In silico clinical trials of BiTE expression by oncolytic viruses reveal the impact of patient heterogeneity on dosage protocol

Problem Statement
Bispecific T-cell engagers (BiTEs) represent a promising immunotherapy that redirects T cells to target cancer cells. However, their clinical application, particularly against solid tumors, is hindered by limited bioavailability and severe systemic toxicities. Oncolytic viruses (OVs) offer a potential solution by serving as vectors to deliver BiTEs locally to the tumor microenvironment (TME), thereby reducing systemic exposure. Specifically, oncolytic measles viruses encoding BiTEs (MV-BiTE) have demonstrated safety and efficacy in murine models. Despite this preclinical success, it remains unknown whether the observed efficacy will translate to a heterogeneous human population. A critical gap exists in understanding how patient-to-patient variability affects treatment outcomes and whether optimal dosage protocols can be identified to maximize efficacy across diverse individuals.

Methodology
The authors developed an in silico clinical trial framework using a system of ordinary differential equations (ODEs) to model the dynamics of MV-BiTE therapy. The mathematical model tracks five key populations: uninfected tumor cells ($U$), dead cells/debris ($D$), oncolytic virus ($V$), BiTE concentration ($B$), and effector CD3+ T cells ($T$). The model incorporates Gompertzian tumor growth, viral infection kinetics (using Michaelis-Menten functions to account for spatial limitations), and T-cell mediated apoptosis enhanced by BiTEs.

To address the lack of human MV-BiTE clinical data, the authors employed a two-pronged approach:

1. Parameterization: Model parameters were estimated by fitting to in vitro data (viral replication kinetics, cell viability, and specific lysis assays) and in vivo data from a syngeneic mouse model (C57BL/6J mice with B16-CD20-CD46 tumors).
2. Virtual Patient Generation: Two methods were used to generate virtual cohorts representing patient heterogeneity:
   • Method 1: Acceptance-rejection sampling to match individual mouse tumor growth trajectories, generating 400 parameter sets per mouse to capture variability in viral infectivity and immune response.
   • Method 2: Utilizing data from a Phase II clinical trial of Talimogene laherparepvec (T-VEC), an oncolytic herpes simplex virus, as a surrogate. The authors converted clinical trial "change from baseline" (CFB) tumor measurements into a probability density function (PDF) and used a genetic algorithm to optimize model parameters (specifically tumor growth rate $r$, T-cell killing rate $k$, and BiTE decay rate $d_B$) to match the human CFB distribution.

Sensitivity analyses (local and global) were conducted to identify key drivers of model output, and various hypothetical treatment protocols (varying injection frequency and dosage size) were simulated to test their efficacy.

Key Results

• Model Validation: The ODE system successfully recapitulated in vitro and in vivo dynamics, with high goodness-of-fit statistics ($R^2 > 0.94$) for tumor volume and lysis data.
• Drivers of Heterogeneity: Sensitivity analysis and cohort clustering revealed that the primary drivers of heterogeneous treatment responses are the underlying effector T-cell killing rate ($k$) and BiTE pharmacokinetics (specifically the decay/clearance rate $d_B$).
• Cohort Classification: Virtual patients were clustered into three response groups: regression, growth, and delayed growth. The "growth" cohort was characterized by low basal T-cell killing efficacy and specific correlations between viral infectivity and BiTE clearance.
• Dosage Protocol Optimization:
  • For patients classified as "non-responders" (those with tumor growth under standard protocols), the simulations predicted that more frequent administration of lower dosages would significantly improve outcomes.
  • Specifically, protocols involving an initial dose followed by frequent, smaller booster injections (e.g., every 3.5 days or weekly) outperformed standard protocols with larger, less frequent doses (e.g., every 2-3 weeks) in reducing final tumor burden for the growth cohort.
  • Patients in the "regression" cohort responded well to MV-BiTEs regardless of the specific administration protocol.
• Human Surrogate Application: When applied to the T-VEC surrogate data, the model confirmed that initial tumor burden alone could not explain the clinical trial's variability; patient-specific parameters (growth rate, immune response, and BiTE clearance) were necessary to match the observed CFB distribution. The model predicted that individuals who would be non-responders under the standard T-VEC protocol could benefit from the optimized, frequent low-dose regimen.

Significance and Claims
The paper claims that in silico clinical trials serve as a vital, cost-effective tool for generating hypotheses regarding novel therapeutics before human trials. The study demonstrates that mathematical modeling can:

1. Capture the impact of patient heterogeneity on treatment efficacy.
2. Identify specific biological parameters (T-cell killing rates and BiTE pharmacokinetics) that determine whether a patient will respond to MV-BiTE therapy.
3. Guide the design of treatment schedules, suggesting that for non-responders, shifting from standard bolus dosing to frequent, lower-dose administrations may improve clinical outcomes.

The authors emphasize that while the model provides robust predictions, these are hypothesis-generating. They explicitly state that the use of T-VEC data as a surrogate for MV-BiTE is a limitation and that further work, including dedicated human clinical trials and in vivo validation of the proposed dosing protocols, is required to verify these results. The work aims to inform future experimental design and the eventual clinical translation of OV-BiTE therapies.