Lung adenocarcinoma (LUAD) is associated with a low survival rate at advanced stages. Although the development of targeted therapies has improved outcomes in LUAD patients with identified and specific genetic alterations, such as activating mutations on the epidermal growth factor receptor gene (EGFR), the emergence of tumor resistance eventually occurs in all patients and this is driving the development of new therapies. In this paper, we present the In Silico EGFR-mutant LUAD (ISELA) model that links LUAD patients’ individual characteristics, including tumor genetic heterogeneity, to tumor size evolution and tumor progression over time under first generation EGFR tyrosine kinase inhibitor gefitinib. This translational mechanistic model gathers extensive knowledge on LUAD and was calibrated on multiple scales, including in vitro, human tumor xenograft mouse and human, reproducing more than 90% of the experimental data identified. Moreover, with 98.5% coverage and 99.4% negative logrank tests, the model accurately reproduced the time to progression from the Lux-Lung 7 clinical trial, which was unused in calibration, thus supporting the model high predictive value. This knowledge-based mechanistic model could be a valuable tool in the development of new therapies targeting EGFR-mutant LUAD as a foundation for the generation of synthetic control arms.
Lung cancer is one of the most frequently diagnosed cancers and the leading cause of cancer mortality worldwide1,2. More than 40% of newly diagnosed lung cancers are in a metastatic state3. Based on European and American guidelines (respectively4,5), the main treatment options currently available for patients with lung adenocarcinoma (LUAD)—representing 40% of all lung cancer6, are surgery, radiation therapy, chemotherapy, immunotherapy, and targeted therapy.
Alterations such as gene mutations or fusion lead to uncontrolled receptor tyrosine kinases (RTK) signaling and an oncogenic signal, leading to strong activation of downstream pathways converging on common signaling effectors that elicit tumor development7. Molecularly targeted therapies have markedly improved clinical outcomes in patients with LUAD defined by the detection of oncogenic mutation or fusion in RTK-like epidermal growth factor receptor (EGFR). The EGFR tyrosine kinase inhibitor (TKI) gefitinib was the first targeted therapy for the treatment of advanced EGFR-mutant LUAD approved by both the European Medical Agency (EMA) and the Food & Drug Administration8. However, as all EGFR-mutant LUAD eventually develop resistance to treatment this disease is still one of the most deadly cancers9.
The development of new molecularly targeted therapies comes with a high human, time, and financial cost10,11,12. Yet, a large amount of data and knowledge resulting from biological experiments of last decades at different scales (from the molecular level to the population level) and in various conditions (in vitro cultivated cells, animal experiments, human studies) are now publicly available for integration to support new insights and progress13. Drug development decision making could benefit from being informed and rationalized by the integration of these heterogeneous data. Knowledge-based mechanistic computational models represent a valuable tool to bridge quantitatively experimental data that are heterogeneous in scale and nature. In particular, they provide insights to study the rare mutation combinations, such as KRAS (Kirsten rat sarcoma gene) or BRAF (B rapidly accelerated fibrosarcoma gene) co-occurring with EGFR mutation. They can provide both the dynamics of the biological entities included in the modeling and clinical outputs of patients. Another interest of mechanistic disease models lies in their modularity: models of other treatments can therefore be easily integrated into such models of physiopathological processes14.
Such a model can be used to predict the clinical outcomes of a virtual patient, implemented as a digital twin of a real world patient, in response to distinct sets of treatments, allowing prediction of clinical outcomes of a target population of such patients, based on the corresponding virtual population (Box 1) and can serve as a support for clinical development of new drugs.
Such mechanistic models have been developed in the past for oncology application. For instance, Milberg et al.15 developed a model of detailed anti-tumor immune response in a context of melanoma, including several immune cell interactions linked to tumor diameter evolution. Dogra et al.16 reported a model linking the pharmacokinetics of several treatments on cell cycle progression in triple-negative breast cancer. Others, such as Barber et al.17 or Yu et al.18 used statistical models to link tumor characteristics with the clinical outcome progression-free survival. In 2020, Nagase et al.19 published a Bayesian model tumor radius evolution of EGFR-mutant NSCLC treated with 1st generation TKI. However, to our knowledge, a mechanistic model that targets the same population, and that links key molecular and cellular cancer evolution actors to disease progression and clinical outcomes, as observed in clinics, is still missing.
Based on the recommendations from EMA20 with respect to physiologically based pharmacokinetic modeling and the American Society of Mechanical Engineers (ASME) Verification & Validation (V&V) 40 standard published by the FDA, model development could be summarized in four main steps: (1) definition of model context of use20,21, (2) construction of a knowledge model describing the patho-physiological interplay of biological phenomena within the context of use, (3) implementation of a computational model by translating knowledge model into mathematical equations, (4) calibration of the model parameters to ensure that simulations reproduce expected behaviors observed in the real world.
In addition to the four main steps, we propose a fifth one (5), namely the validation of the model22, in order to assess model credibility by challenging its predictability in reproducing real-world data that were not used to build the model nor to calibrate it.
We present in this article a mechanistic model built based on those guidelines with the additional validation step (fifth step), integrating multiscale phenomena, with a context of use to predict tumor evolution and disease progression over time of EGFR-mutant LUAD patients treated with gefitinib. Other settings such as additional treatments or placebo in humans are deemed out of the scope of this work. We present here the in silico strategy used to build the In Silico EGFR-mutant LUAD (ISELA) model, its validation ensuring the reliability of its prediction and the use of the model to identify individual characteristics linked to clinical outcome.
Visual predictive checks
As a verification criterion of calibration success, as well as correct estimation of parameter values and distribution amongst the population, visual predictive checks were performed on the experimental dataset used for calibration (Figs. 1, 2 and 3).
For the in vitro calibration of the model, the model faithfully reproduced the time at which ERK and AKT proteins reach their maximal activation levels, and the maximal activation observed (following epidermal growth factor (EGF) or hepatocyte growth factor (HGF) stimulation) as shown in Fig. 1.
On in vitro KRAS mutated spheroid simulation, the model output matched the experimental data, both in terms of tumor radius evolution, and maximal depth for cell viability observed in these conditions (Fig. 2a, b). On the mice xenografted with patient-derived tumor (carrying exon 19 deletion mutation, with or without co-occurrence of PIK3CA mutation), either treated with gefitinib or untreated and based on the dynamics observed in literature, the ISELA model reproduces accurately the evolution of tumor volume over time (Fig. 2c–f).
The selected range of acceptable model output variation, materialized by the error bars, was defined as the maximum between two times the associated standard deviations and 20% of the mean. This approach allows coverage of heterogeneous calibration datasets: we were able to constrain the model on datasets composed of one or few numbers of experiments and/or datapoints, as well as datasets lacking standard deviation. Finally, the penalizations were applied to the selected range of variation without assigning specific weight to the mean. To note, in one condition (namely EGFR mutant with placebo), the simulation tended to provide a tumor volume that was slightly lower than the observed one, while remaining in the experimental uncertainty that was huge in this particular setting.
In the human setting, the ISELA model was able to match a realistic TTP for 97% patients included in the calibration process (one virtual patient displaying a higher TTP than their real world counterpart), supporting its reliability with respect to clinical outcome prediction (Fig. 3).
To conclude, calibration constrained the ISELA model by finding a set of parameter values allowing it to represent biological behaviors consistent with data extracted from the literature. Thus these steps increased the credibility of the ISELA model. However, a validation step is needed to formally assess the performance of the model and in particular its prediction capacity in its context of use.
We performed a validation process to assess the reliability of the ISELA model predictions. We aimed to ensure that the model is able to reproduce biological and clinical behaviors extracted from independent clinical datasets that were not used to build nor calibrate the model. In the following, we compare the ISELA model predictions with the data extracted from ref. 23. The model is deemed as successfully validated if it respects the thresholds set in the “Materials and methods” section on raw coverage and bootstrapped LR-test thresholds.
Generation of the virtual population
As explained in the “Materials and methods” section, we generated a virtual population ten times larger than the real population size. Table 1 provides statistical comparison illustrating how close the generated virtual population is from the clinical data, as detailed in “Materials and methods” section.
The Virtual population did not differ significantly from the real population, for any of the compared characteristics: the virtual population generated was therefore representative of the provided Lux-Lung 7 population characteristics. As a result, simulation outputs can rightfully be compared to real world clinical data, as the inputs match.
We then compared the outputs of the simulations to the Lux-Lung 7 inferred TTP to evaluate whether the ISELA model reproduces accurately the Lux-Lung 7 trial.
Comparison of survival curves between simulated and real population
We provided the output of the model using a classical Kaplan–Meier survival curve, defined by the with the survival curve computed on the entire Virtual Population and with its 95% bootstrapped prediction interval (PI), overlaid with the TTP deduced from the Kaplan–Meier curves extracted from ref. 23 (Fig. 4).
As seen in Fig. 4, the ISELA model fulfills both validation criteria detailed in the “Materials and methods” section: 98.5% of experimental data are covered by the model prediction interval and only 0.6% of bootstrapped LR tests are significant (not able to reject the null hypothesis defined as: no difference between observed and simulated populations). These results support that observed and simulated TTP data are not statistically different. The ISELA model is thus considered as validated, as per the initial objective. As a consequence it increases the credibility of both the model predictions, and its matching of the real world population.
Exploration of the individual tumor size evolution
Exploration of model outcomes within the virtual population of the validated ISELA model was performed on the tumor size evolution dynamics.
The tumor radius evolution over time is an output of the ISELA model that was calibrated on in vitro and on mice with success (see Figs. 1 and 2). Tumor radius can be followed every day during the simulation for each individual patient treated with gefitinib, as displayed in Fig. 5. As expected, the vast majority of the patients show a decrease of tumor size at first, followed by a relapse of the tumor which becomes gefitinib-resistant and increases in size, though this relapse time differs among patients (see Fig. 5).
When stratifying the patients based on the co-occurence of the KRAS mutation, it was noted that tumors harboring the KRAS mutation resist gefitinib, as expected24, compared to other tumors. Indeed, when size was compared at 6 months to baseline, different patterns were observed ranging from an increase in tumor size to a decrease in tumor size, patients harboring KRAS mutation being in the first case (Fig. 6).
To go further and identify the key parameters that impact the change in tumor size and the resulting time to progression, we performed a sensitivity analysis on all individual patients characteristics (Fig. 7).
Both analyses on tumor radius and TTP consistently identified the immune system (2 parameters), neo-angiogenesis (1 parameter), tumor initial size (2 parameters), initial size of the resistant subclone (1 parameter), as well as 1 parameter encompassing the impact of implicit mutations on cell proliferation cancer hallmark as critically impactful on both outputs of interest.
The in silico EGFR-mutant lung adenocarcinoma (ISELA) model presented in this paper is a predictive and reliable mechanistic model of tumor growth evolution for patients treated with gefitinib with TTP as primary outcome. This model includes patients’ individual characteristics variability observed in literature. The model was designed with knowledge and data available in public literature. The calibration outcome and the corresponding visual predictive checks show the successful calibration with more than 85% of patients accurately reproduced. Visual predictive checks are a valuable tool widely used in the field of modeling in particular in pharmacodynamics25,26, which helped build credibility for the calibrations performed. To go even further, and to assess the credibility of the ISELA for prediction, a formal validation of the model output with respect to an independent dataset showed agreement with the real world clinical data. These results underline the capacity of the model in predicting tumor progression in a population of patients with EGFR-mutant LUAD treated with gefitinib.
The predictive accuracy of the model has been validated on population-based data extracted from ref. 23, based on two metrics: bootstrapped log-rank tests (more than 99% of tests were negative) and clinical data coverage (more than 98% of coverage was observed). As a consequence of this validation, the ISELA model predictions are deemed reliable on EGFR mutant LUAD patients at the population level, for patients that do not experienced severe toxicity, death, or treatment discontinuation.
One advantage of the mechanistic approach is that each parameter holds a pathophysiology-related meaning: causality between disease-related biological phenomena is inherent to the knowledge-based model, easing the interpretation of the impact of parameter values on clinical outcomes, especially interesting in the context of uncommon populations. Exploration of rare populations was therefore realized, and we compared EGFR mutant LUAD population with and without KRAS mutation: the obtained results were in line with reported knowledge, namely (i) consistency in the population characteristics: KRAS is an uncommon mutation, around 2.5% reported on trials based on a population with EGFR mutant LUAD27 and 2.35% in the virtual population defined earlier; (ii) consistency in the efficacy of gefitinib in KRAS and EGFR mutant LUAD: first-generation EGFR-TKIs, i.e., gefitinib and erlotinib transiently down-regulates also the activity of mutant KRAS and related downstream signaling pathways24; (iii) consistency in TTP of KRAS and EGFR mutant LUAD patients: Patients harboring KRAS mutations are associated with a shorter time to progression during TKI-treatment24. Being able to reproduce behaviors that were not the focus of this study increases the credibility of to the model.
The ISELA model can be further improved. Currently, individual behavior description should be interpreted with caution, as some characteristics were not available at the individual patient level, thus correlations between descriptors were extrapolated from the calibration process. To the best of our knowledge, it is difficult to access individual data on tumor size evolution over time, tumor mutational burden (e.g., number of driver mutations, number of clones in the tumor), and individual patient characteristics (e.g., age, sex). Access to such individual-based data would improve the calibration process and the predictions made at individual level. Sensitivity analysis of the model identified neo-angiogenesis and immune-related phenomena as the two main drivers of TTP progression and tumor size. These parts of the model currently remain phenomenological. However, they could be further detailed as part of the future development of the ISELA model, in order to better study mechanistically how these phenomena impact on clinical outcomes. This would also help to increase the domain of applicability of the model.
The model could be extended to new contexts of use taking into account new mutations or new treatments and thus be adapted to support several drug development lines. One advantage is that the ISELA model was planned and implemented to allow enhancements as scientific knowledge progresses. Both qualitative and quantitative advances can be used. As a consequence, if new relevant information is found regarding the physiopathology, it can be integrated in the existing model, rather than rebuilding a model from scratch. Finally, to further explore the advantages and drawbacks of the ISELA model one could compare it with mathematical models applied to the same context of use: patients with EGFR-mutant lung adenocarcinoma treated with first-generation TKI.
In silico approaches such as the one presented in this article provide tools to overcome frequent issues related to clinical trials: they notably ensure the clinical equipoise by enrolling the exact same virtual patients in control and investigational arms. As a consequence, in silico models supporting drug development can ease the development of new drugs improving the medical care of patients diseases such as LUAD28,29.
Materials and methods
Development of the ISELA model
The ISELA model is a knowledge-based mechanistic model designed to reproduce tumor size evolution and disease progression of virtual patients matching real world patients with EGFR-mutant LUAD treated with gefitinib, as illustrated in Fig. 8. Together, virtual patients form virtual populations (see Box 1). The clinical outcome deemed of interest is the time to progression based on RECIST (Response Evaluation Criteria In Solid Tumors) criteria30. Briefly, this corresponds to an increase of the largest dimension of the tumor by 20% and of at least 0.5 cm.
A thorough review of more than 250 scientific papers was performed to identify the main phenomena to include into our EGFR mutant LUAD physiopathology model: (1) cell proliferation, cell death, layering of cells in the tumor, carrying capacity due to neo-angiogenesis and limited growth due to the immune system impacting tumor growth, (2) impact of individual mutational profile on these pathways, (3) signaling pathways that are downstream of EGFR activation, (4) tumor heterogeneity stemming from groups of cells sharing the same phenotype, namely tumor clones; and (5) resulting clinical outcomes from physiopathology (Supplementary Information). Due to the model modularity, (6) a gefitinib treatment model was added in order to consider its impact on patients’ physiopathology as described in Table 2. The associated knowledge was validated based on thoracic oncology scientific expertise. This allowed us to uncover the knowns and unknowns of the target population characteristics, and define appropriate simplifying assumptions when needed. This biological knowledge was converted into mathematical equations (ordinary differential equations—ODEs) to computationally model the corresponding biological phenomena. These equations were implemented as groups of mechanistically related equations, or submodels, and the integrated combination is the ISELA model.
The ISELA model accounts for the heterogeneity in tumor with the modeling of a number of tumor clones with distinct genetic background from one patient to another. From a computational point of view, this is ensured by the duplication of each part of the model corresponding to phenomena occuring at the clone level. Namely, the following phenomena are duplicated to account for clone heterogeneity: tumor growth, mutational profile, EGFR, and downstream signaling pathways. As a consequence, depending on the number of clones each virtual patient carries in their tumor, the model runs with a set of 27 to 97 variables, 108 to 258 parameters, and 13 to 83 ODEs (ranges for 2 to 16 clones and associated duplication), as indicated in Table 2. The model structure is illustrated in Fig. 9, and the equations of the model are provided in Supplementary Information.
As indicated in Fig. 9, two model outputs are considered as clinical endpoints: tumor radius and time to progression, deduced from the tumor radius. Yet, the model does not consider censoring due to toxicity, death, or treatment discontinuation. Patients who did not display tumor progression at the follow-up cut-off, that is to say at the end of the simulation, may be considered as being right censored.
Following the model development detailed in the previous section, the model was calibrated as advised in the EMA guidelines20 and the V&V 4021. Calibration aims to find parameter values and distributions such that the model reproduces expected behaviors observed in the real world. It is the first step to ensure the accuracy of a mechanistic model and is performed prior to the validation process. We here describe the calibration protocol applied to the ISELA model, based on the data we found in literature. The corresponding calibration process is composed of successive steps, and each step has as its objective a specific model variable behavior matching one or more specific computational constraints. Since calibration steps are executed sequentially, the first calibration steps are prerequisites for the following steps. They take into account both quantitative and qualitative constraints, to consider and reproduce the heterogeneous and multi-scale data extracted from literature31 details the two first steps of the process, where we aligned the ISELA model with:
In addition, we performed two additional calibration steps to increase the context of use of the ISELA model to humans They both focus on finding values of the parameters related to human neo-angiogenesis, immune system, and treatment-resistant clones to reproduce the time to progression (TTP; i.e., duration between start of the treatment administration and detection of tumor progression) of patients found on literature:
We reproduced the population-level clinical data deduced from the NEJ002 trial46, and deduced correlations between patient descriptors. The extraction of the list of time-to-events (for both PFS and OS) was realized using R package digitize47, using the input survival times from graph reading; and the reported number at risk. TTP was inferred based on the clinical trial PFS and OS, as detailed in ref. 22. Therefore, the NEJ002 TTP dataset was deduced from the lists of time-to-events corresponding to the PFS and OS of the Maemondo/NEJ002 trial. Under the hypothesis that patients who died before disease progression are characterized by the same time to event in the PFS and OS sets, we filtered out PFS events that correspond to patients’ death, leaving only the time-to-events corresponding to disease progression.
These two last steps were not intended to reproduce tumor size evolution over time (as done in steps 1 and 2) since in vivo tumor sizes are rarely reported in the literature in humans. Instead, the goal was to reproduce the TTP, computed from the evolution of the time tumor progression, according to the RECIST criteria. The experimental data that were used in these four calibration steps are listed in Table 3.
To use the same model structure in all settings (in vitro, in mouse, and in human simulations), allometric scaling was used, as described in ref. 31. In a nutshell, allometry theory refers to the impact of the size of living creatures on their characteristics such as morphological and physiological traits. In this paper, we used a common scaling law with the following relationship Z = a × Mb with Z the studied characteristic, M the organism mass, and a and b parameters called allometric coefficient and allometric exponent, respectively48,49. As reference weight for in vitro and mice, we used 2.63 g50 and 23 g51, respectively.
Visual predictive checks are performed as a verification criterion of calibration success (see “Results” section).
As detailed in the next two sections, model validation was assessed on the Lux-Lung 7 clinical dataset, and based on the simulation of a virtual population matching its characteristics.
The Lux-Lung 7 trial (with PFS and OS reported by ref. 23) was selected for three reasons:
The characteristics of the patients enrolled in the trial corresponds to the specified context of use of the ISELA model,
The treatment they received (gefitinib) is consistent with the context of use of the model,
The dataset was neither used for building nor calibrating the ISELA model.
These are reported in Table 4.
To be able to compare the ISELA model TTP to the LUX-LUNG 7 dataset, the disease progression endpoint was similarly derived from clinical PFS and OS, as explained in ref. 22 and detailed in the calibration context. Therefore, the Lux-Lung 7 TTP dataset was deduced from the lists of times-to-event corresponding to the PFS and OS of the Lux-Lung 7 trial. The comparison of OS, PFS, and TTP is provided on Fig. 10. In the absence of information about which patients did not display tumor progression (and died without detectable progression), we assume that the distribution of patients characteristics is not altered in the subsets of patients who displayed tumor progression.
Virtual population generation and statistical analyses for validation
The protocol described in Table 5 was applied to compare the simulated output with the clinical data set.
The data processing and analysis were performed within R Software, version 3.6.1 or above. In particular, we used the following packages: survival, survminer, tidyr, data.table, jsonlite, and ggplot2.
We chose to perform sensitivity analysis based on a tornado approach. In a nutshell, a population of 5000 virtual patients was generated, based on the characteristics of the general population (Supplementary Information), a 50/50 proportion of EGFR mutations (exon 19 deletions and exon 21 L858R point mutation) was used. For each patient characteristic, patients were split in two categories: low value (those with value lower than the median) and high value (those with value higher than the median) and the median output of interest was computed for each category. The value used for the comparison of all parameters is the difference between the median output of interest in the complete virtual population minus the median output of interest in each of these two categories. The resulting values are plotted in tornado plots. The advantage of such an analysis is that it does not rely on statistical hypotheses on the distribution of the impact of the parameter on the output of interest.
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
The model structure, model documentation, and data supporting the conclusions of this study are available on the jinko.ai platform upon request from the corresponding author, C.M.
Model documentation and source code files for submodels are provided in Supplementary Information.
Bradley, S. H., Kennedy, M. P. T. & Neal, R. D. Recognising lung cancer in primary care. Adv. Ther. 36, 19–30 (2018).
Wang, H. et al. Global, regional, and national life expectancy, all-cause mortality, and cause-specific mortality for 249 causes of death, 1980–2015: a systematic analysis for the global burden of disease study 2015. Lancet 388, 1459–1544 (2016).
Maity, S., Pai, K. S. R. & Nayak, Y. Advances in targeting EGFR allosteric site as anti-NSCLC therapy to overcome the drug resistance. Pharmacol. Rep. 72, 799–813 (2020).
Planchard, D. et al. Metastatic non-small cell lung cancer: ESMO clinical practice guidelines for diagnosis, treatment and follow-up. Ann. Oncol. 29, iv192–iv237 (2018).
NCCN. Non-Small Cell Lung Cancer Metastatic. NCCN Guidelines for Patients. https://www.nccn.org/patients/guidelines/content/PDF/lung-metastatic-patient.pdf (2022).
ESMO. Non-small-cell lung cancer (NSCLC). ESMO Patient Guide Series -ESMO Clinical Practice Guidelines. https://www.esmo.org/for-patients/patient-guides/non-small-cell-lung-cancer (2019).
Du, Z. & Lovly, C. M. Mechanisms of receptor tyrosine kinase activation in cancer. Mol. Cancer 17, https://doi.org/10.1186/s12943-018-0782-4 (2018).
FDA. Prescribing information for IRESSA (gefitinib). https://www.accessdata.fda.gov/drugsatfda_docs/label/2018/206995s003lbl.pdf (2015).
Huang, L. & Fu, L. Mechanisms of resistance to EGFR tyrosine kinase inhibitors. Acta Pharm. Sin. B 5, 390–401 (2015).
DiMasi, J. A., Hansen, R. W. & Grabowski, H. G. The price of innovation: new estimates of drug development costs. J. Health Econ. 22, 151–185 (2003).
Martin, L., Hutchens, M., Hawkins, C. & Radnov, A. How much do clinical trials cost? Nat. Rev. Drug Discov. 16, 381–382 (2017).
Kuepfer, L., Lippert, J. & Eissing, T. in Advances in Experimental Medicine and Biology 543–561 (Springer New York, 2011).
Given, L. S. et al. Comprehensive cancer control in the US: 20 years of progress. Cancer Causes Control 29, 1151–1161 (2018).
Eissing, T. A computational systems biology software platform for multiscale modeling and simulation: Integrating whole-body physiology, disease biology, and molecular reaction networks. Front. Physiol. 2, https://doi.org/10.3389/fphys.2011.00004 (2011).
Milberg, O. et al. A QSP model for predicting clinical responses to monotherapy, combination and sequential therapy following CTLA-4, PD-1, and PD-l1 checkpoint blockade. Sci. Rep. 9, 11286 (2019).
Dogra, P. et al. Translational modeling identifies synergy between nanoparticle-delivered miRNA-22 and standard-of-care drugs in triple-negative breast cancer. Pharm. Res. 39, 511–528 (2022).
Barber, P. R. et al. Predicting progression-free survival after systemic therapy in advanced head and neck cancer: Bayesian regression and model development. eLife 11, e73288 (2022).
Yu, J., Wang, N. & Kågedal, M. A new method to model and predict progression free survival based on tumor growth dynamics. CPT: Pharmacomet. Syst. Pharmacol. 9, 177–184 (2020).
Nagase, M., Aksenov, S., Yan, H., Dunyak, J. & Al-Huniti, N. Modeling tumor growth and treatment resistance dynamics characterizes different response to gefitinib or chemotherapy in non-small cell lung cancer. CPT: Pharmacomet. Syst. Pharmacol. 9, 143–152 (2020).
EMA. EMA guidelines. EMA guidelines. https://www.ema.europa.eu/en/reporting-physiologically-based-pharmacokinetic-pbpk-modelling-simulation (2018).
Kuemmel, C. et al. Consideration of a credibility assessment framework in model-informed drug development: potential application to physiologically-based pharmacokinetic modeling and simulation. CPT: Pharmacomet. Syst. Pharmacol. 9, 21–28 (2019).
Jacob, E. et al. Empirical methods for the validation of time-to-event mathematical models taking into account uncertainty and variability: application to EGFR+ lung adenocarcinoma. bioRxiv (2023).
Paz-Ares, L. et al. Afatinib versus gefitinib in patients with EGFR mutation-positive advanced non-small-cell lung cancer: overall survival data from the phase IIb LUX-lung 7 trial. Ann. Oncol. 28, 270–277 (2017).
Santoni-Rugiu, E. et al. Intrinsic resistance to EGFR-tyrosine kinase inhibitors in EGFR-mutant non-small cell lung cancer: Differences and similarities with acquired resistance. Cancers 11, 923 (2019).
Holford, N. The Visual Predictive Check: superiority to standard diagnostic (Rorschach) plots. in 14th Meeting of the Population Approach Group in Europe. https://www.researchgate.net/publication/238684965_The_Visual_Predictive_Check_Superiority_to_Standard_Diagnostic_Rorschach_Plots (2005).
Post, T. M., Freijer, J. I., Ploeger, B. A. & Danhof, M. Extensions to the visual predictive check to facilitate model performance evaluation. J. Pharmacokinet. Pharmacodyn. 35, 185–202 (2008).
Skoulidis, F. & Heymach, J. V. Co-occurring genomic alterations in non-small-cell lung cancer biology and therapy. Nat. Rev. Cancer 19, 495–509 (2019).
Popat, S. et al. Addressing challenges with real-world synthetic control arms to demonstrate the comparative effectiveness of pralsetinib in non-small cell lung cancer. Nat. Commun. 13, https://doi.org/10.1038/s41467-022-30908-1 (2022).
Davi, R. et al. CLRM-09. incorporating external control arm in mdna55 recurrent glioblastoma REGISTRATION TRIAL. Neuro-Oncol. Adv. 3, iv3–iv3 (2021).
Schwartz, L. H. et al. RECIST 1.1—update and clarification: from the RECIST committee. Eur. J. Cancer 62, 132–137 (2016).
Palgen, J.-L. et al. Integration of heterogeneous biological data in multiscale mechanistic model calibration: application to lung adenocarcinoma. Acta Biotheor. 70, https://doi.org/10.1007/s10441-022-09445-3 (2022).
Schoeberl, B., Eichler-Jonsson, C., Gilles, E. D. & Müller, G. Computational modeling of the dynamics of the MAP kinase cascade activated by surface and internalized EGF receptors. Nat. Biotechnol. 20, 370–375 (2002).
Aoki, K., Yamada, M., Kunida, K., Yasuda, S. & Matsuda, M. Processive phosphorylation of ERK MAP kinase in mammalian cells. Proc. Natl Acad. Sci. USA 108, 12675–12680 (2011).
Nakakuki, T. et al. Topological analysis of MAPK cascade for kinetic ErbB signaling. PLoS ONE 3, e1782 (2008).
Guha, U. et al. Comparisons of tyrosine phosphorylated proteins in cells expressing lung cancer-specific alleles of egfr and kras. Proc. Natl Acad. Sci. USA 105, 14112–14117 (2008).
Jagiella, N., Müller, B., Müller, M., Vignon-Clementel, I. E. & Drasdo, D. Inferring growth control mechanisms in growing multi-cellular spheroids of NSCLC cells from spatial-temporal image data. PLOS Comput. Biol. 12, e1004412 (2016).
Ekert, J. E. et al. Three-dimensional lung tumor microenvironment modulates therapeutic compound responsiveness in vitro – implication for drug development. PLoS ONE 9, e92248 (2014).
Freyer, J. P. Role of necrosis in regulating the growth saturation of multicellular spheroids. Cancer Res. 48, 2432–2439 (1988).
Kang, H. N. et al. Establishment of a platform of non-small-cell lung cancer patient-derived xenografts with clinical and genomic annotation. Lung Cancer 124, 168–178 (2018).
Asahina, H. et al. A phase II trial of gefitinib as first-line therapy for advanced non-small cell lung cancer with epidermal growth factor receptor mutations. Br. J. Cancer 95, 998–1004 (2006).
Yang, C.-H. et al. Specific egfr mutations predict treatment outcome of stage IIIB/IV patients with chemotherapy-naive non–small-cell lung cancer receiving first-line gefitinib monotherapy. J. Clin. Oncol. 26, 2745–2753 (2008).
Wu, J.-Y. et al. Lung cancer with epidermal growth factor receptor exon 20 mutations is associated with poor gefitinib treatment response. Clin. Cancer Res. 14, 4877–4882 (2008).
Vasconcelos, P. E. et al. EGFR-a763_y764insfqea is a unique exon 20 insertion mutation that displays sensitivity to approved and in-development lung cancer EGFR tyrosine kinase inhibitors. JTO Clin. Res. Rep. 1, 100051 (2020).
Yasuda, H. et al. Structural, biochemical, and clinical characterization of epidermal growth factor receptor (EGFR) exon 20 insertion mutations in lung cancer. Sci. Transl. Med. 5, 216ra177–216ra177 (2013).
Sugio, K. et al. Prospective phase ii study of gefitinib in non-small cell lung cancer with epidermal growth factor receptor gene mutations. Lung Cancer 64, 314–318 (2009).
Maemondo, M. et al. Gefitinib or chemotherapy for non–small-cell lung cancer with mutated EGFR. N. Engl. J. Med. 362, 2380–2388 (2010).
Wei, Y. & Royston, P. Reconstructing time-to-event data from published Kaplan-Meier curves. Stata J 17, 786–802 (2017).
Pérez-García, V. M. et al. Universal scaling laws rule explosive growth in human cancers. Nat. Phys. 16, 1232–1237 (2020).
Smil, V. Laying down the law. Nature 403, 597–597 (2000).
West, G. B., Woodruff, W. H. & Brown, J. H. Allometric scaling of metabolic rate from molecules and mitochondria to cells and mammals. Proc. Natl Acad. Sci. USA 99, 2473–2478 (2002).
Vellers, H. L., Letsinger, A. C., Walker, N. R., Granados, J. Z. & Lightfoot, J. T. High fat high sugar diet reduces voluntary wheel running in mice independent of sex hormone involvement. Front. Physiol. 8, (2017).
Park, K. et al. Afatinib versus gefitinib as first-line treatment of patients with EGFR mutation-positive non-small-cell lung cancer (LUX-lung 7): a phase 2b, open-label, randomised controlled trial. Lancet Oncol. 17, 577–589 (2016).
Jamal-Hanjani, M. et al. Tracking the evolution of non-small-cell lung cancer. N. Engl. J. Med. 376, 2109–2121 (2017).
Re, M. D. et al. Understanding the mechanisms of resistance in EGFR-positive NSCLC: from tissue to liquid biopsy to guide treatment strategy. Int. J. Mol. Sci. 20, 3951 (2019).
Ma, C., Wei, S. & Song, Y. T790M and acquired resistance of EGFR TKI: a literature review of clinical reports. J. Thorac. Dis. 3, 10–18 (2011).
Yang, F. et al. Relationship between tumor size and disease stage in non-small cell lung cancer. BMC Cancer 10. https://doi.org/10.1186/1471-2407-10-474 (2010).
We would like to thank Janssen-Cilag France for supporting the project. We thank M. Margreiter for his valuable participation in the early phases of this project.
The authors declare the following financial competing interests: A.L.H., J.L.P., A.P.M., E.P., E.J., J.B., R.K., N.C., and C.M. are employed by Novadiscovery SA. R.T. and L.L. are employed by Janssen-Cilag. M.D. reports receipt of honoraria for academic/accredited talks from AstraZeneca, Guardant Health, MSD Oncology, BMS, Takeda. M.D. reports membership of an advisory board and consultancy for AstraZeneca, MSD Oncology, BMS, Pfizer, Roche, Takeda, Boehringer Ingelheim, Janssen Oncology, Amgen, AbbVie, Elevation Oncology, Eli Lilly. M.D. reports receipt of research grants (funds to institution) from Eli Lilly and Nanostring.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
L’Hostis, A., Palgen, JL., Perrillat-Mercerot, A. et al. Knowledge-based mechanistic modeling accurately predicts disease progression with gefitinib in EGFR-mutant lung adenocarcinoma. npj Syst Biol Appl 9, 37 (2023). https://doi.org/10.1038/s41540-023-00292-7