Accurate prediction of tumor growth and doubling times for triple negative breast cancer (TNBC) allows for patient-specific assessment of tumor aggressiveness


TNBC is the most aggressive of breast cancer subtypes and lacks targeted therapies. Thus, tumor aggressiveness is an important factor to determine escalation of care for TNBC patients. Currently, aggressiveness is evaluated based on tumor spread and markers of cell growth. Although these features are significantly associated with tumor volume doubling time (TVDT), TVDT distributions are not simply separated by any of these criteria. Therefore, for single patients, no clear predictor of TVDT (tumor aggressiveness) exists. We sought to test the SimBioSys TumorScope (TS), a biophysical modeling software, to determine whether it could be used for predicting tumor growth and TVDT in TNBC patients.

We used the TS software to simulate tumor growth in 30 TNBC patients. We compared the calculated TVDTs from our simulations to values obtained from literature. Using our metabolic modeling platform, we simulated tumor average specific growth rates (SGRs) from RNA-seq data taken from the TCGA BRCA data set. We found that coordinated gene expression profiles act to regulate SGR differentially across breast cancer receptor subtypes (p = 2×10-16), with ER+/PR+ tumors growing slowest (SGR = 0.49 %/day +/- 0.33, median +/- stdev) and TNBC tumors growing fastest (SGR = 1.03 %/day +/- 0.35), which match well with previously reported values of SGR in breast cancers (reported SGR = 0.175 +/- 0.979 %/day for Luminal A and SGR = 1.003 +/- 1.121 %/day for TNBC, mean +/- stdev).

We next tested how these growth rates are integrated with tumor biology to produce TVDTs on the order of 100-200 days in TNBC patients. We selected 30 patients with TNBC whose MRI data and electronic medical record (EMR) information are available in our TumorBank to use for this study. We then selected representative TNBC tumor metabolic models and mapped patients to these metabolic models based on patient phenotype. We used clinical measurements to predict unspecified model parameters for TS simulations, resulting in a model with a single free parameter (growing fraction of tumor cells). We tuned this parameter within reported values to simulate TVDT in our patient cohort. Simulating the growth of these tumors resulted in a median TVDT = 163 days with a standard deviation of 111 days, closely matching clinically measured TVDTs in TNBC patients (reported mean TVDT = 127 +/- 48 days).

In summary, TS biophysical modeling software accurately predicts TVDTs in TNBC patients. Importantly, this was accomplished without parameter fitting outside of physiologically-measured values and serves as a powerful validation of the our modeling approach. In the future, this high accuracy modeling framework can be further personalized by accounting for additional patient features, including, but limited to, patient-specific mitotic index, blood chemistry data, and tumor-specific RNA-sequencing.

View the full publication here.