As a first year fellow in medical oncology at Memorial Sloan Kettering (MSK), I was about to end a five year stretch of clinical training. I am a physician scientist (MD-PhD) with my PhD training in genetic epidemiology and I was longing to get back into research. My PhD was in germline genetics of ovarian and breast cancer, and I initially thought I wanted to stay in this field. But then I found out about a potential opportunity to study clonal hematopoiesis (CH). CH is the presence of a clonal expansion of hematopoietic stem or progenitor cells in individuals who do not have hematologic disease. While CH remains benign in most people, in some cases it represents the first step in a multi-phase progression to malignancy. Many solid tumor patients at MSK have their blood sequenced using MSK-IMPACT for clinical purposes and you can use this to detect clonal hematopoiesis. We know from epidemiologic data that being exposed to certain types of chemotherapy put solid tumor patients at a much higher risk of developing myeloid neoplasms, such as acute myeloid leukemia and myelodysplastic syndrome. So I wanted to understand if exposure to certain types of therapy resulted in a higher frequency of CH. Understanding the relationship between CH and therapy can shed light into the mechanisms driving therapy-related myeloid disease (tMN) and help to develop interventional strategies. 

When I was initially discussing the project with my mentors for this project at MSK, Drs. Ross Levine, Ahmet Zehir and Elli Papaemmanuil, we discussed perhaps focusing on a few tumor types. But I thought I could apply my training in epidemiology to develop methods of extracting relevant data from the electronic health record (EHR). That way we could include the whole cohort rather than reviewing the chart to extract clinical data from a subset. We (Elli, Ross, Ahmet and I) all agreed that it was essential that we collect a large volume of sequential samples in patients getting therapy to understand the dynamics of how therapy might be promoting pre-existing CH or inducing new mutations. With these goals in mind, I set off.

The first part, extracting high quality clinical data from the EHR, was harder than I thought. I worked extensively with the informatics group at MSK including Daniel Kelly and Stuart Gardos on integrating data from multiple sources in the MSK EHR to maximize accuracy of the clinical data. The next hurdle was figuring out how best to statistically model the relationship between CH and exposure to different therapies. For this, I collaborated with Dr. Lindsay Morton at the National Cancer Institute (NCI). Lindsay was a post-doc at the NCI when I was a PhD student. Since then, she has become a leading expert in the epidemiology of therapy-related leukemia. We worked together to apply established epidemiologic methods for modeling the long-term complications of oncologic therapy to this new question. 

From this data we showed that CH was more common in people exposed to radiation therapy and specific types of chemotherapy, including topoisomerase II inhibitors and platinum. People who had been exposed to higher doses of these therapies had even higher frequencies of CH. But the association between CH and therapy wasn’t present for all genes - it was the strongest for genes involved in the DNA-damage response pathway (DDR), such as TP53. From the sequential sampling data, we realized that, in the vast majority of cases, therapy wasn’t inducing new CH but rather selecting for pre-therapy mutations in DDR CH. This meant that we could identify, even before therapy, which patients were most likely to have CH clones that expanded with therapy. 

However, the vast majority of patients who get cytotoxic therapy won’t develop tMN. So what are the characteristics of CH that put patients at the highest risk of developing tMN? Within my dataset only a small number had developed a therapy-related myeloid neoplasm. So, I reached out to several collaborators who had similar data: Drs. Nancy Gillis, Koichi Takahasi, Todd Druley, Andrew Young, Chris Gibson and Ben Ebert. We performed a combined analysis and were able to identify CH mutational and clinical features that placed solid tumor patients at the highest risk of tMN. 

I next wanted to understand if this model could prove useful clinically for an oncologist in balancing the risks of tMN against the benefits of chemotherapy. To answer this you need to establish how much CH and other clinical factors can identify patients both at a high and low absolute risk of tMN. I couldn’t directly model the absolute risk of myeloid disease in the combined dataset due to the sampling strategies in some of the studies. However, one of my PhD advisors, Dr. Monterserrat Garcia-Closas, and her collaborator, Dr. Nilanjan Chaterjee had developed a method to combine data from multiple sources to model absolute risk. We applied this method to model the absolute risk of secondary leukemia in women with breast cancer. This suggested that yes, for a very small proportion of women at the highest risk of leukemia based on CH and other features, the risks of tMN might actually outweigh the benefits of chemotherapy. This isn’t ready for clinical application right now, but it suggests the development of formal risk stratification models for tMN might be clinically relevant. 

Now, as a recently independent investigator at Washington University, I am following up on this work. My group is analyzing longer-term prospective studies in solid tumor patients receiving therapies that put them at a high risk of tMN to better understand the biology of therapy-related leukemia. We are also conducting a larger-scale study of the risk factors for tMN in solid tumor patients to develop clinical tools that will more accurately estimate the risk of tMN. Thank you for reading the paper and its journey!