The Impact of QSP Modeling on the Design and Optimization of Gene Therapy Approaches

ABSTRACT

Quantitative Systems Pharmacology (QSP) is increasingly utilized to support the design and translation of gene therapies. This perspective outlines the application of QSP modeling across three domains of gene therapy: mRNA‐based therapeutics, adeno‐associated virus (AAV) vectors, and genome editing systems. We highlight opportunities for dose optimization, biomarker interpretation, and mechanistic understanding, while addressing current limitations in model generalizability, data sparsity, and translational relevance. Examples include QSP platforms for lipid nanoparticle (LNP)‐delivered mRNA, physiologically based pharmacokinetics (PBPK)‐informed AAV biodistribution models, and CRISPR‐Cas9‐based editing systems. These case studies demonstrate QSP's value in de‐risking development and personalizing therapies for rare and complex diseases.

1. Introduction

Quantitative Systems Pharmacology (QSP) modeling is increasingly recognized as a tool to address challenges of rare diseases like inborn errors of metabolism and hematological disorders, with its ability to inform personalized dosing and trial design [1] by integrating patient‐specific factors like genetic mutations and biomarkers [2]. For personalized dosing, a QSP model for acid sphingomyelinase deficiency (ASMD), a rare lysosomal storage disorder, was explored to guide enzyme replacement therapy with olipudase alfa and to link pathophysiology to pharmacology [3].

In the development of gene therapy modalities, QSP has the potential of offering mechanistic insights that bridge preclinical findings and clinical outcomes. In this perspective, we highlight the unique opportunities and challenges of applying QSP modeling across three key areas of gene therapy innovation. We first discuss ribonucleic acid (RNA) therapeutics and messenger RNA (mRNA)‐based vaccines, notably those utilizing lipid nanoparticle (LNP) delivery systems, where QSP can enhance understanding of intracellular dynamics and immunogenicity. We then focus on AAV and other viral vector‐based gene therapies, where modeling can inform vector biodistribution, expression kinetics, and durability of effect. Finally, we explore gene editing technologies, such as clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR‐associated protein 9 (Cas9) delivered via LNPs, emphasizing how QSP can support optimizing editing efficiency and minimizing off‐target effects.

2. RNA Therapeutics and Vaccines

Quantitative Systems Pharmacology (QSP) has been increasingly applied to mRNA vaccines and therapeutics, providing frameworks that can be translated into rare disease contexts where empirical data are especially limited. Although few QSP models have been developed directly for rare disease mRNA vaccines, recent work in broader populations provides important methodological advances that can be repurposed for gene therapy in small, vulnerable cohorts.

For example, Selvaggio et al. [4] introduced one of the earliest mechanistic QSP models for mRNA vaccines, capturing early post‐injection events such as cellular uptake, antigen translation, and antigen presentation to identify design parameters most strongly influencing immune responses. Dasti et al. [5] extended this into a multiscale QSP framework, linking molecular‐level processes (mRNA internalization, endosomal escape, antigen translation) with tissue‐level immune dynamics, and successfully calibrated the model to both BNT162b2 and mRNA‐1273 across dosing regimens, age groups, and vaccine products. Although developed in the context of COVID‐19, these models provide a reusable template for gene therapies and mRNA vaccines in rare diseases, where similar questions—optimal dosing, immune durability, and variability across special populations—must be addressed with minimal patient data.

Other examples illustrate how QSP can bridge gaps between data‐rich and data‐sparse settings. Miyazawa et al. [6] applied a minimal PBPK‐QSP model to explore how mRNA stability, translation efficiency, and endosomal escape determine protein expression, insights that are equally critical when designing mRNA constructs for rare genetic disorders. Dogra et al. [7] used an immune‐response QSP model to simulate booster strategies and predict breakthrough infection risk in COVID‐19, an approach directly translatable to assessing vulnerability windows in rare disease patients with compromised immunity. In a rare disease setting, Paris et al. [8] showed how a pediatric QSP model for spinal muscular atrophy could capture biomarker dynamics (pNfH) under therapy, illustrating that the same mechanistic frameworks can be tuned for very small patient populations.

Collectively, these studies demonstrate that even when QSP models originate from large‐scale or pandemic applications, they can be adapted to rare diseases by recalibrating parameters, integrating disease‐specific biomarkers, and simulating digital twin populations. In rare diseases, where clinical trial sizes are constrained and validated biomarkers are scarce, leveraging the mechanistic depth of existing vaccine QSP models offers a path to de‐risk development, personalize therapy, and anticipate long‐term outcomes in ways traditional empirical methods cannot.

3. AAV/Viral Vector Gene Therapies

Recombinant adeno‐associated viruses (AAVs) can be considered the most widely used and successful platform for in vivo gene therapy. Different AAV‐based gene therapy products have been approved by the Food and Drug Administration (FDA) for the treatment of various genetic disorders such as Hemophilia, Leber congenital amaurosis, spinal muscular atrophy, Duchenne muscular dystrophy (DMD), etc. This therapeutic modality has some unique and challenging aspects from a dose‐selection and clinical‐development perspective [9].

In order to avoid an immune response, each patient can typically be dosed once with AAVs [10]. For rare diseases which can often be severe in nature, the dose given to patients needs to be accurate and efficacious. A recent publication demonstrated that allometric and weight‐based scaling approaches are only ~40% accurate in predicting transgene expression [11]. This can be attributed to the complex exposure–response relationship between AAV vector being dosed and transgene expression/pharmacodynamic effect, which are potentially driven by the differences in:

• •AAV vector capsid (guides biodistribution/tropism),
• •AAV vector genome (guides differences in tissue expression),
• •Specific attributes of the transgene product (i.e., secreted protein vs. intracellular protein),
• •Route of administration (local vs. systemic),
• •Cross‐species differences in biodistribution and expression.

Given the complexities associated with biodistribution and the mechanism of action associated with AAV‐based therapeutics, it is crucial to generate sufficient preclinical data that helps to understand and capture the specific biological processes involved in vector biodistribution and transduction [12, 13].

To translate such complex biological insights into actionable predictions, modeling approaches like PBPK and QSP have been increasingly leveraged. Figure 1 outlines the sequential stages in the development of QSP models tailored to gene therapy applications.

FIGURE 1.

Administration site: intravenous (IV), intrathecal (IT), intra‐muscular (IM), intra‐cisterna magna (ICM), intra‐parenchymal (IP). Biodistribution: physiologically based pharmacokinetic (PBPK) models are used to characterize the PK in different organs (e.g., liver, muscle, CNS, etc.) Cellular uptake: rate constants are used for each intracellular trafficking step. Transgene expression: can be modulated by immune response, reducing AAV transduction efficiency, negative feedback on gene expression, and saturable binding or transport. ChatGPT was used solely to assist in improving the figure design.

PBPK and QSP models have been effectively applied to AAV gene therapy. Liu et al. [14] developed a mechanistic PBPK model describing AAV biodistribution, intracellular trafficking, and transgene expression. Pfizer created a QSP model for liver‐targeted AAV gene therapy in hemophilia B, integrating preclinical data to support dose predictions [15]. Certara recently developed a modular framework for mechanistic modeling and interspecies scaling for AAV‐based gene therapy [16]. Similarly, AstraZeneca adapted a systems biology framework into a QSP model to simulate clinical outcomes [17]. Table 1 summarizes QSP applications across gene therapy modalities.

TABLE 1.

QSP applications across gene therapy modalities.

Modality	Therapeutic example	Modeling focus	Key reference
LNP‐mRNA	COVID‐19 vaccines	Immune response and booster optimization	[7]
LNP‐mRNA	Protein replacement	mRNA kinetics, protein translation	[6]
AAV	Hemophilia B	Vector biodistribution, transgene expression	[14, 15]
CRISPR‐Cas9 (LNP)	NTLA‐2001 (TTR amyloidosis)	Knockdown kinetics, biomarker response	[18, 19]
CRISPR‐Cas9 (LNP)	PCSK9 (LDL‐C)	Feedback modeling, lipid regulation/knockdown	[18]

4. Gene Editing (CRISPR/Cas9) Systems

Gene editing (GE) has emerged as a revolutionary strategy for treating genetic disease by modification or correction of disease‐causing genetic mutations. Pioneered by CRISPR/Cas9, derived from bacteria, CRISPR/Cas9 showed great ability as an effective GE technology with its two components: the guide RNA (gRNA) and a non‐specific CRISPR‐associated endonuclease Cas9. The gRNA identifies the desired DNA target sequences at the genomic locus, and the Cas9 nuclease, once activated, induces a double‐stranded break (DSB). The ends are then rejoined/repaired meanwhile gene insertions or deletions (i.e., Indels) are introduced causing gene knockout and reduced production of the protein of interest, or, a functional gene, delivered via other means can be inserted at a specific locus resulting in increased production of a functional protein.

The subsequent precision editing technologies: base editing (BE) and prime editing (PE) are emerging and on the rise. BE was sought after to attain base substitutions (correcting point mutations) in cellular DNA without inducing double‐stranded breaks (DSBs) whereas PE systems enable precise genome modifications in various cell types and organisms with reduced undesired indels at both target and non‐target sites. The GE systems are delivered directly in vivo using viral vectors (VVs), viral‐like nanoparticles (VLPs), or lipid nanoparticles (LNPs).

QSP modeling with its granularity, layering drug kinetics atop systems‐level insights is a powerful tool to describe biodistribution, cellular uptake as well as intracellular and intranuclear fate of GE multicomponents and to generate virtual populations and perform trial simulations. Indeed, QSP models were employed to describe PK, PD, immunogenicity, biomarker response and the subsequent translation from animal to human. QSP models were initially built integrating biological, pharmacological, and physiological data into a mathematical framework according to biologically plausible hypotheses and were confirmed and/or verified (learn & confirm concept via preclinical experiments to refine hypotheses and improve predictions) to project the first‐in‐human dose and PK/PD predictions.

A mechanistic QSP model was developed for NTLA‐2001, a GE LNPs system of CRISPR/Cas9 and mRNA aiming at reducing the circulating amyloid‐forming transthyretin (TTR) protein (i.e., gene knockout). The model building commenced using literature data as well as mice and nonhuman primates (NHPs) and translated physiologically to whole body across species [18, 19, 20]. The model was able to capture the hallmarks of LNPs PK following intravenous (IV) administration where rapid decline from peak (rapid opsonization, phagocytosis into the mononuclear phagocyte system [MPS] and uptake in hepatocytes via low density‐lipoprotein [LDL] receptor‐mediated endocytosis or macropinocytosis) was followed by a secondary peak (exocytosis) and a log‐linear elimination (lysosomal degradation). PD was modeled in NHPs using an indirect response model. The QSP model captured the reduction in serum TTR protein (biomarker) in polyneuropathy patients at low doses and reflected saturation at higher doses of the dose–response curve. The QSP model was extended to also describe the cascade of events of TTR amyloidosis from liver TTR production to tetramer formation, folded dimers and monomers (first‐order formation) and misfolded aggregates downstream to fibril formation (with saturable clearance mechanism) and deposition in nerves or other organs. The model predicted that accumulated fibrils could be cleared at dose levels associated with greater TTR knockout.

A similar QSP model was built for another CRISPR‐based investigational therapeutic intended for proprotein convertase subtilisin/kexin type 9 (PCSK9) and LDL cholesterol serum reduction whereby PD was modeled via a feedback loop model at the cellular level [18]. Hence, QSP models integrating pathophysiology with drug effects proved their merits in predicting clinical outcomes.

In line with the FDA's roadmap [21] and plan [22] to leverage New Approach Methodologies (NAMs), computer simulations and artificial intelligence to predict drugs' in vivo behavior, the advent of 3D mini‐organs (organ‐on‐chips, OoCs also called human‐on‐chips, HoCs), derived from induced pluripotent stem cells (iPSCs) or directly from patient biopsy samples, was shown to replicate the structural and functional characteristics of real tissues, allowing researchers to study disease mechanisms in a highly personalized environment [23]. Rich mechanistic data from these patient‐derived organoid models as well as GE therapy trials may encourage investigators to pursue developing mechanistic QSP models as a feasible and valuable tool for reducing unnecessary exposure to ineffective or toxic drugs, improving patient safety and treatment outcomes, optimizing dosing, and predicting long‐term outcomes [24]. In addition, NAMs can support model qualification by providing independent experimental data (e.g., in vitro assays, organoids, and organ‐on‐chip systems) that cross‐validate QSP predictions. This enhances regulatory confidence and allows iterative refinement of model components in areas where clinical data are sparse. Table 2 lists challenges and opportunities for QSP in gene therapy.

TABLE 2.

Challenges and opportunities for QSP in gene therapy.

Challenge area	Description	Potential solutions
Sparse biomarker data	Limited time‐course data	Use of virtual populations, data augmentation
Immunogenicity	Poorly characterized for novel vectors	Iterative model calibration, NAM integration
Long‐term dynamics	Gene editing permanence and delayed effects	Mechanistic modeling of disease progression
Model qualification	Regulatory acceptance	Leverage NAMs and cross‐validation

5. Discussion and Conclusion

Generally, only 35% of companies initiate QSP modeling before clinical candidate selection, suggesting underutilization in early stages where robust predictions could have the most impact [25]. For gene therapy for rare diseases, significant strides were taken for QSP model development; yet some lack the systems‐level integration typical of QSP. Other challenges that still exist are (1) complexity of gene editing and the long‐term effects, necessitating novel modeling approaches to capture permanent genetic change, (2) data limitations where long‐term follow‐up (LTFU) data (beyond 1–2 years) are limited, and off‐target effects or clonal diversity remain understudied, do complicate model validation, (3) disease heterogeneity and patient‐specific factors (e.g., age (pediatrics), baseline protein levels, disease severity) require personalized QSP models, which are resource‐intensive and (4) the need for scientific validation of the organoids, organ‐on chips and other NAMs used to generate data for PBPK/QSP model building and the qualification of the QSP models to advance gene editing therapeutics for rare diseases. There are significant ongoing efforts to develop gene therapies for rare autoimmune disorders [26]. Existing PBPK‐QSP based modeling approaches can potentially be refined to incorporate immune‐activation modules to offer significant insights in this modality. In addition to the above discussion, QSP has the potential to personalize rare disease therapeutics in virtual patients and digital twins of actual patients by mechanistically simulating the disease progression of rare diseases and accounting for phenotypic heterogeneity [27].

Disclosure

The opinions expressed are the employees' own and not those of their respective organizations.

Conflicts of Interest

The authors declare no conflicts of interest. N.R. is an employee of Alexion, AstraZeneca Rare Disease and has received salaries and stocks. E.A.C. is an employee of Abbvie and has received salaries and stocks. G.M.L.M.‐T. is an employee of Neurocrine Biosciences and has received salaries and stocks.