You are standing on top of a large, grassy hill. As it slopes down, the ground is uneven, forming pockets and smaller hilltops. More grassy slopes roll out around you to meet the horizon on all sides – a landscape of peaks and valleys, large and small, wildly irregular.

You have with you a large bag of slightly deflated basketballs. Ignoring the ludicrous and perplexing circumstances that must have led to you standing on a hill with so many useless basketballs, you decide you might as well toss one or two down the hill, to see what happens – they don’t bounce, but they’ll roll. Soon you've emptied the bag, sending dozens rolling off in every direction. Most of them find their way to a few deep, obvious valleys, before coming to rest. But one of them seems to be stuck – it’s caught in a groove along the hill that you stand on, but the groove is so slight that you hadn't noticed it before.

It may seem alarming, but that basketball may be the reason we haven’t cured cancer yet.

A crash course in epigenetics

It’s probably not immediately obvious how you waking up on a grassy hill has anything to do with cancer genetics (long conference after-party?), but this landscape is based upon sophisticated ideas about molecular evolution. Just bear with me while I stitch the two ideas together.

Let’s start simple. If you are a mutant white-coloured moth, you are probably more likely to be eaten than your peers because you're easier for predators to spot. The trait has lowered your fitness, and if we graphed this, lighter colour would coincide with lower fitness for your species. In the next generation, there will probably be fewer white moths because more of them will get eaten in this generation. You can put all sorts of traits on a graph like this, in order to try and predict the effects of natural selection on evolution.

In a similar way, we can investigate change at other levels of biology by looking at what is changing, and the forces that can induce the change.

Our cells change all the time, but that change is not genetic. Even though your skin cells look nothing like your muscle cells, they all function from the same genetic code, they just use this code in different ways. Instead of looking at how the population’s genetics change from generation to generation (as in evolution), we’re looking instead at how each cell’s genetic regulation changes as it is subjected to a variety of forces.

The regulation of genes being turned on or off is the realm of “epigenetics”. Epigenetic reprogramming allows a cell to change and adapt in response to its physical environment and its underlying genetic make-up. It must use its epigenetic options to “cope” and find an optimal regulatory scheme for the set of genes it has inherited, as well as its physical location and circumstances in the body.

The protein and genetic networks that comprise a cell are held in order by checks and balances that link the entire system together: everything from the composition of the proteins to local ion concentrations has direct consequence on multiple levels of the cell’s activity, and keeping the systems that respond to these levels balanced allows the cell to thrive. This is actually a huge problem for geneticists: we’re not able to isolate one part of the cell and test how it modulates the cell’s fitness independently (separating it from the environment and other genes and cellular components). In order to be realistic, we must consider interaction effects. For example, in a recent paper in PloS Genetics , researchers at Michigan State University demonstrate how unexpected interactions between genes can complicate our predictions of each individual gene’s activity. Because the gene’s functionality depends so intimately on its circumstances, any attempt at isolation is not only artificial, but could give us information that differs substantially from what happens in the real world because it neglects these interaction effects entirely.

Hills and valleys

Back to the rolling hills and basketballs for a moment. This land of many grassy hills is actually a three-dimensional graph of potential epigenetic states, and the basketballs are cells. The north-south direction can represent the state of the gene A, and east-west represents gene B. By “state”, I mean the epigenetic regulation of the gene – is it allowing proteins to move in and transcribe RNA? Is the gene product correctly formed? Is the gene partially or completely silenced? Is the gene product being alternatively spliced (with chunks added or missing)? Each “state” is a different point on the north-south or east-west axis of this graph.

In the real world, this graph would have much more than three dimensions (you’d need a dimension for each gene), but good luck graphing that for a blog post.

The z-axis (up-vs-down) can be thought of as a short-term fitness; we’re talking about how cells change their regulation, not how genetics change over evolution, but the forces are similar. For the purposes of this article, I will refer to this as “fitness” as well, because the concept (a force pushing the cell into an optimal state) bears enough in common with our original definition. The low points on our graph are where the cell’s short-term fitness is highest.

In our grassy hills metaphor, the balls are attracted to the valleys because they are pulled by gravity: in the same way, the cell state can be “drawn” to fitness valleys in this graph by selection. Balls don’t roll back up: I find this easy to remember because, coming from Halifax, I hold a deep personal resentment for hills.

If the system were simple, the graph would be simple: when gene A and gene B are working normally, the fitness would be highest, and this would be represented in the graph by a single, deep valley. Everywhere else would be a hill, because that would be a deviation from normal and would likely have negative effects to reduce the cell’s fitness. Roll a ball, and it will settle in the valley. However, under realistic circumstances where we have a diverse and complicated set of interacting variables, the graph is not one simple slope and one valley. Genetic interaction creates many peaks of many different sizes, just like in the 3D graph we just looked at. The basketballs you rolled off the hill earlier, thanks to the complex terrain, could roll to all sorts of different locations under the influence gravity. On our graph, the pressure on the cell to find an optimal stable state can pull them into any number of valleys, just like gravity.

Genetic interactions and their effects on cell fitness are extremely difficult to predict a priori, and because of this, construction and interpretation of accurate, useful graphs requires excessive computational power and theoretical expertise. Modern biology is only beginning to explore their potential. Instead of two genes, we want to look at dozens, sometimes hundreds at a time. It seems computational biology is finally ready to take on this challenge. Recently, researchers have begun to tease apart the implications of systems like this one.

Adaptability is key

In May of this year, Dr. Sui Huang of the Institution for Systems Biology published a paper outlining how genetics and epigenetics can interact to influence cancer evolution. First, he outlines how cells will act on a graph similar to our grassy hills: as the population of cells evolves, different lineages settle into fitness valleys where their gene regulation program is optimized. The mathematics underlying the production of these graphs is intense, and for the purposes of this post I will instead be focusing on the salient points of the theory as it applies to cancer. (If your physics- and math-envy runs as deep as mine, I highly recommend reading the original paper after this article).

A closer look reveals how this theory extends to the differentiation of cell types within our bodies as we develop: groups of cells go about dividing, and depending upon the developmental stage, genetics, and local environment, different valleys will be accessible as the cell rolls down its fitness hill. My supervisor once referred to these changing cells as being within a search space, where they are met with several different pressures and must adapt to them by switching genes on or off. Being in a search space means they’re still rolling, “searching” for the lowest part of the valley.

If a cell is in a valley, that means it has a certain set of genes turned on and other genes turned off. Since the optimal states a cell can find should be more or less the same in healthy people, we see recurring patterns and have named the different cell types according to them. Now we finally have names for our valleys: each corresponds to a different cell type, with a Neural Valley, an Epidermal Valley, a Muscle Valley, and so on. Once settled in a valley, the cell rarely changes types without outside influence, because it would have to enter states that are less preferable to climb back over a hill. Other valleys are thus inaccessible to the differentiated cell in absence of a force to push them there. Your basketballs won’t roll back to you: they prefer to be stable.

Cancer cells and instability

Cancer cells have it a bit tougher than healthy cells. Not only are they subjected to increasingly unfamiliar environments (eg. reduced oxygen within the tumor, higher energy demands, loss of adhesion), but they also often experience increased mutation rates, which can gradually dismantle the cell's genome. Recall that when we’re talking about adaptability of cells, they are unable to change their mutated genes back; all they can do is change which genes are turned on or off. Mutation can cause genes to become activated or deactivated unexpectedly, changing the cell’s options for running epigenetic programs; this means the cancer cell sees a constantly shifting set of valleys and peaks as the underlying genetic code changes. In more math-speak, this is because the parameters of possible states the cell can enter change with each mutation. Every time a mutation alters something, the cell enters a new search space, and must find a gene regulation state in which fitness will be maximized under these new demands.

With new fitness valleys created on the graph, the cancer cell can adapt by falling into new states. When a cell is under pressure, it does not have the time or resources to search further and explore many cell states: selection pulls it into the nearest dip downward without consideration to whether this is a true valley, or a local minima (like your ball that got caught randomly on the side of a hill, instead of rolling into the valley). Further, in the same way that a normal differentiated cell can’t change cell types on its own, the cancer cell can get stuck in suboptimal locations without the means to overcome the selective forces holding it there. The resulting “cell type” it settles into may not be a normal or healthy way for the cell to function, but it’s stuck there.

In fact, Huang argues, this may be how cancer arises in the first place: the first mutations need only to pull down a key hill in the epigenetic graph, allowing access to a new valley. Also, because this isn’t a cell type that is normally expressed, it hasn’t been subjected to natural selection in order to iron out its more destructive consequences on the rest of the body. For example, whereas normal gene regulation programs give the cell the ability to self-destruct, this part of the program is very often lost in cancer, leading to unchecked growth.

This idea provides a model by which cancer cells evolve, by taking both genetics (the code) and epigenetics (the gene state) into account. If we examine cancer as a collection of altered cell states in this way, the mechanism of drug tolerance emerges immediately. As Huang argues, the addition of a drug will indeed kill certain cells, but it will also change the fitness landscape and push the remaining cells into a new search space (see Fig 4). Imagine that the cell is producing an abnormal protein that increases its growth rate, and we have a drug that will shut down the pathway making that protein. If you treat with this drug, you are changing the regulatory network of the cell – essentially picking it up from the valley and tossing it somewhere else. Many cells may die as the pathway is shut down, but those that don’t die immediately (due to variation in cancer cell lineages) will tumble into a search space again, switching to any state that could give them a chance of surviving the onslaught of the treatment. Drugs that induce mutations will create the same effect as regular mutations, changing the hills and valleys, which will also push most of the cells into a search space.

Perhaps, he argues, this is exactly what produces the aggressiveness we see in late-stage cancer. By changing the cell’s function we literally give these cells the opportunity to find more malignant forms that had previously been inaccessible: we prop them up on a new fitness hill and let them find new valleys that they never could have entered on their own.

Keeping it together

This model outlines a situation in which any alterations we make to the system will have deeply complex implications beyond the goals of the treatment itself. We're brought face-to-face with the biggest problem in biology: everything is connected, and everything influences everything else within a system. Under this model, the reduction of drug development to “Drug X will disable Protein Y to kill Cell Type Z” is not only disingenuous, but dangerous. As the familiar reader knows, this has been the guiding principle of cancer therapeutics for decades: with the advent of molecular targeted therapies, biologists have attempted to reduce the cancer problem to a simple search for individual "driver" mutations, the soft spots of the cancer's program where we can pull a string and send the entire network into shambles. But is this really the best approach? Through our efforts to study the aspects of tumor biology, have we obscured relevant information that can only be seen in the intact system?

Since the rise of genetics and advanced molecular biology, cancer research is a field that has become almost as obsessed with understanding aberrant metabolism as it is with devising therapies. Multidisciplinary, systems-level approaches such as this may help us address the challenges we’re encountering on the genetic level. If the answer lies within the patterns of the system, perhaps the difficulty of developing new drugs is partly rooted in inherent determination of modern science to reduce, partition, isolate, and classify: it’s becoming increasingly clear that this artificial isolation is undermining our analysis, not clarifying it. Maybe the problem wouldn’t seem so complicated if we were willing to step back and rethink our theories on a wider scale... a change in scenery may show us the patterns we’ve been missing, and let us know if we’ve missed the forest for the trees.


Chari S, Dworkin I (2013). The Conditional Nature of Genetic Interactions: The Consequences of Wild-Type Backgrounds on Mutational Interactions in a Genome-Wide Modifier Screen. PLoS Genet. 9(8): e1003661. doi:10.1371/journal.pgen.1003661

Velenich A, Gore J. (2013). The strength of genetic interactions scales weakly with mutational effects. Genome Biology. 14:R76. doi:10.1186/gb-2013-14-7-r76

Huang S. (2013). Genetic and non-genetic instability in tumor progression: link between the fitness landscape and the epigenetic landscape of cancer cells. Cancer Metastasis Rev. doi: 10.1007/s10555-013-9435-7

Huang, S., Ernberg, I., & Kauffman, S. (2009). Cancer attractors: a systems view of tumors from a gene network dynamics and developmental perspective. Seminars in Cell & Developmental Biology, 20(7), 869–876. doi:10.1016/j.semcdb.2009.07.003.