How Pharmacokinetic and Pharmacodynamic Variability Shape Responses to Combination Therapy

A new modeling study suggests that the biggest source of variation in how patients respond to medicines may not be how quickly their bodies absorb, distribute, or eliminate a drug, but how their biological systems process the signal that the drug creates. The finding could challenge a central assumption behind many dosing strategies: that measuring drug concentrations in the blood is enough to predict whether treatment will work. In simulations of single-drug and two-drug treatments, pharmacodynamic variability—the differences among patients in receptor signaling, pathway sensitivity, feedback regulation, and downstream biological responses—often outweighed pharmacokinetic variability. The analysis also found that adding a second drug could make responses more consistent, even when patients retained substantial differences in drug exposure. The results, published in Pharmacology Research & Perspectives, offer a mechanistic explanation for a clinical puzzle seen across combination therapies: some patients benefit even when the concentration of one medicine falls below its conventional target.

The study focused on interindividual variability, or IIV, the biological spread that causes the same dose to produce different effects in different people. Pharmacokinetics describes what the body does to a drug, including absorption, distribution, metabolism, and clearance. Pharmacodynamics describes what the drug does to the body after it reaches its target. A patient may therefore have a low concentration because the drug is cleared rapidly, yet still experience a strong response if receptors, signaling proteins, or feedback circuits are unusually sensitive. Another patient may reach the same concentration but respond weakly because the relevant pathway is less responsive. Conventional therapeutic-drug monitoring is designed mainly to control pharmacokinetic variation. The new work argues that this approach can overlook the biological processes closest to the final therapeutic effect, where small differences in pathway behavior may be amplified.

To investigate the problem, the researchers developed a unified pharmacokinetic–Signal-Reaction-Stimulus-Response, or PK-SRSR, framework. They modeled two hypothetical drugs, A and B, with similar molecular weight, apparent clearance, and dosing frequency. At steady state, the simulated concentration was determined by dose, bioavailability, clearance, and dosing interval. That exposure was then converted into molecular signals using equations that represented receptor-level drug effects and the activity of endogenous biochemical pathways. The model separated parameters into three broad categories: exposure parameters, drug-level pharmacodynamic parameters, and system-level pharmacodynamic parameters. Exposure was represented by apparent clearance for each drug. Drug-level parameters included maximal signal strength and the concentration producing half of that signal, commonly related to potency or EC50. System-level parameters described baseline endogenous stimulation and the sensitivity of biological pathways, including interactions between the pathways activated by the two drugs.

The SRSR component was designed to capture a crucial feature of biology: medicines do not operate in isolation. They mimic, amplify, or interfere with endogenous signals, and those signals can interact through enzyme-mediated biochemical reactions. In the model, the signal generated by each drug depended on its steady-state concentration, its maximal effect, and its potency parameter. The final response was then calculated from the interaction of the two signals and their pathway sensitivities. An interaction index classified the drug relationship as additive when the index equaled one, antagonistic when it exceeded one, or synergistic when it fell below one. The researchers fixed the combination potency at a normalized value and varied other parameters across simplified values chosen to represent weak, moderate, and strong signaling regimes. These were not clinical measurements from named medicines or patients, but controlled computational conditions intended to reveal how variability propagates through a biological network.

The team used Monte Carlo simulations to generate response distributions under hypothetical physiological conditions. In these simulations, each model parameter could vary among individuals according to a log-normal distribution, a common choice for positive biological quantities that fluctuate multiplicatively rather than symmetrically. Variability was tested at coefficients of variation ranging from zero to 60 percent, with 45 percent treated as a moderate level. The researchers simulated monotherapy with 10 milligrams of drug A and combination therapy with 10 milligrams each of drugs A and B. They also explored doses of drug A from zero to 200 milligrams, with repeated simulations used to estimate how uncertainty in concentration and response changed across dose levels. To identify which parameters mattered most, they applied Sobol variance-based sensitivity analysis to 1,000 simulated individuals. A first-order index measured the direct contribution of one parameter to response variance, whereas a total-order index included its interactions with all other parameters. The gap between the two indices therefore revealed the importance of nonlinear parameter interactions.

Under monotherapy, the strongest influences came from system-level parameters rather than clearance or drug-specific potency. In several interaction settings, the response was especially sensitive to S0B, the baseline signal associated with pathway B, and βB, the apparent potency of that pathway. Under additive conditions, each produced first-order and total-order sensitivity values above 49 percent, while most other parameters contributed less than 5 percent. Under synergistic conditions, S0B and βB still dominated, with first-order contributions above 34 percent, followed by smaller effects from the maximal signal of drug A and the interaction-related parameter βiA. Antagonistic conditions showed a similar pattern, with S0B and βB accounting for much more of the response variability than the remaining parameters. The result is striking because drug A was the medicine being administered, yet the simulated response could be governed more strongly by the behavior of a biological pathway associated with the other, untreated signaling system.

Combination therapy changed the ranking. Once drug B was added, direct drug-effect parameters became much more important, while the influence of system-level variation generally declined. Under synergy, the maximal signal strengths of drugs A and B and the potency of pathway B were among the leading contributors. Drug A’s maximal signal had a first-order contribution above 25 percent and a total-order contribution above 27 percent; drug B’s corresponding values were 17 and 24 percent, while βB contributed about 32 percent directly and 34 percent overall. Under antagonism, maximal signal and pathway sensitivity parameters shared influence, with SmaxA, SmaxB, βA, and βB each contributing roughly one-fifth to nearly one-third of response variance. Under additivity, βB remained particularly influential, with first-order and total-order values of 51 and 54 percent, followed by the maximal signal of drug B. These patterns indicate that the second medicine did not simply add another source of uncertainty. By acting on a modulatory pathway, it could reshape the network so that baseline biological differences mattered less.

The overall simulations produced the clearest numerical evidence of this stabilizing effect. When every model parameter carried a 45 percent coefficient of variation, the coefficient of variation in response during monotherapy was 59.0 percent for additive interactions, 52.3 percent for antagonistic interactions, and 52.6 percent for synergistic interactions. With both drugs present, those values fell to 44.2, 42.4, and 33.0 percent, respectively. In the simulated synergistic combination, response variability was therefore reduced by almost 20 percentage points compared with monotherapy. Increasing variability in any parameter category increased uncertainty in the outcome, but system-level pharmacodynamic parameters had the largest overall effect, followed by drug-level pharmacodynamic parameters and exposure parameters. As the dose of drug A rose from zero, clearance initially became relevant, then its contribution declined at high doses. Drug-effect parameters showed a similar dose-dependent pattern, while system-level parameters were most influential in the absence of treatment and dropped sharply after a low dose was introduced.

The model offers a possible explanation for why combination treatments can succeed despite apparently inadequate plasma concentrations of one component. If drug B stabilizes feedback or compensates for differences in pathway sensitivity, the final response may become less dependent on the exact exposure to drug A. A concentration target derived from monotherapy could consequently misrepresent the dose required in a combination regimen, because the second drug has changed the exposure–response relationship. The authors argue that, when pharmacokinetic optimization is used for combinations, all agents should be considered jointly rather than adjusting one drug in isolation. More broadly, the findings support response-guided strategies that incorporate pharmacodynamic biomarkers, receptor or signaling characteristics, and patient-specific biological information. Such an approach could eventually complement blood-level monitoring with measurements of what the treatment is doing inside the relevant biological system.

The findings remain a mechanistic hypothesis rather than a ready-made clinical dosing rule. The researchers used hypothetical drugs, normalized parameter values, assumed distributions, and steady-state conditions in which concentrations, receptor occupancy, and downstream signaling were treated as being near equilibrium. Real treatments often involve delayed effects, receptor trafficking, adaptive changes, time-dependent feedback, and distinct patient subpopulations whose biology cannot be represented by a single smooth distribution. The study also did not specify particular signaling pathways or validate its predictions against clinical data. Its value lies instead in showing how apparently modest differences in biological regulation can dominate the variability of treatment response and how pathway interactions may suppress that variability. Experimental and clinical studies will be needed to determine whether the simulated rankings hold in real diseases. If they do, precision pharmacotherapy may need to move beyond the question of how much drug reaches the bloodstream and ask a more consequential one: how does each patient’s biological network transform that exposure into a response?