A new approach to detect cancer was recently developed by an award-winning Brazilian scientist.
Dr. Daniel De Carvalho, Associate Professor at the Dept. of Medical Biophysics, University of Toronto, Senior Scientist and Canada Research chair at the Princess Margaret Cancer Centre, leads the development of a new early detection methodology that combines identification of DNA methylation changes and artificial intelligence, that can increase cancer cure chances. His work has been featured by NEJM, Cell, Nature Medicine, JCO, and the Wall Street journal and Dr. De Carvalho have received multiple awards for his research on cancer epigenetics.
In our interview, he describes his methodology, the current stage of cancer detection and cure, and the relevance of AI and machine-learning in his approach.
BayBrazil: Could you describe how this new methodology works and why the current cancer testing may be inaccurate at an early stage?
Daniel: Currently, blood-based tests for cancer early detection are based on the rational that when a cancer cell dies, it releases proteins and DNA in the blood stream.
Traditional approaches are based on detecting some of these proteins. For example PSA (Prostate-Specific Antigen) for prostate cancer is a clinically used test. A problem with this approach is that not only prostate cancer cells release PSA, but also normal prostate cells. If someone has a non-malignant prostate condition, it may give a false-positive result suggesting prostate cancer. Therefore, the test has low specificity (ie. too many false-positive results). This can lead to over-diagnosis and complications associated to false-diagnosis (distress to the patient, unnecessary medical procedures, and so on).
More recent approaches are based on identifying circulating tumor DNA in the blood. These approaches look for cancer-specific mutations. Since these mutations are much less common in normal cells, the test can be more specific. However, finding these mutations is similar to try to find a needle in the haystack. There are not many recurrent mutations to look for and the circulating tumor DNA is very diluted in normal cell free DNA (for early detection, much less than 1% of the circulating DNA come from tumor cells, everything else come from normal cells). Therefore, these approaches lack sensitivity (too much false negative, as a person can have cancer but the test not be able to detect it). Another problem of looking for mutations is that they are not tissue-specific. The same mutated gene can be involved in breast cancer, pancreatic cancer and lung cancer, for example. Clinically, an early detection test needs to be able to inform where the cancer is.
Our approach is also based on identifying circulating tumor DNA in the plasma. However, instead of looking for mutations, we try to identify DNA methylation changes. These are epigenetic modifications that occur at thousands to millions of places in the DNA. These epigenetic changes are digital fingerprints of tissue of origin. For instance a liver cell and a lung cell from the same person have almost exactly the same DNA sequence (roughly 50% from the mother and 50% from the father), but they have very distinct DNA methylation patterns, that reflects their different cell lineage. In cancer, there is a massive reprogramming of the DNA methylation patterns, allowing us to discriminate between normal cells and cancer cells and allow identifying the tissue of origin. Moreover, because of the large number of modifications it is easier to identify the tumor DNA. It is like looking for a needle in the haystack, but there are thousands of needles in the haystack.
BayBrazil: Is there any estimate of the chances of cure comparing cancer detection at early stage with the traditional detection?
Daniel: It varies from cancer to cancer. Lung cancer and colorectal cancer are good examples where tumors detected at screening stage have
BayBrazil: How important is the use of artificial intelligence in this methodology? How relevant is technology contribution to improve this type of research?
Daniel: It is very relevant. Our approach is not a traditional biomarker approach, where one looks at a specific gene or marker and tells if there is more than expected (PSA for example) or look for a specific mutation in a specific cancer gene. We obtain a genome-wide signal of methylation levels across millions of genomic regions, then we train a machine-learning classifier to understand that a specific profile is likely associated with a person with pancreatic cancer, while a different profile is associated with a person with lung cancer or a healthy cancer-free individual. As our data-sets increases and the AI/ML field improve, we hope to have a test that is more sensitive and specific for cancer early detection. Moreover, since DNA methylation patterns are associated with each normal tissue, these tests could be applied for many other diseases, from metabolic diseases, cardiac disease, etc. Since it is all based on the same test, the data analyses aspect of this is fundamental.
BayBrazil: What are the next steps in the research and when may it be available to patients in general? In the future, may this test be done as a routine annual checkup?
Daniel: It is important to clarify that this is a research project. Many more validations need to take place before it can be offered to the population. The goal is to have a much larger number of samples to train our classifiers, ideally using samples from health studies. These studies follow hundreds of thousands of people for many years, and many of the participants develop cancer at some point, so blood samples are available before the cancer diagnosis. Then, once we are confident with the classifiers, a prospective trial need to test the accuracy for regulatory purposes, before the test can be offered. It is likely that a high-risk population (heavy smokers for example) is the best place to start.
BayBrazil: Is the early detection the only way to improve the cancer cure or is there research and development working on the cure in advanced stages of cancer?
Daniel: Most of the research today is focusing on treating advanced diseases. Immunotherapy is a good example.
BayBrazil: How committed are the commercial companies in the development of cancer detection and cure?
Daniel: Very committed. There are a few high-profile companies based in California trying to develop novel cancer early detection approaches. Grail and Freenome are two big examples, using genomics and AI.
BayBrazil: How is cancer research in Brazil? Is any academic team or company developing studies in the field?
Daniel: There is a lot of high quality cancer research in Brazil, mostly in the academic setting. I’m not aware of specific research on this field of liquid biopsy for cancer early detection.
BayBrazil: Tell us about your career and if it would be possible to develop this research in Brazil.
Daniel: I did my PhD training in immunology at University of Sao Paulo, my supervisor was Dr. Gustavo Amarante-Mendes, an expert in cancer immunology. Then, I did a postdoctoral training in Los Angeles under the supervision of Dr. Peter Jones a pioneer in the field of cancer epigenetics. After this postdoctoral training, I started my own group at Princess Margaret Cancer Centre / University of Toronto, focusing on merging Cancer Epigenetics, Cancer immunology and computational biology.
Regarding the second part of your question, I believe there is a strong scientific community in Brazil focusing on cancer research and very important resources in terms of clinical samples from a large and diverse population. However, more funding is necessary to develop new technologies.