Multiple Myeloma (MM) is an incurable, but treatable bone marrow resident plasma cell cancer. Despite a cohort of effective anti-MM agents, patients’ initial response to therapy is followed by several phases of relapse characterized by increasingly short-lived responses as shown in the figure above, where each blue disc is a tumor burden measurement made in the clinic shown on the y-axis and the treatment received by the patient is highlighted. The inter-patient

variability of tumors typically found in MM patients makes finding the right therapy, at the right time, for a given patient, a challenge. However, by co-culturing cancer cells obtained from a patient in an ex vivo (outside a living organism, human) reconstruction of the bone marrow serves as an ideal framework to test the efficacy of standard of care drugs for a given patient prior to prescribing therapy. EMMA (Ex vivo Mathematical Myeloma Advisor) develops patient-specific mathematical models that are informed by data obtained from an ex vivo assay involving a co-culture of patient-derived myeloma cells from a bone marrow biopsy, along with human bone marrow stromal cells, patient-derived plasma, fetal bovine serum, and media in a multiwell plate as shown in the figure below. Each well is seeded with a specific concentration of a drug and imaged using a microscope for over 4 days.


An image processing algorithm is used to identify percent live cells (highlighted in red, in the videos) at every time point, for five different concentrations, for each of the 32 drugs. The two time lapse videos (above) show live multiple myeloma cells (highlighted in red) responding to 5uM of Doxorubicin and 5uM of Selinexor, respectively. The videos and the plots below them indicate that Patient180's response to the same concentration of the two drugs is different. Further, different patients respond differently to the same concentration of a given drug. This is attributed to the intratumoral heterogeneity that may vary from patient-to-patient and from time-to time for a given patient. This necessitates quantifying intratumoral heterogeneity for each patient and estimating sensitivity for each of the subpopulations to 32 drugs in order to simulate the 90 day response of that patient for each of the 32 drugs within five days post biopsy.


EMMA is uniquely equipped to capture patient-specific response across time and concentration as shown in the figure above (left). At the center of the EMMA framework lies the concept of drug induced damage, which can be perceived as a precursor to cell death. The effect of this damage is proportional to the duration each cell is exposed to the drug and accumulates over time eventually leading to cell death. Intratumoral heterogeneity is modeled by assuming that the tumor is composed of several subpopulations, each with a specific threshold for the drug-induced damage.


Four models are considered (depicted below), each accounting for heterogeneity differently. The first model (1 Population - No Distribution) assumes the tumor is composed of one homogeneous population, where all the cells have the same threshold. The 2 Population - No Distribution Model assumes two homogeneous sub populations. The 1 Population - Distribution model assumes that several subpopulations exist, where each subpopulation's threshold is part of a normal distribution. The 2 Population - Distribution model assumes two normally distributed thresholds for the subpopulations. The four models are fit to the same data as shown in the figure above (right). A statistical nodel identification tool called Akaike Information Criterion (AIC) is used to identify the model that best fits the data over noise.


Multiple myeloma cells inside the human body aren't exposed to a constant concentration of the drug as is the case in the ex vivo reconstruction of the bone marrow used to estimate patient-specific drug sensitivity. The drug is typically cleared from the blood at a certain rate, which is estimated in a Phase I clinical trial. This information is coupled with the patient-specific model parameters estimated above

to simulate a patient's clinical response for upto 32 drugs. These results are compiled into a report, which is currently featured in the weekly tumor board meetings to aid physicians in the process of clinical decision making.