Mathematical modeling and methodology to identify patient-specific immunological landscapes in CML treatment using TKI cessation and dose reduction data

Chronic myeloid leukemia (CML) is an example of how mathematical models can help on understanding and describing cancer treatment. In the last years, the paradigm in CML treatment with tyrosine kinase inhibitors (TKI) changed from a life-long treatment to a scenario where patients with good response can stop treatment and remain in treatment free remission (TFR). Although it is still not clear which are the mechanisms and markers that identify those patients, recent evidence suggests that the immune response is crucial for maintaining TFR. Here, we present an ODE model for CML treatment and the role of an anti-leukemic immune response. Keeping the model as simple as possible we show that it fits well to 21 individual time courses under standard treatment. However, the optimal fits are not unique, which leads to ambiguity in the predictions about the outcome of treatment cessation. To overcome it, we show that additional data after TKI stop allows to capture the information necessary to use the model for making predictions. Applying this methodology to those 21 patients and calculating the multiple basins of attraction of stable equilibria in the patient-specific calibrated model, we identify three qualitatively different 'immunological landscapes' among which the patients are distributed. One set corresponds to those patients that require complete CML eradication to achieve TFR, meaning in practice a lifelong therapy or a likely recurrence after TKI stop. A second class corresponds to those patients where the immune system controls residual CML cells after treatment cessation if a certain threshold is achieved. A third class corresponds to patients where the immunological control of CML is achieved only if intricated balance between TKI effects and immune activation is achieved. Mathematically, this corresponds to phase portraits where one basin of attraction presents a topological defect arising from a heteroclinic bifurcation, and model simulations suggest that such optimal balance leading to TFR can be achieved with protocols of dose reduction. Finally, we show that the information necessary to classify the patient’s immunological landscape can be obtained not only from TKI stop data, but also from measuring the effects of TKI dose reduction during a six-month period. This provides a general strategy consisting of three phases: standard treatment, then standard reduced treatment and accurate observation of response, then model-based patient-specific treatments based on the previous phase. Summed up, these results illustrate the potential of mathematical modeling to the era of personalized medicine, with CML as a concrete example, but potential to more complex cancers, and also illustrates the difficulties that mathematical oncologists may encounter on this way, such as parameter unidentifiability and possibilities to circumvent it.