Introduction to Single-cell Physiologically-Based Pharmacokinetic (scPBPK) Models

This paper introduces single-cell physiologically-based pharmacokinetic (scPBPK) models that incorporate expression-dependent processes via weighting functions to simulate and analyze cellular drug concentration heterogeneity, demonstrating their utility through case studies on AZD1775 and midazolam.

The Gist

The Big Idea: From "Crowd Average" to "Individual Stories"

Imagine you are a doctor trying to understand how a medicine works in the human body. For decades, scientists have used a tool called PBPK models (Physiologically-Based Pharmacokinetic models).

Think of a standard PBPK model like looking at a crowd from a helicopter. You can see the whole group moving together. You know the average speed of the crowd, the average size of the group, and where the crowd is going. It's great for general planning, but it hides the details. You can't see if one person is running fast while another is walking slowly, or if one person is wearing a heavy backpack while another is empty-handed.

This paper introduces a new tool called scPBPK (single-cell PBPK). This is like landing the helicopter and looking at every single person in the crowd individually. It allows scientists to see how a drug behaves inside one specific cell versus its neighbor, revealing hidden differences that the "helicopter view" misses.

---

The Secret Ingredient: "Expression-Dependent" Processes

To make this "single-cell" view work, the authors had to account for something called Expression-Dependent (ED) processes.

The Analogy: The Factory Workers
Imagine the liver is a massive factory that breaks down drugs. In the old "helicopter view," we assumed every worker in the factory was identical and worked at the exact same speed.

But in reality, workers are different.

• Worker A might be super strong and break down drugs very fast.
• Worker B might be tired and work slowly.
• Worker C might have a broken tool and work even slower.

In biology, these "workers" are proteins (enzymes) inside cells. The amount of protein a cell has is called its "expression." The paper's breakthrough is realizing that drug metabolism depends on how much "worker" (protein) a specific cell has.

To model this, the authors used a Weighting Function. Think of this as a dice roll for every single cell.

• Some cells roll a 6 (lots of workers, fast drug breakdown).
• Some roll a 1 (few workers, slow breakdown).
• Most roll a 3 or 4 (average speed).

They used a specific type of dice roll called a Negative Binomial distribution (a fancy statistical tool often used in genetics) to make sure the randomness looked like real human biology.

---

The Two Experiments: A Tale of Two Drugs

The authors tested their new "single-cell microscope" on two different drugs to see how it changed the results.

1. The Drug AZD1775 (The "Traffic Jam" Drug)

• The Scenario: This drug tries to enter the brain. The brain has a security gate called the Blood-Brain Barrier (BBB).
• The Process: There are "pumps" (like P-gp and ABCG2) that try to kick the drug out of the brain. These pumps are Expression-Dependent.
• The Result: Because the pumps vary wildly from cell to cell (some cells have huge pumps, some have tiny ones), the drug levels inside the brain cells became highly chaotic.
• The Takeaway: In the old model, everyone in the brain had the same amount of drug. In the new model, some cells were swimming in the drug, while others had almost none. This is huge for cancer treatment because if a tumor cell has low drug levels, the cancer might survive.

2. The Drug Midazolam (The "Fast Lane" Drug)

• The Scenario: This drug is broken down by the liver.
• The Process: The liver cells have enzymes that chew up the drug. This is also Expression-Dependent.
• The Result: Surprisingly, the drug levels in the liver cells were almost identical for everyone, even though the enzymes varied!
• Why? Imagine a highway. Even if some cars (enzymes) are slow and some are fast, if the traffic flow (drug moving into the cell) is incredibly fast, the cars don't have time to pile up or get stuck. The "traffic" moves so fast that the differences in the workers don't matter. The drug levels stayed smooth and uniform.

---

Why Does This Matter?

1. Finding the "Hidden" Failures
If you treat a patient with a drug, the "helicopter view" might say, "Great, the average drug level is perfect!" But the "single-cell view" might reveal, "Oh no, 30% of the cancer cells didn't get enough drug to die." This helps explain why some treatments fail even when the math looks right.

2. Connecting to the Future
We are living in the age of "Omics" (genetics, proteins, etc.), where we can read the DNA of individual cells. This new model is the bridge that connects that genetic data to how drugs actually move in the body.

3. The Next Step
Right now, this model predicts how drugs move (Pharmacokinetics). The authors hope to soon connect this to how drugs act (Pharmacodynamics). Imagine a model that doesn't just tell you how much drug is in a cell, but predicts exactly how that specific cell will react based on its unique genetic makeup.

Summary

This paper builds a microscope for drug movement. It moves us from thinking of the body as a bag of identical marbles to a complex city of unique individuals. By using math to simulate how different cells have different "workforces," scientists can finally predict why a drug might work for one person (or one cell) but fail for another.

Technical Summary

1. Problem Statement

Traditional Physiologically-Based Pharmacokinetic (sPBPK) models are a cornerstone of drug development, treating tissues as homogeneous compartments (lumped vascular, interstitial, and intracellular spaces). However, these models fail to capture cellular heterogeneity, which is increasingly recognized as critical due to advances in multi-omic technologies (scRNAseq, proteomics).

• The Gap: There is currently no established framework to model drug disposition at the single-cell level.
• The Challenge: Standard models assume uniform enzyme expression and transport rates across a tissue. In reality, gene and protein expression vary significantly between individual cells, leading to heterogeneous drug concentrations and metabolic rates that standard PBPK models cannot predict.
• The Need: A modeling framework is required to bridge the gap between whole-organ pharmacokinetics and single-cell biology to better understand treatment failure, drug resistance, and variable efficacy among cell subpopulations.

2. Methodology

The authors propose a Single-cell Physiologically-Based Pharmacokinetic (scPBPK) framework that modifies standard ODE-based PBPK models to account for expression-dependent (ED) processes.

Core Framework

• Model Structure: The model retains the standard whole-body structure (arterial/venous blood, organs) but subdivides specific organ compartments (e.g., liver, brain) into a 3-subcompartment structure: Vascular, Interstitial Fluid (IF), and Intracellular (IC).
• Differentiation of Cell Populations: The Intracellular compartment is split into:
  1. Single Cells ($n_{sc}$): A defined number of individual cells (e.g., 10,000) simulated individually to capture heterogeneity.
  2. Bulk Cells ($n_{bulk}$): The remaining cells in the tissue, treated as a homogeneous pool to ensure mass conservation.
• Expression-Dependent (ED) Processes: Key processes like membrane transport or metabolism are identified as ED. Their kinetic parameters (e.g., $V_{max}$ for metabolism, transport clearances) are not constant but are modulated by a Weighting Function ($W_i$).
• Weighting Function:
  • Defined as a Negative Binomial (NB) distribution, $W_i \sim NB(\mu, \theta)$, reflecting the statistical properties of scRNAseq data.
  • $V_{max,i} = (V_{max,total} / n_{tot}) \times W_i$.
  • This introduces stochastic variability in metabolic capacity or transport rates for each simulated cell.
• Mass Conservation: The model ensures that the sum of fluxes and clearances across all single cells and bulk cells equals the total organ-level clearance observed in standard PBPK models.

Case Studies

Two distinct drug models were developed to test the framework:

1. AZD1775 (Adavosertib): A WEE1 kinase inhibitor.
   • Focus: Blood-Brain Barrier (BBB) transport.
   • ED Processes: Three active transporters (P-gp efflux, ABCG2 efflux, OATP1B1 uptake) and passive diffusion.
   • Implementation: Converted a 2-compartment brain model into a 3-compartment scPBPK model with 4 clusters of 2,500 cells each.
2. Midazolam (MDZ): A CYP3A4/3A5 probe drug.
   • Focus: Hepatic metabolism.
   • ED Process: Metabolism by CYP enzymes (Michaelis-Menten kinetics).
   • Implementation: Converted a Minimal PBPK (mPBPK) model into a scPBPK model with the liver as the 3-compartment structure.

3. Key Contributions

• Novel Mathematical Formalism: The first derivation of ODEs that simultaneously track bulk tissue kinetics and individual cell kinetics within the same physiological framework.
• Integration of Omics Data: The use of Negative Binomial distributions to parameterize cellular heterogeneity, directly linking scRNAseq statistical properties to pharmacokinetic modeling.
• Demonstration of Heterogeneity Mechanisms: The paper proves that cellular heterogeneity in drug concentration is not guaranteed; it depends on the ratio of mass transport rates ($h$) to metabolic capacity ($V_{max}$).

4. Results

The simulations yielded contrasting results for the two drugs, highlighting the conditional nature of cellular heterogeneity:

• AZD1775 (BBB Transport):
  • Result: Significant heterogeneity was observed in brain interstitial fluid (IF) and intracellular (IC) concentrations.
  • Observation: Different clusters showed up to a 4-fold variation in maximum concentration.
  • Cause: The heterogeneity in transporter expression (ED processes) at the BBB directly translated to concentration differences because transport was the rate-limiting step.
• Midazolam (Hepatic Metabolism):
  • Result: No significant heterogeneity in intracellular concentrations was observed, despite high variability in single-cell metabolic clearances ($V_{max}$).
  • Observation: Single-cell and bulk cell concentration profiles were nearly identical.
  • Mechanism: The mass transfer coefficient ($h_2$) between interstitial fluid and intracellular space was very high relative to $V_{max}$. Rapid transport homogenized the drug concentration across cells before metabolism could create disparities.
  • Variance Analysis: A biphasic relationship was found between the $h_2/V_{max}$ ratio and concentration variance. Maximum variance occurred at a ratio of ~0.01. At very low ratios (transport-limited) or very high ratios (metabolism-limited), variance decreased.

5. Significance and Future Implications

• Predicting Treatment Failure: scPBPK models can identify subpopulations of cells (e.g., tumor cells) that fail to reach therapeutic concentrations (e.g., $>EC_{50}$) due to specific expression profiles, explaining why a drug works in some patients but not others.
• Mechanistic Insight: The framework allows researchers to deconvolute whether variability is driven by transport limitations or metabolic differences.
• Pathway to scPK/PD: This work lays the foundation for Single-cell PK/PD models, linking cellular drug exposure directly to single-cell pharmacodynamic responses (e.g., cell death, signaling changes).
• Synergy with Multi-omics: The model is designed to be compatible with modern single-cell data, allowing for the integration of transcriptomic and proteomic datasets to refine drug development strategies and personalized medicine.

In summary, this paper establishes a rigorous mathematical framework to move pharmacokinetics from the "average cell" to the "individual cell," revealing that cellular heterogeneity in drug exposure is a dynamic property determined by the interplay between transport kinetics and metabolic capacity.