Hill Effects and Nonlinear Indirect Responses

Hill Effects Within an Indirect-Response Model

A concentration-effect function can be steep or shallow, but that shape does not by itself determine how quickly an indirect response changes. The response also has turnover: production adds to the response, while loss removes it. This tutorial connects a Hill concentration-effect function to that turnover, focusing on how the concentration-response shape changes the equilibrium and trajectory while the state equation remains linear in the response \(R\).

Let \(C\) be drug concentration, \(k_{\mathrm{in}}\) the baseline production rate, and \(k_{\mathrm{out}}\) the baseline first-order loss rate. The untreated equilibrium is \(R_0=k_{\mathrm{in}}/k_{\mathrm{out}}\). A Hill function modifies either production or loss; at a fixed concentration, it changes the rate terms but does not introduce a power of \(R\) or another nonlinear function of the response.

For inhibition, \(C_{50}\) is often written \(IC_{50}\); for a stimulatory effect, it may be written \(SC_{50}\). The four models below use \(H(C)\) to represent the fractional concentration-dependent effect and \(I_{\max}\) or \(S_{\max}\) for its maximum magnitude. These amplitudes are not the Hill coefficient: \(n_H\) shapes the concentration dependence, while the amplitude sets the largest effect.

Equilibria and Turnover Under Constant Exposure

The following results assume constant concentration, positive baseline rates, and nonnegative effect amplitudes. For inhibitory production, take \(0\leq I_{\max}\leq 1\), so the production rate cannot become negative. For inhibitory loss, the same bound ensures a nonnegative loss multiplier; at finite concentration, the effective loss rate remains positive. Stimulatory production and loss have positive multipliers.

These expressions follow directly from setting the rate of change to zero to find the equilibrium. For example, inhibited loss gives \(0=k_{\mathrm{in}}-k_{\mathrm{out}}\{1-I_{\max}H(C)\}R_{\mathrm{ss}}\). Solving for \(R_{\mathrm{ss}}\) and using \(R_0=k_{\mathrm{in}}/k_{\mathrm{out}}\) gives the stated ratio. To obtain the trajectory, write the constant-concentration equation as \(dR/dt=q-kR\), where \(q\) is the constant production rate and \(k>0\) the effective loss rate. Since \(R_{\mathrm{ss}}=q/k\), the difference \(x=R-R_{\mathrm{ss}}\) obeys \(dx/dt=-kx\), hence \(x(t)=x(0)e^{-kt}\). This also shows why the response approaches its equilibrium exponentially under fixed exposure.

Changing production changes the equilibrium but leaves the response time constant \(1/k_{\mathrm{out}}\) unchanged. Changing loss affects both equilibrium and time constant. Inhibited loss raises the equilibrium and slows the approach to it; stimulated loss lowers the equilibrium and speeds the approach. This distinction is important even when two models predict similar equilibria.

What the Hill Coefficient Changes

The Hill function is monotone in concentration for \(C>0\). Its slope with respect to log concentration makes the role of \(n_H\) especially clear. Taking the derivative gives

At \(C=C_{50}\), \(H=0.5\), so the log-concentration slope is \(n_H/4\). A larger \(n_H\) therefore makes the concentration transition steeper around \(C_{50}\), while leaving the half-effect concentration unchanged. The response equilibrium inherits this concentration dependence through the relevant model equation. For example, inhibited production has \(R_{\mathrm{ss}}/R_0=1-I_{\max}H(C)\); inhibited loss has \(R_{\mathrm{ss}}/R_0=1/\{1-I_{\max}H(C)\}\), which is a nonlinear transformation of the same Hill function.

The time course also depends on where the Hill function enters. For a fixed concentration, production modulation changes the target equilibrium but not the exponential rate of approach. Loss modulation changes the target and the exponential rate. For a changing concentration, the coefficients in the state equation change over time. The equation is still linear in \(R\), but the fixed-concentration exponential formula no longer describes the entire trajectory as one exponential.

Assumptions, Diagnostics, and Common Pitfalls

The formulas presume a well-mixed effect-site concentration is available to the model, the chosen Hill function is an adequate description of concentration dependence, and the response follows a single first-order turnover process. They also presume fixed model parameters over the prediction period. If concentration is not constant, an instantaneous equilibrium calculated at each time is only the equilibrium that would apply if that concentration were held fixed. It is not generally the actual response at that time, because \(R\) retains the history of earlier production and loss.

Useful checks include confirming that \(H(C)\) lies between zero and one, that \(H(C_{50})=0.5\), and that the computed equilibrium has the expected direction relative to baseline. Under fixed concentration and positive effective loss, a response starting below its equilibrium should increase toward it, while one starting above should decrease toward it. A computed response that moves away from the equilibrium signals an algebraic or sign error, or a violation of the model assumptions.

• Confusing steepness with maximum effect. Increasing \(n_H\) steepens the concentration transition; it does not increase \(I_{\max}\) or \(S_{\max}\).
• Using the production-modulation time constant for every model. The rate is unchanged only when production is modified. Modifying loss changes \(k_{\mathrm{eff}}\).
• Treating equilibrium as the observed response. Equilibrium is reached only after sustained fixed exposure and sufficient time. A finite-time response may remain far from it.
• Allowing a negative loss rate. A negative effective loss rate is not a stable turnover model. Check the loss multiplier and the allowed range of effect amplitudes.

A Hill model is more flexible than the \(n_H=1\) Emax form when the concentration-effect transition is steeper or shallower than that form can describe. But the flexibility comes with a need for informative concentration coverage: data concentrated far below or far above \(C_{50}\) may provide little information about the transition’s shape. A direct concentration-response model may describe an equilibrium relationship, but it does not by itself represent the delay caused by turnover. Conversely, an indirect-response model adds dynamic structure, and its predictions rely on the assumed production and loss mechanism. A good fit does not alone establish that this mechanism is correct.