Predicting response and identifying responders to combination cancer immunotherapy in melanoma using Quantitative Systems Pharmacology (QSP) models
Rukmini Kumar (1), Kannan Thiagarajan (1), Lakshman Jagannathan (1), Liming Liu (2), Kapil Mayawala (2), Dinesh DeAlwis (2), Brian Topp (2)
(1) Vantage Research (2) MSD
Introduction:
QSP models at various levels of physiological detail can support decision making in various stages of the cancer immunotherapy pipeline, from early discovery to clinical development. We have developed models of immune-mediated tumor killing, calibrated the model to two immune therapies, and used it to predict response to combination therapy. Further, we have used the model to identify key patient characteristics of responders to double immune combination therapy.
Objectives:
* Develop QSP model of appropriate physiological detail to address various questions in a Cancer Immunotherapy drug development program
* Use model to predict characteristics of melanoma patients that benefit from combination of two immune therapies.
Methods:
QSP model development was initiated by extensive review of the public literature describing clinical observations and the mechanistic drivers of immune-mediated tumor killing. The model was first calibrated to a therapy that converted incactive CD8s to active CD8s in the tumor then expanded to include therapy that increases CD8 density in the tumor. The model has five ordinary differential equations that capture the growing tumor, as well as aspects of the immune system such as activated and inactivated cytotoxic CD8 T cells, ‘lumped’ pool of pro-inflammatory (T helper) and anti-inflammatory (T Regulators) cell types. Each Virtual Patient (VP) has five Virtual Tumors (‘target lesions’ for RECIST scoring). Post treatment change in the sum of tumor diameters, as well as stochastic simulation of non-target lesion and new metastatic lesions are used to generate RECISTv1.1 scores. The model parameters are constrained directly by literature when data is available, or constrained to match published clinical data, and these are documented.
Results:
A Virtual Population (VPop) was calibrated to publicly available clinical data on immune monotherapy. The mechanistic tumor dynamic model was calibrated to observed changes in the sum of longest tumor diameter (waterfall plots) such that 20% or more of VPs showed greater than 20% increase in tumor diameter). The stochastic model was then calibrated to reflect greater than 40% of patients with progressive disease (via RECISTv1.1), some despite reduced target tumor mass (due to new metastatic lesions, non-target growth or rebound). Other aspects of published clinical characteristics such as baseline distribution of density of immune cells (Percent Tumor Infiltrating Cells are < 20% of total cells in tumor), baseline tumor size distribution, and change in CD8 cell densities on therapy (median of 2 fold increase from baseline) were also matched. Rates of metastases, within patient correlation of Percent Tumor Infiltrating Cells, killing rate and other parameters that were not constrained directly by data were estimated in the calibration process.
Simulation of combination therapy predicted dramatic reduction in tumor mass typified by waterfall plots but more modest improvements in RECIST scores. For the most part, both immune therapies were predicted to be effective in tumors that were inflamed prior to therapy. Non-inflamed tumors were predicted to be non-responsive to either monotherapy or the combination. However, there is a subset of virtual patients that responded to the combination but neither monotherapy. These VPs were found to have modest infiltration of immune cells, or lower levels of the T helper cells at baseline.
Conclusions:
While there is great excitement around the prospects of double immune therapy, these combinations are unlikely to be effective in “cold” tumors that are not recognized by the immune system. Significant expansion of the responder population may require therapy directed at stimulating an immune response (making tumors “visible” to the immune system) rather than adding to a pre-existing immune response. QSP models can be effectively used for patient stratification in Cancer Immunotherapy combination programs.
References:
[1] Gadkar K et al., A six stage workflow for robust application of systems pharmacology, CPT Pharmacometrics Syst. Pharmacol. 5 (2016), pp. 235-49.
[2] Ribas A et al., Association of Pembrolizumab with Tumor Response and Survival Among patients with advanced Melanoma, JAMA, 315 (2016), pp. 1600-1609.
[3] Hamid O, et al., Safety and tumor responses with lambrolizumab (anti-PD1) in melanoma, N. Engl J Med, 369 (2013), pp. 134-144.
[4] Schachter J et al., Pembrolizumab versus Ipilimumab for advanced melanoma: final overall survival results of a multicentre, randomised, open-label phase 3 study (KEYNOTE-006), Lancet, 390 (2017), pp. 1853-1861
[5] Wolchok JD et al., Ipilimumab monotherapy in patients with pretreated advanced melanoma: a randomised, double blind multcentre, phase 2, dose-ranging study, Lancet Oncol. 11 (2010), pp. 155-164.
[6] Long GV et al., Standard-dose pembrolizumab in combination with reduced-dose ipilimumab for patients with advanced melanoma (KEYNOTE - 029): an open-label, phase - 1b trial, Lancet Oncol., 18 (2017), pp. 1202-1210.