Introduction: Oncology drug development relies heavily on in vivo models such as patient derived xenografts (PDX) for preclinical drug efficacy testing. However, PDX models are limited by long tumor establishment times, high engraftment failure rates, high experimental costs [1] and arbitrary drug dosing. Thus, developing a systematic pipeline to prioritize the most efficacious drugs at optimal doses for testing will substantially save experimental costs. The Genomics of Drug Sensitivity in Cancer (GDSC) database [2] catalogues a large database of cancer cell lines tested against cancer drugs in the form of large-scale in vitro drug screens, and has been used to find novel drug targets, but has yet to be used to improve PDX study design. A comprehensive database of PDX tumor growth studies also exists in the Mouse Models of Human Cancer database (MMHCdb) [3]. We thus, aim to develop a pipeline to link in vitro drug sensitivity data (GDSC) to in vivo PDX outcomes (MMHCdb). This was done using pharmacokinetic (PK) indices, a method of combining PK profiles of drugs with the half-maximal inhibitory concentration (IC50) of in vitro drug sensitivity screens. Being able to predict PDX outcomes from in vitro data and the accompanying in vivo drug dose will help us to better prioritize effective drugs and the optimal doses to be tested in PDX mice.
Methods: Tumor volume data from the two largest cancer subtypes; lung adenocarcinoma (LUAD) and triple-negative breast cancer (TNBC) was extracted from MMHCdb. In vivo drug efficacy was calculated as Tumor Growth inhibition (TGI), the percentage reduction in tumor volume in treated versus control groups. In vitro IC50 values were compiled from GDSC and relevant cell lines of the same cancer subtype to PDX were identified. Cell lines without an IC50 within GDSC screening concentrations were discarded due to unreliable IC50 estimates. When a single match had multiple cell lines and PDX models, median values for both groups were calculated. One- and two- compartment PK models for studied drugs cisplatin, cyclophosphamide, docetaxel and doxorubicin were developed using in vivo mouse PK data from literature and verified using visual predictive check. Unless otherwise stated, all datapoints were derived using WebPlotDigitizer. PK indices were calculated by simulating PDX drug dosing regimens and adding the relevant median IC50 as a threshold. PK indices included the area under the effect curve (AUEC) [4], the ratio of area under the concentration-time curve to median IC50 (AUC/IC50) and the duration of drug exposure above median IC50 (Time>IC50) and were correlated with TGI via Pearson correlation.
Results: A total of 173 in vitro IC50 values with matching drugs and PDX models were identified from GDSC. A one-compartment PK model was developed for cyclophosphamide, while two-compartment PK models were developed for cisplatin, docetaxel and doxorubicin. All models had linear elimination. Among the PK indices AUEC performed the best in correlating with TGI (R² = 0.739), followed by AUC/IC50 (R² = 0.665) and Time>IC50 (R² = 0.615), with median IC50 exhibiting the weakest correlation with drug efficacy (R² = 0.436). AUEC was estimated to be 558.54 and 88.47 to achieve complete response (95% TGI) or partial response (50% TGI) in PDX respectively, based on the response evaluation criteria in solid tumors (RECIST) criteria defined by the MMHCdb.
Conclusion: Using public databases, GDSC and MMHCdb, we were able to demonstrate AUEC as a useful PK index to predict in vivo drug efficacy in PDX models and suggest the relevant drug exposure level required to achieve either complete or partial response. Thus, this pipeline allows us to prioritize efficacious drugs and optimize the drug doses to be tested in PDX studies, thereby saving time and costs in oncology drug development.
References
1. Y. Liu et al., Signal Transduction and Targeted Therapy. 8 (2023).
2. W. Yang et al., Nucleic Acids Research. 41, D955–D961 (2012).
3. D. M. Krupke et al., Cancer Research. 77, e67–e70 (2017).
4. C. Chen, S. M. Lavezzi, L. Iavarone, CPT Pharmacometrics & Systems Pharmacology. 11, 1029–1044 (2022).