Precision Medicine for Cancer: A view from the trenches
In 1956, Peter Nowell, a chemist turned physician turned cancer researcher, has just accepted a faculty position at the University of Pennsylvania after working at the Naval Radiological Defense Laboratory in San Francisco where he conducted research on radiation and bone marrow transplantation.
He was particularly interested in leukemias and lymphomas, cancers of white blood cells, the infection-fighting cells of the immune system. In particular, he was interested in a type of cancer called Chronic Myelogenous Leukemia (CML) for which there was no cure at the time. Patients who were diagnosed with the disease had a median survival of 3–5 years.
Back then, it was not yet clear that cancer was caused by genetic changes, and Dr. Nowell took the unpopular approach of looking into the chromosomes of cancer cells taken from people who were inflicted with the disease. Being new to the field, he looked for more scientists who could share the effort and help him explore that peculiar relationship between chromosomes and cancer. Unfortunately, he could not find anyone who was interested in such an endeavor, and was forced to look outside his campus, eventually finding Dr. David Hungerford, a young graduate student who was working at the Fox Chase Cancer Center, studying leukemia cells for a dissertation on human chromosomes. Together, the two observed something very strange about the chromosomes of CML patients. One of the 46 chromosomes was abnormally short. Having no clue how that chromosome was created or what its function was, they examined more cells from more people with CML, and sure enough, that odd, small chromosome was hard to miss in all of them. This tiny chromosome was later named the “Philadelphia Chromosome”, after the city where the two researchers discovered it, and as time went by, and more and more data started accumulating, the two had found the Philadelphia Chromosome in 95% of patients with CML.
Later on, and as some more advanced technologies reached maturity during the early 1980s, a Dutch researcher named Nora Keisterkamp, showed that the Philadelphia Chromosome is created by the fusion of two genes that code for 2 proteins — BCR and ABL, which under normal conditions are entirely separate.
Activity of proteins within cells is tightly regulated. They can be turned on or off using chemical modifications. However, the BCR-ABL fusion is unique in that it creates a protein without an “off switch”. The protein is stuck in “On” mode which leads to constitutive cell division that is not subjected to any regulatory signal from the outside, and thereby, forces the cells to divide and divide in a murderous, ferocious manner until the death of their host.
Sadly, even at that point in time there was nothing that could be done for those CML patients. Even when the cause for cancer was very well understood by the scientific community, there was nothing, except a dangerous and often fatal stem cell transplantation, that could stop BCR-ABL from making cells divide unstoppably.
Getting to the 1990s, after being rejected from Dana Farber Cancer Institute, and having his marriage fall apart, Dr. Brian Druker, decided to make a career change and move from Harvard to the Oregon Health and Science University and embark on a mission to find a cure forCML.
His quest led him to Nick Lyndon who worked at a Swiss pharmaceutical company Ciba-Geigy (which would merge with Sandoz in 96’ to form Novartis) that apparently had a molecule that binds the ABL protein and inhibits its action. Together they embarked on a long journey to develop what will eventually be called Gleevec, or imatinib.
After a quick succession of clinical trials, imatinib was approved by the FDA in 2001 and was considered a monumental success, a real triumph for the fight against a disease that was responsible for thousands of deaths each year. With imatinib, 98% of patients on the drug remained cancer free 5 years after treatment, an effect never seen before in the history of the disease. Not only were patients essentially cured, the treatment also had much fewer side effects than traditional chemotherapy, a treatment regimen that targeted all dividing cells in the body indiscriminately.
The success of imatinib was one of the sparks that ignited the precision medicine paradigm — can cancer be successfully treated when the treatment is guided by the genetic profile of the tumor?
Before the era of DNA sequencing and the subsequent rise of precision medicine, treatment protocols for cancer patients were determined mostly by the type of tissue from which the tumor originated (i.e. lung, liver, colon) and it was based on a cocktail of poisonous molecules that decimated all dividing cells. Following the imatinib success, pharma companies started racing to find cures targeted for specific pathogenic DNA alterations with the hope of replicating the remarkable achievement of imatinib and CML. Furthermore, the importance of basic research and its contribution to the fight against the disease was acknowledged, and money from both public and private sources started flowing into research labs around the world with the hope of finding the next BCR-ABL, or biomarker, which would enable focused, targeted therapies.
Twenty years have passed since imatinib has made its appearance on the market. A lot has happened in those 2 decades: we have identified many more molecular alterations that cause cancer, some of which already have treatments available. We have developed very sophisticated DNA sequencing methods that have lowered the cost of the test and consequently made them available to almost every cancer patient in the developed world. The capacity of such tests was increased as well Commercial tumor DNA sequencing tests offered by companies such as Foundation Medicine can now map DNA alterations in close to 500 different genes. The research community has also created massive, peta-byte (1 peta-byte = 1000 terabytes = 1 million gigabytes) scale databases such as the Catalogue of Somatic Mutations in Cancer (COSMIC) and The Cancer Genome Atlas (TCGA) to allow every researcher in the world access to high quality data.We made remarkable progress in understanding cancer and the role the immune system plays in disease progression, as was remarkably shown by James P. Allison and Taskuku Honjo who received a nobel prize in 2018 “for their discovery of cancer therapy by inhibition of negative immune regulation”. Although their discovery (that cancer could be treated by stimulating parts of the immune system to react to the tumor) was made in the 90s, the first drug that utilized that important insight (ipilimumab) was approved by the FDA only in 2011 and today constitutes an extremely effective weapon in the battle against the disease.
Despite all of these monumental achievements, the promise of precision medicine is far from being fulfilled as of 2019. First, it benefits only a small percentage of patients who have an actionable genetic alteration. For example, in a study conducted at the MD Anderson Cancer Center that involved 2600 patients, only 6.4% were found to be eligible for a therapy that matched their tumor genetic profile. Similarly, clinical trials that investigate the benefits of precision medicine, such the MSK-IMPACT and MOSCATO, report 9%-18% success rates in matching patients to targeted therapies.
And unfortunately, the identification of an actionable alteration is far from being a guarantee for a successful treatment. Four years ago, the results of a phase II clinical trial named “SHIVA”, were published, and were both surprising and disappointing. This was the first time precision medicine was put to the test. Could it provide statistically significant benefit to patients when compared to the standard of care? Or in more technical terms, can cancer be successfully treated solely based on the mutations present in patient’s tumor rather than the type of tissue and cells involved? 195 patients were selected to participate in the trial of which the vast majority had a refractory disease that metastasized and ceased to respond to conventional treatments. Half of the patients were given treatments based on the mutations present in the tumor and half were given the standard of care that was chosen by their physicians.
Surprisingly, the SHIVA trial failed to show a statistically significant difference between the two groups of patients. Both had died at about the same rate and so, the contribution of precision medicine to improving patient outcome sustained a serious blow after all the hype and promise it generated.
The results of the SHIVA trials, although published more than 4 years ago still echo to this day. People continue to debate/voice their opinion on the subject in major academic journals. Opponents of the precision medicine approach argue that we still lack a good understanding of most molecular mechanisms that lead to cancer. We understand bits and pieces, rely on anecdotes, but are not yet able to see the full, immensely complex picture. Cancer is rarely caused by one faulty gene, it typically takes 5–15 mutations in key proteins in very specific positions to create a fully fledged disease, and treating patients with drugs aimed at one faulty protein is a sure recipe for relapse. On the other hand, advocates of the precision medicine paradigm argue that the trial included too few patients and therefore was not sufficiently powered, also, the algorithm used to match patients to treatments was not optimal and should be improved. And most importantly, we are just in performance will improve over time.
Newer studies, such as the I-PREDICT that was published in Nature Medicine earlier this year,reported that 49% of the patients were successfully matched to a targeted therapy. Patients in this study were given a combination of drugs, to cover as many pathogenic molecular alterations as possible, and remarkably, patients who had most of their cancerous gene variants covered by treatments had statistically significant longer Progression Free Survival (PFS) periods (the period in which the tumor doesn’t grow or form new metastases). Could this mean that we are on the right track to cancer eradication? Hardly. Even if the differences in PFS are statistically significant, they are still small ( just a few months). Too soon the cancer is back on its feet, violent as ever before with resistance to the previous line of drugs showing how evidently, we are still very far from a real cure. Why is that?
Some cancers are caused by a handful of driving mutations that occur in proteins that control key cellular processes such as proliferation and cell death (apoptosis) but for most cancers, the culprit is far from being obvious. First and foremost, mutations found in most patients usually lead to cancer in synergy with other co-occurring mutations. Second, sequencing the genome of a tumor will occasionally reveal no pathogenic mutations. For example, some breast cancers are caused by a mutation in the DNA sequence that controls the expression levels of a protein (the expression level of a protein equals to the amount of the protein in the cell) and finding the presence of such alteration requires sequencing the RNA within the cell (RNA is translated to protein and its amount within the cell correlate with the amount of the protein). In another example, mutations in non protein coding regions were found to cause a disruption in the packaging of DNA, leading to activation of genes related to cell proliferation.
The immense complexity of the inner workings of cells and specifically the errors in those inner workings, which eventually lead to cancer, are fertile grounds for AI.
However, the vast majority of recent AI studies aimed to diagnose or predict treatment to the disease concentrate mainly on image based domains such as pathology and radiology.
Being image based domains, studies in pathology and radiology can leverage various deep learning models and methodologies developed for other domains of image analysis and indeed, they show impressive achievements. One notable example is a study by Google that was published on 2017 showing a dermatologist-level classification of skin cancer with deep neural networks. The neural network used in this study was trained on a data set called imagenet, which contained images like cars, planes and houses, unrelated to cancer or any skin condition at all. The authors simply fine-tuned this model, in a process called transfer learning, to identify malignant skin lesions. Although studies of this kind hold tremendous value, primarily for augmenting the capabilities of caregivers and diagnosticians, they provide little insight into the biology of the disease.
The amount of data that can be gathered for each cancer patient is staggering. Genomics (the DNA sequence of tumor cells), transcriptomics (the amount of RNA in tumor cells that corresponds to the abundance of cellular proteins), personal data such as age, race and so on, origin of the tumor (e.g. lung/colon) and even data about bacterial populations across the body — the microbiome and the metabolites that are produced by them. All those components cross interact; for example, there is a substantial amount of evidence that the microbiome interacts directly with our brain and may also affect the outcome of cancer treatments such as immunotherapy. Currently, there are very few patients for which this amount of data is available, and therefore very little insight can be gathered from them. There is a reason to be optimistic, however. The amount of data that is collected from patients is rapidly increasing, new treatment modalities — such as oncolytic viruses, microbiome transplantations, siRNAs, CAR-T cells and more are becoming available at an increasing rate. All those would hopefully be utilized in the future to buy more time for people who are battling this horrendous disease. On the other hand, our expectations should be properly set. In my opinion, cancer is with us to stay for the foreseeable future. There will be some successes — people will live with a stable disease for longer time periods, some of them may even be completely cured. However, the likelihood that a resistance mechanism will be discovered by the tumor increases significantly with time — tumor cells are under selective pressure exerted by both drugs and the immune system. Once such mechanism is discovered by one of the tumor cells, the new and resistant cells will out-compete their progenitors and re-initiate the disease. In the future, with plenty of treatment modalities available, we may be able to treat each such recurrence with a different drug, extending people’s lives to the average life expectancy.
