Malvina Marku 1,2, Loic Ysebaert 1,2,4, Etienne Chatelut 1,2,4, Jean Baptiste Woillard 3, Melanie White Koning 2,4
1 Institut Universitaire du Cancer De Toulouse - Oncopole (Toulouse, France), 2 Inserm, Cancer Research Centre of Toulouse (Toulouse, France), 3 University of Limoges (Limoges, France), 4 University of Toulouse (Toulouse, France)
Background: Chronic Lymphocytic Leukaemia (CLL) exhibits heterogeneous therapeutic responses, with a substantial subset of patients developing resistance to standard treatments. Current pharmacokinetic/pharmacodynamic (PK/PD) approaches quantify drug exposure but lack mechanisms to capture downstream molecular adaptations. On the other hand, gene regulatory networks (GRNs) characterise cell-state dynamics and transcriptional regulation [1], offering insights into adaptive resistance at the molecular scale. To date, PK modelling and GRN dynamics have not been integrated into a unified framework to mechanistically link drug exposure with cancer cell regulatory rewiring and dynamic behaviour. In this project, we aim to develop a hybrid PK-GRN modelling framework that quantifies drug exposure and network-level transcriptional responses to ibrutinib therapy in CLL, enabling the prediction of CLL cell behaviour and the identification of regulatory mechanisms driving resistance.
Methods: In our previous study, including 94 CLL patients under ibrutinib, a mechanistic PK/PD model was constructed to describe plasma concentrations of the therapeutic agent, and relate this to lymphocyte counts over time [2], [3]. To characterise intracellular regulatory changes, we analysed longitudinal single-cell RNA sequencing data from one CLL patient’s peripheric blood mononuclear cells (PBMC) sampled at three time points: baseline (M0), response at 3 months (M3), and when the patient showed relapse at 27 months (M27) [4]. Building on these data, we performed various bioinformatics analyses on the CLL compartment, including trajectory inference [5] and GRN inference [6], thus identifying the major cell fate lineages and causal interactions between genes at the single-cell level. To quantify how ibrutinib-induced perturbations propagate through the regulatory landscape, we applied a Random Walk with Restart (RWR) algorithm [7] to the baseline (M0) GRN, thus modelling signal diffusion from ibrutinib-target nodes and estimating the relative reachability of response- and relapse-associated genes within the pre-treatment network topology. Lastly, we combined the inferred GRN with the patient-specific PK profile within a hybrid modelling framework, where the continuous drug profile drives the dynamic modulation of TF activities and downstream regulatory modules. In this exploratory phase, this hybrid model enabled simulation of network adaptation as a function of time-varying ibrutinib exposure.
Results: Trajectory analysis revealed two distinct CLL cell fates emerging under treatment: a responsive lineage enriched at M3 and a resistant lineage predominating at M27. Interestingly, GRN inference showed that the resistant lineage is characterised by increased activity of TFs associated with inflammatory signalling pathways, previously reported in resistance mechanisms in CLL [8]. RWR analysis applied to the M0 network demonstrated that genes specifically upregulated at relapse occupy reachable positions from ibrutinib-perturbed regions of the network. RWR effectively prioritises a strongly connected subset of validated interactions, achieving high early precision (maxPR = 0.5). This suggests that relapse-associated programs are not randomly acquired but are embedded within the baseline regulatory topology and progressively activated under therapeutic pressure.
Lastly, the preliminary hybrid PK-GRN simulations suggested that varying ibrutinib concentration can influence the stability properties of the inferred regulatory network and the relative prevalence of resistant-like attractor states. These results suggest that exposure-dependent network rewiring may quantitatively determine cellular phenotypic composition over time. Given the exploratory nature of these simulations, further validation in a larger patient cohort will be required. Nonetheless, this framework provides a proof-of-concept for integrating pharmacokinetics with regulatory network dynamics.
Conclusions: We present a proof-of-concept PK-GRN framework that links systemic ibrutinib exposure to molecular dynamics in a CLL study, providing a mechanistic bridge between drug concentration and phenotypic adaptation. This framework opens several perspectives including (i) extension of classical PD models by embedding network-defined CLL phenotypes (responsive and resistant states) as mechanistically informed compartments, (ii) identification of early regulatory biomarkers predictive of resistance emergence, and (iii) coupling this hybrid PK-GRN model with Machine Learning algorithms may enable data-driven optimisation of treatment strategies, by identifying exposure profiles that minimise the emergence or stabilisation of resistant attractor states.
References:
[1] M. Marku et al., ‘Data driven network inference and longitudinal transcriptomics unveil dynamic regulation in Chronic Lymphocytic Leukaemia models’, Npj Syst. Biol. Appl., vol. 12, no. 1, p. 24, Jan. 2026, doi: 10.1038/s41540-025-00645-4.
[2] F. Gallais et al., ‘Population PK-PD Modeling of Circulating Lymphocyte Dynamics in Chronic Lymphocytic Leukemia Patients Under Ibrutinib Treatment’, Clin. Pharmacol. Ther., vol. 110, no. 1, pp. 220–228, 2021, doi: 10.1002/cpt.2189.
[3] F. Gallais et al., ‘Population Pharmacokinetics of Ibrutinib and Its Dihydrodiol Metabolite in Patients with Lymphoid Malignancies’, Clin. Pharmacokinet., vol. 59, no. 9, pp. 1171–1183, Sep. 2020, doi: 10.1007/s40262-020-00884-0.
[4] S. Cadot et al., ‘Longitudinal CITE-Seq profiling of chronic lymphocytic leukemia during ibrutinib treatment: evolution of leukemic and immune cells at relapse’, Biomark. Res., vol. 8, no. 1, p. 72, Dec. 2020, doi: 10.1186/s40364-020-00253-w.
[5] M. Lange et al., ‘CellRank for directed single-cell fate mapping’, Nat. Methods, vol. 19, no. 2, pp. 159–170, Feb. 2022, doi: 10.1038/s41592-021-01346-6.
[6] N. Kumar, B. Mishra, M. Athar, and S. Mukhtar, ‘Inference of Gene Regulatory Network from Single-Cell Transcriptomic Data Using pySCENIC’, in Modeling Transcriptional Regulation: Methods and Protocols, S. MUKHTAR, Ed., New York, NY: Springer US, 2021, pp. 171–182. doi: 10.1007/978-1-0716-1534-8_10.
[7] A. Valdeolivas et al., ‘Random walk with restart on multiplex and heterogeneous biological networks’, Bioinformatics, vol. 35, no. 3, pp. 497–505, Feb. 2019, doi: 10.1093/bioinformatics/bty637.
[8] U. Rozovski, M. J. Keating, and Z. Estrov, ‘Targeting inflammatory pathways in chronic lymphocytic leukemia’, Crit. Rev. Oncol. Hematol., vol. 88, no. 3, pp. 655–666, Dec. 2013, doi: 10.1016/j.critrevonc.2013.07.011.
Reference: PAGE 34 (2026) Abstr 12012 [www.page-meeting.org/?abstract=12012]
Poster: Oral: Drug/Disease Modelling - Oncology