Sunday, March 3, 2013

Darwinian Dynamics of Cancer

In the NEJM article by Aparicio and Caldas, the authors make a case for a better understanding and application of clonal genome analysis in diagnostics, prognostics and treatment of cancers[1]. As the authors states: 

The notion that most cancers are ecosystems of evolving clones has implications for clinical application; we review these, with particular focus on epithelial cancers. Darwin's theory of evolution was originally developed in the context of speciation. It has proved to be a fundamental property of biologic systems, including human cancers. Although tumor evolution has been a foundational concept in cancer biology for decades, a new focus has arrived with the advent of methods that make comprehensive sequencing of cancer genomes tractable for the first time.

Now the authors again and again come back to two premises:

1. Clonal Hypothesis. Namely that cancer is purely clonal and that there is a sing cell that has become aberrant and from that cell all others progress. On the one hand that generally applies to hematopoietic cancers and may apply more broadly, but one may ask where the cancer stem cell theory fits in and further that there are on a cell by cell basis complex and variant genetic profiles.

2. Darwinian Evolution. Namely that cells continue to evolve from the initial stem cells and that it is this Darwinian evolution which drives the difficulty of dealing with cancer.

The authors do raise the need to understand cancer cells not as we do today in some aggregate but in a manner which we can do today, namely o a cell by cell basis. One agrees with this goal, one achievable today, albeit at some substantial cost.

I believe that what the authors are getting at is that to develop better diagnostics, prognostics and therapeutics that we must recognize the following:

1. We must look at cancers on a cell by cell basis.

2. We must know the “age” of each cell, namely how many mitotic changes between the initial precipitating event and the cell we are examining.

3. We must know also the status of the parent cells, the peers and the descendents; thus opening up knowledge to what changes occurred where and when. This opens the door to understanding the dynamics of cancer cell progression.

Thus we need a temporal map of cancer cells.

I would argue that we also need a spatial map. The reason for this is as follows:

1. Tumors spread and mutate but besides the somatic genetic failings of a cancer cell we also have the epigenetic failings, miRNA and methylation for example, plus we have the environmental failings, namely cancer cells seem to adapt and prosper in their new environments, oftentimes actually employing the new environment for proliferation advantages.

2. Cancer cells will change as they mutate, and the somatic genetic change can be influenced by their past as well as their current environment.

3. There is a question as to how predictable these changes are. Can we estimate what will happen next knowing the past history of a cell. Is such a change a Markovian stochastic process such that some Bayesian approach may be employed to understand what will happen next and where it will happen.

4. Therapeutics must match not only the current defects, namely kinase inhibitors and BRAF blockers, but most anticipate the next cancer cell break through.

The authors do relate to some of these observation as follows: 

Fixed somatic mutations can be used to infer lineage relationships between cells. Large-scale chromosomal aberrations have been used for decades to elucidate clonal structure in cancers. Until very recently, the ability to systematically enumerate single-nucleotide mutations within tumors was limited. However, the advent of next-generation sequencing devices has dramatically reduced the cost and increased the scale of genome sequencing. Moreover, most next-generation sequencing devices can provide a measure of allele prevalence for almost any aberration found in a genome.

As we have discussed before, we can now take multiple samples of tumor cells, from various parts of the body, and then determine both genetic and epigenetic variances. They continue: 

In a mixed population of tumor clones, three key concepts relate the prevalence of fixed aberrations measured in the whole population (allele prevalence) to that of the underlying clones. First, clonal-mutation prevalence is a compound measure of the population abundance of the mutation in question and is a function of the size of each clonal population bearing the mutation. Second, a clonal genotype refers to the set of common fixed mutations that define a clone. Third, clonal lineage defines the relationship between clones as they evolve over time. Clonal relationships are usually conceptualized as branched, treelike structures. However, many complex topologies may exist with the potential for extinction events, the independent occurrence of secondary mutations (including identical secondary mutations), and mutations occurring at different scales (from chromosome aberrations to single-nucleotide variants) in the genome.

What the authors speculate about is the use of blood serum to isolate circulating tumor cells and perform the analysis we have discussed. Namely they state: 

Even more exciting is the ability to use direct sequencing of plasma DNA to identify the mutations in circulating tumor DNA (ctDNA), effectively transforming a blood sample into a “liquid biopsy.” This approach has the potential to be used, for example, in patients with multiple or inaccessible metastases to characterize the mutational complement in some or all of the metastatic lesions. This has been reported for one gene, PIK3CA, in which mutations were identified in the plasma of 28 to 29% of patients with metastatic breast cancer. The presence of the mutation in PIK3CA in plasma DNA correlated with the mutation status of PIK3CA in a metastatic-tumor specimen collected synchronously with the blood sample. The development of ctDNA genotyping for liquid biopsy will depend on the demonstration that mutations are sufficiently sensitive and specific to monitor tumor burden.

As we know, tumor cells diffuse, flow and proliferate, and that a simple model for this is one where we have a temporal-spatial diffusion like equation which stipulates the number of cancel cells for each location and time point in the inflicted human target. This model, however, fails in presenting ongoing mutations of genes and epigenetic factors as well. Namely it is a pure clonal model, that it, it depicts the single mutated clone going forth and multiplying. We know that such a model is of interest but of limited value. One could enhance the mode by making the unit controlled by the diffusion model a vector, where the vector entities are various genetic states, thus we may start with an n1(x,t) due to BRAF V600 and then have a new n, say n2(x,t) which is MEK, and so forth. This yields a Markovian progression in a diffusion model.

This type of modeling based upon measureable gene expressions are critical for prognostic efforts and therapeutics.

However, back to the authors statement, the use of serum markers is a problem because we do not know if the serum marker is coming or going. Namely is the cell we have selected one on the way to the liver or coming from the liver. We know that most tumor cells in the serum are merely flowing from one place to another, they diffuse through the walls of the blood stream and then multiple in an organ, not the blood stream. Thus the marker may be of some use but is just a single metric. I believe it fails in the goal of the spatio-temporal Markovian model.

The authors conclude: 

A greater issue is that most agents are developed to inhibit single targets (or a couple of closely related targets). Targeting the most common driver mutation alone will not succeed if mutations that confer resistance are already present as minor clones in the cancer (see above). The genetic variability of cancer and its capacity to evolve mean that most single-target approaches select for resistant clones, which expand and become dominant. Durable control of viral replication was achieved in HIV therapy only when triple combinations of antiretroviral agents, sufficient to suppress clonal evolution of the virus, were developed. In the context of cancer research, better knowledge of how single targets could be combined from the outset is essential. The ability to follow which clonal genotypes are sensitive and which are resistant could be valuable in both the early stages (xenograft studies) and late stages (phase 1–2 trials) of drug development.

This is of course a critical issue. We know from imatinib and CML that we get a slowdown and regression but it returns and then enters a blast phase. We know with decitabine and MDS we stop methylation but for just so long, and we know with vemurafenib that we stop BRAF issues but again just for so long.

The previous work make the following clear:

1. When we understand aberrant pathways we can develop a therapeutic to address it.

2. Cancer cells change, just how we may not fully understand, and they do so in a Darwinian manner, and thus we must find another pathway control to stop the next step, and so forth.

The authors have thus raised a critical issue. We argue however that it need a workable paradigm to address its usefulness going forward.