: R831R834.

Sterelny, Kim. 2007. “Evolvability”. Philosophy of Biology, ed. Matthen, Mohan; Stephens, Christopher, 163-178. North-Holland: Oxford.

Sterelny, Kim. 2011. “Evolvability Reconsidered”. In The Major Transitions in Evolution Revisited, ed. Calcott, Brett and Sterelny, Kim, 83-100. MIT Press 83-100.

Wagner, G. P., & Altenberg, L. 1996. “Perspective: complex adaptations and the evolution of evolvability”. Evolution 967-976.

Waters, C.K., 2007. Causes that make a difference. The Journal of Philosophy, 104: 551-579.

Woods, R. J., Barrick, J. E., Cooper, T. F., Shrestha, U., Kauth, M. R., & Lenski, R. E. 2011. “Second-order selection for evolvability in a large Escherichia coli population”. Science 331: 1433-1436.

27

Woodward, J., 2005. Making things happen: A theory of causal explanation. Oxford University Press.

Young, N. M., Wagner, G. P., & Hallgrímsson, B. 2010. “Development and the evolvability of human limbs”. Proceedings of the National Academy of Sciences 107:3400-3405.

28