Modeling Antagonism from Shared Targets and Limited Resources

Why Mechanism Can Produce Antagonism

A combination can perform worse than one of its components even when the model contains no parameter explicitly labelled “antagonism.” One route is direct competition: two agents occupy the same target, but the agent that displaces the other produces a weaker effect. A second route is competition for a finite downstream resource, such as a shared adaptor or signaling capacity. In either case, the outcome follows from binding, capacity, and efficacy rather than from an added negative interaction coefficient.

The distinction between antagonism and simple diminishing returns matters. If two drugs each suppress a pathway, a response ceiling can make their combined effect less than the sum of their separate effects. But a monotone ceiling alone does not make the combination less effective than the better monotherapy. To obtain that stronger form of antagonism, one agent must, for example, displace a more efficacious agent or consume capacity in a way that reduces the higher-efficacy contribution.

A Shared-Site Occupancy Model

Definition: Let $C_A$ and $C_B$ be the free concentrations of two agents at a shared binding site, and let $K_A$ and $K_B$ be their equilibrium dissociation constants. Define $x_A=C_A/K_A$ and $x_B=C_B/K_B$. If the site can be unbound or occupied by either agent, and the binding is at equilibrium, the fractions of sites occupied by A and B are $$ \theta_A=\frac{x_A}{1+x_A+x_B}, \qquad \theta_B=\frac{x_B}{1+x_A+x_B}. $$ The unbound fraction is $1/(1+x_A+x_B)$, so the three fractions sum to one.

This model assumes mutually exclusive binding to one site, equilibrium binding, a well-mixed system, and free concentrations that are not appreciably depleted by target binding. It also assumes that one site’s occupancy does not change another site’s binding affinity. These assumptions should be checked against the intended biology; for example, cooperative binding or strong target-mediated depletion would require a different model.

Let $\varepsilon_A$ and $\varepsilon_B$ denote the fractional inhibitory effects produced by occupied sites, relative to the effect of a fully effective inhibitor. Here $0\leq\varepsilon_i\leq1$. For a shared target whose occupancy generates inhibition, the normalized inhibition is

I=\varepsilon_A\theta_A+\varepsilon_B\theta_B =\frac{\varepsilon_A x_A+\varepsilon_B x_B}{1+x_A+x_B}.

The efficacy parameters are properties of the occupied target in this model; they are not binding affinities. A low $K_B$ can make B a strong competitor even if $\varepsilon_B$ is small. That combination—strong occupancy but weak effect per occupied site—is a direct mechanism for antagonism.

Worked Example: A Partial Inhibitor Displaces a Stronger Inhibitor

Suppose, hypothetically, A and B compete at one target. At the measurement time, $C_A=9$ nM and $K_A=1$ nM, while $C_B=9$ nM and $K_B=9$ nM. Let $\varepsilon_A=1.0$ and $\varepsilon_B=0.20$. Calculate inhibition with A alone, B alone, and both agents present.

The dimensionless concentrations are $x_A=9/1=9$ and $x_B=9/9=1$. With A alone, $\theta_A=9/(1+9)=0.9$, so $I_A=1.0(0.9)=0.90$, or 90% normalized inhibition. With B alone, $\theta_B=1/(1+1)=0.5$, so $I_B=0.20(0.5)=0.10$, or 10% normalized inhibition.

With both present, $\theta_A=9/11$ and $\theta_B=1/11$. Therefore, $$ I_{AB}=\frac{1.0(9)+0.20(1)}{11} =\frac{9.2}{11}=0.8364. $$ The combination gives 83.64% inhibition, less than the 90% with A alone. B has reduced the response by about 6.36 percentage points by taking some target occupancy from the fully effective inhibitor. This is antagonism relative to A monotherapy, not merely a failure to achieve the sum of the monotherapy effects.

When Does the Lower-Efficacy Agent Reduce the Effect?

The condition can be derived by comparing the combination with A alone at the same $x_A$. With A alone, $I_A=\varepsilon_A x_A/(1+x_A)$. For $x_B>0$, subtracting gives

I_{AB}-I_A = \frac{x_B\left[\varepsilon_B(1+x_A)-\varepsilon_A x_A\right]} {(1+x_A+x_B)(1+x_A)}.

The denominator is positive. Therefore the combination reduces inhibition relative to A alone exactly when

\varepsilon_B(1+x_A)<\varepsilon_A x_A, \qquad\text{or equivalently}\qquad x_A>\frac{\varepsilon_B}{\varepsilon_A-\varepsilon_B} \quad\text{when }\varepsilon_A>\varepsilon_B.

The threshold does not depend on the size of $x_B$, provided $x_B>0$, in this idealized model: any positive B occupancy reduces A’s effect once the threshold is crossed. The magnitude of the reduction does depend on $x_B$, and approaches a limit as B increasingly dominates occupancy. If B is at least as efficacious as A, this particular mechanism cannot make the combination worse than A alone. The same comparison can be made with the labels reversed to assess whether A antagonizes B.

Finite Downstream Capacity

The same mathematics can represent agents competing for a finite downstream resource rather than a molecular binding site. For instance, suppose a shared adaptor has total capacity $R_T$. At equilibrium, A-bound adaptor, B-bound adaptor, and free adaptor divide that total pool. If $x_A$ and $x_B$ represent their dimensionless demands for the resource, the fractions assigned to A and B are $\theta_A$ and $\theta_B$ above. If occupied adaptor generates suppressive output with efficacy $\varepsilon_i$, the output is $I_{\max}(\varepsilon_A\theta_A+\varepsilon_B\theta_B)$, where $I_{\max}$ is the output at full occupancy by a fully effective agent.

Worked Example: Competition for a Limited Adaptor Pool

Consider a hypothetical system with $I_{\max}=100$ response units. A and B compete for a finite adaptor pool, with $x_A=x_B=4$, $\varepsilon_A=0.80$, and $\varepsilon_B=0.20$. Find the output with A alone and with both agents.

With A alone, the A-bound fraction is $4/(1+4)=0.8$. The output is $100(0.80)(0.8)=64.0$ response units. With both agents, the total denominator is $1+4+4=9$, so $\theta_A=4/9$ and $\theta_B=4/9$. Thus, $$ I_{AB}=100\left(0.80\frac{4}{9}+0.20\frac{4}{9}\right) =100\left(\frac{4.0}{9}\right)=44.44 $$ response units. The combination produces less output than A alone because B occupies part of the limited pool while generating less output per occupied fraction.

This example illustrates an important modelling choice: the resource must actually be limiting and shared, and occupancy must be linked to output in the stated way. Merely naming a component “capacity” does not establish competition. Evidence that increasing one agent reduces the other agent’s engagement, or that the resource pool is finite over the relevant range, supports this mechanism.

Saturation Is Not Automatically Antagonism

A different model may send two noncompeting upstream inputs into a saturable downstream response. For example, if $U_A$ and $U_B$ are nonnegative contributions to an activating input, a saturating output might be $Y=Y_{\max}(U_A+U_B)/(K+U_A+U_B)$, for positive $Y_{\max}$ and $K$. The derivative with respect to either input is positive:

\frac{\partial Y}{\partial U_B} =\frac{Y_{\max}K}{(K+U_A+U_B)^2}>0.

Adding B therefore cannot lower this output under the stated model, although the incremental gain can become small as the response approaches $Y_{\max}$. That is diminishing return, not antagonism relative to A alone. A claim of antagonism needs a clearly stated comparison: for example, less inhibition than A monotherapy, less benefit than an additive reference, or a negative interaction on a specified response scale. The introductory material on selecting a reference null for combination synergy is relevant when a null model, rather than monotherapy, defines the comparison.

Worked Example: Exposure Overlap Changes Competition

Suppose a hypothetical target has $K_A=2$ nM and $K_B=4$ nM, with $\varepsilon_A=1.0$ and $\varepsilon_B=0.25$. At an early measurement, $C_A=8$ nM and B is absent. At a later measurement, A remains at 8 nM and $C_B=4$ nM. Assume binding equilibrates rapidly at each measurement.

Early, $x_A=8/2=4$ and $x_B=0$, so $I=4/(1+4)=0.800$, or 80.0% inhibition. Later, $x_A=4$ and $x_B=4/4=1$, giving $$ I=\frac{1.0(4)+0.25(1)}{1+4+1} =\frac{4.25}{6}=0.7083. $$ At the later measurement, inhibition is 70.83%, 9.17 percentage points below the early A-only value. The calculation does not imply that B always antagonizes A: it shows that, while their concentrations overlap at the target, B’s occupancy reduces the contribution from A under this model. If their exposures do not overlap, this particular competition is absent at that time.

Assumptions, Diagnostics, and Alternatives

The occupancy equations describe equilibrium at a specified time, not the full evolution of concentrations, binding, or downstream states. Applying them requires concentrations at the effect site, credible affinity estimates, and sufficiently rapid equilibration. If binding is slow, the receptor pool is depleted from solution, or the response has a substantial delay, the calculated equilibrium occupancy may not represent the observed effect. A static calculation also cannot explain persistence after concentrations change; that requires explicit dynamic states.

Useful diagnostics include checking whether measured target engagement is consistent with the predicted occupancy fractions, whether the combination response falls below the more effective monotherapy at matched exposure, and whether changing the competitor’s concentration changes the response in the predicted direction. If the data show a concentration-dependent reduction in A occupancy as B increases, that supports competition more directly than a response curve alone. Conversely, if B reduces response without changing A engagement, investigate alternative mechanisms such as downstream inhibition, toxicity, altered exposure, or measurement artifacts rather than forcing a shared-site explanation.

Mechanistic competition models differ from the earlier material on Bliss independence and Loewe additivity. Those approaches define reference expectations from single-agent effects or dose equivalence; occupancy models instead encode a specific mechanism. A mechanistic model can explain why a combination departs from a reference, but it does not make the result independent of the endpoint or response scale. The most useful analysis may compare a mechanistic model with a reference-null analysis, provided each model’s assumptions and comparison are stated.

Do not infer shared binding from antagonism alone. Multiple mechanisms can produce the same response pattern.
Do not equate low affinity with low efficacy. $K_i$ controls occupancy in this formulation; $\varepsilon_i$ controls effect per occupied fraction.
Check the monotherapy comparison. A combined response below an additive expectation is not automatically below either monotherapy.
Check the measurement time. Concentration overlap and equilibration determine whether the static competition calculation applies.

Key takeaway: A negative interaction coefficient is not required for antagonism. If a lower-efficacy agent takes occupancy or finite capacity from a higher-efficacy agent, the resulting mechanistic model can predict less effect than the stronger monotherapy. A saturating response by itself generally predicts diminishing returns, not a decrease below that monotherapy.

Check Your Understanding

Answer each question, then compare with the brief solutions below.

If $x_A=3$, $x_B=1$, $\varepsilon_A=1$, and $\varepsilon_B=0.2$, calculate $I_{AB}$. Is it less than A-alone inhibition?
For $\varepsilon_A=0.8$ and $\varepsilon_B=0.2$, find the condition on $x_A$ for B to reduce inhibition relative to A alone when $x_B>0$.
Why does a monotone saturating response to $U_A+U_B$ not, by itself, establish antagonism?
Name two assumptions of the shared-site equilibrium model that could fail in a biological system.

Solutions. 1. $I_{AB}=[1(3)+0.2(1)]/(1+3+1)=3.2/5=0.640$. A alone gives $3/(1+3)=0.750$, so the combination is lower. 2. The threshold is $x_A>\varepsilon_B/(\varepsilon_A-\varepsilon_B)=0.2/0.6=0.3333$. 3. Adding a nonnegative input increases or leaves unchanged the saturating response; it can reduce the marginal gain but not the response below A alone. 4. Examples include rapid equilibrium, mutually exclusive binding, negligible depletion of free drug, and absence of cooperativity.