Quantitative systems pharmacology model predictions for efficacy of atezolizumab and nab-paclitaxel in triple-negative breast cancer

Abstract

Background: Immune checkpoint blockade therapy has clearly shown clinical activity in patients with triple-negative breast cancer, but less than half of the patients benefit from the treatments. While a number of ongoing clinical trials are investigating different combinations of checkpoint inhibitors and chemotherapeutic agents, predictive biomarkers that identify patients most likely to benefit remains one of the major challenges. Here we present a modular quantitative systems pharmacology (QSP) platform for immuno-oncology that incorporates detailed mechanisms of immune-cancer cell interactions to make efficacy predictions and identify predictive biomarkers for treatments using atezolizumab and nab-paclitaxel.

Methods: A QSP model was developed based on published data of triple-negative breast cancer. With the model, we generated a virtual patient cohort to conduct in silico virtual clinical trials and make retrospective analyses of the pivotal IMpassion130 trial that led to the accelerated approval of atezolizumab and nab-paclitaxel for patients with programmed death-ligand 1 (PD-L1) positive triple-negative breast cancer. Available data from clinical trials were used for model calibration and validation.

Results: With the calibrated virtual patient cohort based on clinical data from the placebo comparator arm of the IMpassion130 trial, we made efficacy predictions and identified potential predictive biomarkers for the experimental arm of the trial using the proposed QSP model. The model predictions are consistent with clinically reported efficacy endpoints and correlated immune biomarkers. We further performed a series of virtual clinical trials to compare different doses and schedules of the two drugs for simulated therapeutic optimization.

Conclusions: This study provides a QSP platform, which can be used to generate virtual patient cohorts and conduct virtual clinical trials. Our findings demonstrate its potential for making efficacy predictions for immunotherapies and chemotherapies, identifying predictive biomarkers, and guiding future clinical trial designs.

Trial registration: ClinicalTrials.gov NCT02425891.

Keywords: breast neoplasms; computational biology; immunotherapy; systems biology; tumor microenvironment.

Figure 1

QSP model diagram (A) and workflow (B). The model is composed of four compartments: central, peripheral, tumor, and tumor-draining lymph node, which describe cycles of immune activation in lymph nodes, T cell trafficking to the tumor, killing of cancer cells, immune evasion, and antigen release and lymphatic transport. ARG-I, arginase I; AT, activated T cell; MAPC, mature antigen presenting cell; NO, nitric oxide; QSP, quantitative systems pharmacology; NT, naïve T cell; TEFF, effector T cell; TH, T helper cell; Treg, regulatory T cell. Modified from ref .

Figure 2

Rate of response (left) and the best overall response (right) in model-predicted tumor diameter of 100 randomly selected virtual patients. Response is assessed by RECIST V.1.1 in atezolizumab monotherapy (A), nab-paclitaxel group (B), and atezolizumab+nab-paclitaxel group (C). Median (thick lines) and individual (thin line) rate of response are shown in PD (red), SD (purple), and PR/CR (blue) subgroups. CR, complete response; nab, nanoparticle albumin-bound; PD, progressive disease; PR, partial response; SD, stable disease.

Figure 3

Pretreatment distributions of potential predictive biomarkers in responders and non-responders. Statistical significance is calculated by Wilcoxon test. Atezo, atezolizumab monotherapy 1200 mg every 3 weeks; Combo, atezolizumab 840 mg every 2 weeks+nab-paclitaxel 100 mg/m² Q3/4W; MDSC, myeloid-derived suppressor cell; Nab-P, nab-paclitaxel 100 mg/m² Q3/4W; NR, non-responders; R, responder.

Figure 4

Subgroup analysis of the combination therapy in virtual patient cohort. The total 900 virtual patients are divided into eight subgroups based on the pretreatment values of selected biomarkers, and the objective response rates in each subgroup are calculated with 95% Agresti-Coull CIs. MDSC, myeloid-derived suppressor cell.

Figure 5

ROC analysis of potential predictive biomarkers in combination therapy. Cut-off values are selected based on the range of PD-L1 molecules on APCS, pretreatment effector T cell density, tumor mutational burden, and Teff to regulatory T cell ratio. For each cut-off value, response status (R vs NR) is predicted for each virtual patient by comparing the pretreatment amount of the potential predictive biomarker to the cut-off value. Sensitivity (true positive rate) is plotted against 1 − specificity (true negative rate) for each biomarker. APCs, antigen-presenting cells; AUC, areas under curve; NR, non-responders; R, responders; ROC, receiver operating characteristic.

Figure 6

Effects of parameters on objective response. For each parameter of interest, 900 virtual patients are sorted by the parameter values in ascending order and evenly divided into nine subgroups. The response status of each subgroup in the combination therapy is plotted against the corresponding median parameter values. MDSC, myeloid-derived suppressor cells.

Figure 7

Model simulation of sequential therapies using various nab-paclitaxel doses and schedules. Top row (A–C) represents the median tumor volume after 8 weeks of each dose regimen; middle row (D–F) represents the median CD8⁺ T cell density at week 8; and bottom row (G–I) represents the median Treg density in the tumor at week 8. Administration of nab-paclitaxel starts on day 1 (A, D,G), week 2 (B, E, H), and week 4 (C, F, I) on reaching initial tumor diameter.