Integrating proximate and ultimate causal frameworks is not a trivial task; proximate and ultimate causes are typically studied using different methodologies, work on different timescales, and depend on different underlying conceptual foundations. Similar to cancer translational research, evolutionary developmental biology (EvoDevo) aims to explore the connections between the development of individual organisms and their evolutionary transformation to discover causal-mechanistic explanations of individual traits involved in population-level events (Laubichler 2007; Hamilton 2009). Thus I propose turning to EvoDevo for assistance in integrating proximate and ultimate causal frameworks.
EvoDevo can be considered to have two research axes: (1) the evolution of development, or how developmental processes and programs change over time; and (2) the developmental basis of evolution, or how developmental processes and properties affect evolutionary trajectories (Müller 2007; Love 2015). These axes are sometimes separated into “EvoDevo” and “DevoEvo,” respectively (Hall 2000; Gilbert 2003), but here I will consider them together as “EvoDevo.” The discipline of EvoDevo is heterogenous in its methods and questions, pulling from evolution, development, molecular biology, paleontology, and many other disciplines (Müller 2008; Love 2015), and uses key concepts to orient research problems and questions. These concepts include, among others, novelty, modularity, and evolvability (Arthur 2002; Müller 2007; Love 2015).
Though the goals of EvoDevo and cancer translational research are slightly different (namely explanation versus intervention), they have similar desires—integration of population- and individual-level causes, causal-mechanistic explanations of population level events, and understanding how knowledge of individuals and populations inform one another. We can also see how cancer biology can map onto EvoDevo’s research axes: cancer is complex with developmental features evolving over the lifetime of the cancer (EvoDevo), and cancer systems have variation in relevant properties that facilitate change in certain directions (DevoEvo). Thus I suggest cancer translational research follow EvoDevo’s lead and explore how attending to the conceptual issues can help orient research programs.
Cancer researchers do sometimes invoke evolutionary developmental biology but it is often when evolution and development are known to be involved (Kumar et al. 2017) or making use of discovered biology such as details pertaining to certain signaling pathways (Alekseenko et al. 2018). The leaders of EvoDevo have identified cancer as a place where EvoDevo could be helpful in the future, but, again, they make use of the progress in biological discovery by connecting cancer with important developmental pathways and stem cells (Moczek et al. 2015). The use of EvoDevo’s approach and conceptual issues remains untouched but potentially powerful. Here I will apply a notion of modularity and then a related concept of evolvability. In both cases, I will discuss how the concept is used in EvoDevo and then how it could be applied to cancer translational research. I use modularity to show how it is not just the general concept that I find helpful for cancer translational research, but specifically the application of EvoDevo’s version of modularity. The discussion applying evolvability to cancer research is briefer as it is the concept itself that I apply rather than the history of the concept as well.
Modules and Modularity in EvoDevo
One way EvoDevo has integrated evolution and development is through the notion of modularity (von Dassow and Munro 1999). There are many definitions of modularity but a module is generally understood to be an entity or part that is discrete or autonomous in some ways but also integrated within a larger whole (usually the rest of the organism) in other ways (Wagner et al. 2007). Modules can be processes or structures (Raff 1996), and be instantiated at any hierarchical level. For example, fragments of cis-regulatory DNA (Arnone and Davidson 1997), morphogenic fields, signal transduction pathways, gene regulatory networks (Gilbert and Bolker 2001), and leaf primordia (Gass and Bolker 2003) can all be modules. Furthermore, organisms themselves can be considered modules of higher-level individuals, such as superorganisms (Schlosser and Wagner 2004). Thus, modules can be embedded in higher-order modules (Schlosser 2004).
Modularity involves part-whole relationships, but evolutionary biologists and developmental biologists have thought about those relationships differently. Evolutionary biologists consider modules to be dissociable subunits or parts of a larger system, typically the adult organism (Gass and Bolker 2003). Evolutionary modules are parts autonomous enough to change without appreciably changing other aspects of the organism. Rasmus Winther (2005) calls this a partitioning strategy, where you can understand the whole as a sum of the parts. On the other hand, developmental biologists refer to modules as collectives of entities and processes that act in some unified way to perform a function (Bolker 2000). In this view, a strategy of articulation is used when the relationship or interactions between parts is just as important, if not more, than the parts themselves (Winther 2005). In sum, for evolutionary approaches, the autonomy of a module is usually foregrounded, whereas in developmental approaches, the interactivity or integration between modules is more important.
Regardless of whether autonomy or integration is foregrounded, biologists still recognize that invoking modules and modularity requires being concerned with both autonomy and integration. One way that EvoDevo combines the evolutionary and developmental approaches to modules is that evolutionary modules are the phenotypes that result from particular developmental modules (Gass and Bolker 2003). However, there is not a one-to-one relationship between developmental modules and resulting evolutionary modules (Bolker 2000; Schlosser and Wagner 2004); multiple modules of the same type can be used in a single pathway or process, or modules can overlap by sharing elements (Schlosser 2004). Thus, what is critical for these EvoDevo researchers in bridging the developmental and evolutionary approaches is how modules are individuated and interact with one another. Once meaningful modules are identified, the interactions between them become the focus because it is the changes in interactions that result in phenotypic differences and therefore evolutionary change.Footnote 11
This EvoDevo approach to modularity combines evolutionary and developmental approaches to modules and modularity in a particular way (evolutionary modules are the products of developmental modules). By doing this, the EvoDevo researchers shift attention away from proximate and ultimate causation because they can no longer consider the proximate causes (developmental modules) separately from the ultimate causes (the variation linked to evolutionary modules). Though EvoDevo is concerned with the identification of modules (as are evolutionary biologists) and how they interact with each other (similar to developmental biologists), it is the changes in interactions over time that get foregrounded for EvoDevo because it is those changes that lead to evolutionary change and population-level events.
Using this version of modularity (where the focus is on changes in interactions over time) as a way to integrate different causal frameworks forces us to shift our attention in two main ways. First, as noted, we shift our attention away from investigative strategies that are specific to proximate or ultimate causation, and thereby avoid treating proximate and ultimate causation separately. Second, as I will discuss below, a causal perspective based on this version of modularity inherently requires the consideration of relationships between multiple levels of organization and across different timescales (e.g., developmental and evolutionary). Consideration of multiple levels and different timescales is not new, but adopting this modularity-based perspective provides a rationale for focusing on the changes in interactions between the levels and specifically across time. Thus, a modularity-based causal perspective is one that focuses on the changes in levels of integration of an identified part (module) with respect to other parts (including the organism as a whole) and how those changes affect the system of interest over time.
Modules and Modularity in Cancer Translational Research
The EvoDevo interpretation of modularity presented here was not meant to replace the evolutionary or developmental interpretations with an intermediate notion but rather integrate parts of each interpretation to allow for the discussion of multiple levels and timescales. Similarly, with cancer I do not aim to necessarily replace the biomarker or evolutionary approaches to cancer translational research, but rather provide a framework for interpreting more complex accounts of causation. Without the modularity framework, investigations usually concerned either the individual (for the biomarker approach) or populations (for the evolutionary approach). With a modularity framework in place, we can identify many potential modules in cancer—various molecular pathways, cells, tumors, tissues, the circulatory system, and/or the immune system, for example. Which modules are meaningful depends on the questions being asked or the phenomena of interest. Because the emphasis is on changes in interactions between modules, no level or system is prioritized but discussions of causal importance are still possible because the same structure for understanding causes holds across levels.
For clarity, let’s consider a generic cancer as it progresses from tumorigenesis to metastasis. At various times, different parts (modules) become the focus and sometimes the success of the cancer depends on increased interactions and sometimes decreased interactions. Cells normally interact in specific and controlled ways to form tissues, but a cell can become cancerous when it gains enough autonomy (i.e., it becomes highly modular) to defect from or cheat within the tissue. As that cell reproduces and grows into a tumor, it progressively loses some autonomy and becomes integrated with its descendent cells in the tumor. As the tumor continues to grow, nutrient and waste transport become important and its dependence on angiogenesis (the development of new blood vessels) and interactivity with the circulatory system increases. For the cancer to spread (metastasize), cells use the circulatory system to get to a new location. Once in a distant tissue, it must integrate into the tissue to evade the immune system. Additionally, when treatment is administered (especially immunotherapy), the cancer forms a different, complex iterative relationship with the immune system.
The interactions between modules and the relationships between the parts and whole are constantly changing depending on the tumor microenvironment, as long as the cancer exists. These multilevel changes in relationships across different timescales facilitate and constrain the evolution of the cancer in certain ways, thus affecting any possible cancer-related treatments. For example, if the increased dependence on the circulatory system is necessary for the evolution of the tumor, then decreasing or preventing the integration with the circulatory system would constrain the cancer’s ability to develop and evolve. Likewise, if the cancer’s survival is dependent on autonomy from the immune system, being able to decrease the cancer’s autonomy will be beneficial for the patient.
What this framework shows is that it is not necessarily better for modules to have higher degrees of autonomy or integration. It is sometimes assumed higher degrees of autonomy are beneficial as that allows independent parts to vary without affecting the function of other parts. But what the above example shows is that in some situations it will be beneficial for there to be high degrees of autonomy (e.g., beginning stages of tumorigenesis) and some situations where it will be beneficial to be integrated (e.g., colonizing distant tissues). And, of course, which is better will depend on which point of view gains the benefits (the host/patient or the cancer).
Additionally, knowing which parts of the organism to consider as meaningful parts and wholes (i.e., what should be studied) is only possible if cancer is considered in its temporal aspect. Different modules become important during different stages of cancer progression. During tumorigenesis, the meaningful part might be the cell and the meaningful whole might be the tissue in which that cell resides. And the question to ask relates to what changed such that the cell was able to become more autonomous within that tissue. The answer could be molecular, ecological, environmental, or some combination. Later in tumor growth, one might want to focus on the relationships between the tumor and the circulatory system. How and why is angiogenesis induced? How do those changes affect the evolutionary trajectory of the cancer? Furthermore, during metastasis, it will be important to investigate the relationships between the clusters of cells that are shed from the tumor, the circulatory system through which they travel, the immune system which they must evade, and the new tissues which they ultimately must colonize. How do some cells evade the immune system while others cannot? Why is a cluster of cells more successful at metastasis than single cells? How do metastases succeed in the new tissues such that they integrate enough to evade the immune system but stay autonomous enough to remain cancerous?
The shift the modularity frameworks brings is shifting from identifying that cells are able to defect from and cheat within a tissue to understanding how those cells are able to. Cells defect and cheat more often than cancers arise, so what are the differences between defections when cancers arise and when they do not? As well, millions of cancer cells circulate throughout the circulatory system but never manifest in disease (Massagué and Obenauf 2016). What are the differences between the circulating tumor cells that eventually die and those that will successfully survive to form metastases?
Tracking multilevel causal interactions through the progression of cancer has the potential to open up novel clinical treatment options that were previously hidden in the divide between proximate and ultimate causation. For example, immunotherapy appears to bridge proximate and ultimate perspectives by striving for the precision of the biomarker approach but also taking advantage of the evolving immune system to combat evolving cancerous systems. Researchers want to find proteins that are over-expressed in all and only cancer cells (such as PD-L1; Iwai et al. 2002) across all time points so that the immune system can continue to target and kill the cancer cells regardless of how the cancer evolves or its state of progression. However, if we apply what I have discussed here regarding the complexity of multilevel causal interactions across time, we can see there is still much more work to be done in order to predict and manage the side effects (e.g., autoimmune disorders) and potential complications or failures of immunotherapies (especially regarding metastasis and recurrent tumors). That is, treatments that make use of both proximate and ultimate causal frameworks still fail to predict complications of the treatment. This modularity perspective automatically suggests that changing the relationship (interactions) between the cancer and the immune system will affect the evolution of the cancer as well as the relationship between the immune system and the patient (i.e., increased chances of developing autoimmune disorders). Knowing what else each module of interest interacts with can point researchers towards potential complications as changing one interaction (e.g., increasing the interactions between the cancer and the immune system) is likely to also change other interactions (such as the interaction between the immune system and other parts of the patient).Footnote 12
As this example shows, identifying differences (i.e., the framework used in the biomarker approach) alone is not problematic. In fact, it is necessary. What are the differences that allow for some cheating cells to become cancerous and others to die? These differences make for potential places of intervention. The modularity framework reminds us that these differences need to be fully interpreted and that causes involve multiple integrated levels that need to be taken into consideration.
Evolvability in EvoDevo
Evolvability was not a concept of concern to developmental biology or much of evolutionary biology (particularly population genetics) but has, at times, been considered a central concept of EvoDevo (Hendrikse et al. 2007; Minelli 2009). Like modularity, evolvability has many definitions. We can consider the study of evolvability as the study of what facilitates evolutionary change (Love 2015). More specifically, evolvability is defined as “an organism’s capacity to generate heritable phenotypic variation” (Kirschner and Gerhart 1998, p. 8420), or “the capacity of a developmental system to evolve” (Hendrikse et al. 2007, p. 394). Despite slight differences in what level generates the variation or evolves, what is important here is the ability to generate heritable variation.
Evolvability and modularity are related, as it is often considered that high degrees of modularity are required for the evolution of complex phenotypes. But evolvability includes other characteristics besides modularity or autonomy. Marc Kirschner and John Gerhart have identified multiple properties of evolvable systems (i.e., systems that facilitate the generation of heritable variation; Gerhart and Kirschner 2003). The most well-known include compartmentation or modularity, weak regulatory linkage, and exploratory behavior (for more properties see Kirschner and Gerhart 1998). Each of these properties (including those not discussed here) reduces dependence on other parts while providing robustness and flexibility.
Evolvability in Cancer Translational Research
Evolvability is often studied at the molecular and developmental levels so it seems to be a promising way to integrate the molecular biomarker and evolutionary approaches. As well, genetic variation is known to be associated with faster progression, aggressiveness, and persistence of cancers (Maley et al. 2006; Park et al. 2010). Thus, understanding and identifying properties that facilitate the generation of variation will be important to the development of successful treatments and other clinical applications.
Treatments that use evolutionary theory are often changes in the dosing regimen or treating the microenvironment rather than the cells themselves (Gatenby et al. 2009; Enriquez-Navas et al. 2016). These approaches aim to manage the cancer rather than fully eliminate it. Some researchers have asked, though: What if we could target the evolvability instead? What are the characteristics of evolvable systems and can we target them? (Rosenberg and Queitsch 2014; Fitzgerald and Rosenberg 2017). Essentially the project is to target the drivers of evolution, rather than the products.
Genetic variation is usually thought to be the product of genetic instability due to mutations in DNA repair mechanisms or chromosomal rearrangement. A recent study has shown that, across various cell lines and samples, TGF-beta (transforming growth factor-beta) signaling downregulates DNA repair, which leads to genetically diverse populations. Additionally, exposure to TGF-beta seems to be connected to chemotherapy resistance and (thus) increased adaptability (Fitzgerald and Rosenberg 2017; Pal et al. 2017).
Whether the acquired resistance to chemotherapy actually results in more dangerous cells or cancer is still an open and empirical question but theory suggests that exposure to TGF-beta is causally associated with cancer progression and persistence. Can we target parts of the TGF-beta signaling pathway as a way of slowing evolutionary processes present in cancer progression? There are already clinical trials that target this pathway (Herbertz et al. 2015) but how successful these treatments are also depends on what other processes TGF-beta signaling is involved in. Aspects of wound healing, immune responses, and embryo development use TGF-beta signaling (Fitzgerald and Rosenberg 2017; Pal et al. 2017), reminding us that processes and systems are interconnected and have complex relationships. Thus, we might not want to fully knock down TGF-beta, as having genetic diversity is key for the success of other processes vital to healthy adult life. More work is needed to understand under what circumstances targeting evolvability would be beneficial with respect to treating the cancer, while minimizing side effects.
Again, this example shows that the identification of biomarkers is not problematic. This, and other studies, have identified TGF-beta as upregulated in cancer cells, thus making it a biomarker. However, its presence is reinterpreted within the evolutionary framework. This reinterpretation suggests criteria for why we might think targeting certain molecules or proteins might be more successful than others.