Transporting Emax and Hill Parameters Across Species

Why Parameter Transport Requires Separate Decisions

A concentration-response curve is not a single parameter. Its upper limit, the concentration associated with half of that limit, and the steepness around that concentration describe different features of the response. Agreement between species in one feature does not establish agreement in the others. For example, two species may approach the same maximum effect while differing in potency; alternatively, they may have similar potency but different maxima or curve shapes.

Transporting parameters therefore means imposing specific constraints on a model, not assuming that two curves are interchangeable. The constraint should be justified by the response endpoint, the concentration scale, the biology, and the information in the data. Earlier material on unbound exposure and species differences in target affinity is relevant: using comparable concentrations and accounting for known affinity differences can make a shared potency assumption more interpretable, but it does not prove that the remaining PD parameters are shared.

Definition: For species $s$, a stimulatory Emax-Hill concentration-response model is $$ E_s(C)=E_{\max,s}\frac{C^{h_s}}{EC_{50,s}^{h_s}+C^{h_s}}, \qquad C\geq 0. $$ Here $E_s(C)$ is the drug-attributable response at concentration $C$, $E_{\max,s}>0$ is the maximum effect, $EC_{50,s}>0$ is the concentration producing half of that species' maximum effect, and $h_s>0$ is the Hill coefficient controlling curve steepness. Concentration and $EC_{50,s}$ must use the same units; $E_{\max,s}$ uses the response units.

At $C=EC_{50,s}$, the fraction in the model is $1/2$, whatever the value of $h_s$. At concentrations far below $EC_{50,s}$, the response is small relative to $E_{\max,s}$; at concentrations far above it, the response approaches $E_{\max,s}$. These properties explain why data near the midpoint inform potency and slope, while observations near a genuine plateau are especially important for estimating the maximum. The response here is defined on a comparable, drug-attributable scale; differences in baseline or disease state are not resolved by this curve.

What It Means to Share a Parameter

A shared parameter constraint sets the corresponding value equal across species. For two species A and B, a shared maximum means $E_{\max,A}=E_{\max,B}$, while allowing $EC_{50,A}$, $EC_{50,B}$, $h_A$, and $h_B$ to differ if the model permits. A shared potency constraint sets $EC_{50,A}=EC_{50,B}$; a shared Hill constraint sets $h_A=h_B$. These are distinct assumptions and should be considered separately before imposing them together.

A fully shared curve constrains all three parameters to be equal. A partially shared model constrains only a subset, and a species-specific model estimates all three separately. Partial sharing is often the most useful scientific comparison: it can represent, for example, a conserved efficacy ceiling with species-dependent potency. Whatever structure is chosen, the concentration-response function, response definition, and units must be comparable between species.

For the constraints to be meaningful, assume that the measured concentration represents the concentration relevant to the effect, the endpoint is measured comparably, and the observations can be described by a stable concentration-response relationship. If the effect is delayed relative to concentration, if the relevant exposure differs from the measured exposure, or if assay scales differ, an apparent parameter difference may reflect those issues rather than a true species difference in the curve. Conversely, imposing equality does not correct such problems.

Worked Example: Shared Maximum, Different Potency

Suppose hypothetical response data support a common maximum of 100 response units, but potency differs. For illustration, let species A have $EC_{50,A}=2$ concentration units and species B have $EC_{50,B}=5$ concentration units, with $h_A=h_B=1$. Calculate the predicted responses at $C=10$ and $C=30$.

With $h=1$, the model is $E_s(C)=100C/(EC_{50,s}+C)$. At $C=10$, species A is predicted to have $$ E_A(10)=100\frac{10}{2+10}=100\frac{10}{12}=83.33\text{ response units}. $$ Species B is predicted to have $$ E_B(10)=100\frac{10}{5+10}=100\frac{10}{15}=66.67\text{ response units}. $$ At $C=30$, the predictions are $$ E_A(30)=100\frac{30}{2+30}=93.75\text{ response units}, \qquad E_B(30)=100\frac{30}{5+30}=85.71\text{ response units}. $$

The model has the same maximum in both species but does not predict equal responses at a given concentration. Species B requires a higher concentration to reach the same fraction of its maximum. These calculations illustrate a possible parameterization; the specified values would need support from data, especially observations approaching the plateau. The example assumes that the two species' responses are expressed on the same scale and that concentration is comparably defined.

Why Plateau and Curve-Shape Information Matter

The fitted parameters can be strongly dependent on the concentrations studied. If all observations are well below $EC_{50}$, then the data may mainly define the rising portion of the curve. A larger $E_{\max}$, a different $EC_{50}$, or a different $h$ can sometimes produce similar predictions over that narrow range. If no species has observations near saturation, apparent agreement in fitted maxima is weak evidence for a truly shared maximum. A shared-parameter fit may simply be borrowing the same poorly identified value for both species.

Likewise, a few concentrations clustered near $EC_{50}$ may provide information about the midpoint but little about the slope on both sides of it. A shared Hill coefficient should not be treated as established merely because two curves cross at the midpoint: by definition, both curves produce half of their respective maxima at their own $EC_{50}$. Shape comparisons need observations spanning concentrations below and above the midpoint, preferably with enough coverage to assess the approach to the plateau.

Worked Example: A Shared Maximum Does Not Imply a Shared Prediction

Consider a hypothetical species B with concentration $C=200$ concentration units, $EC_{50,B}=2$ concentration units, and $h_B=1$. Suppose its maximum is constrained to the shared value $E_{\max}=100$ response units. The prediction is $$ E_B(200)=100\frac{200}{2+200} =100\frac{200}{202} =99.01\text{ response units}. $$ This is close to, but below, the shared maximum of 100 response units, as expected at a concentration 100 times the $EC_{50}$.

The arithmetic $100\times200/(2+2)=5000$ is correct for that written expression, but it is not the Emax-Hill prediction at $C=200$. The denominator in the model must include $C$, so it is $2+200$, not $2+2$. Using $2+2$ while retaining 200 in the numerator mixes different concentrations and produces a value that exceeds the assumed maximum. In this model, the response fraction $C/(EC_{50}+C)$ cannot exceed 1 for nonnegative concentrations, so the predicted response cannot exceed $E_{\max}$.

This example also shows a limit of the constraint: a response near 100 at one high concentration is consistent with the shared maximum but does not demonstrate that the species have the same maximum. A meaningful comparison requires data that can distinguish a common plateau from separate plateaus.

Worked Example: Different Hill Coefficients, Same Midpoint

Suppose two hypothetical species have $E_{\max}=100$ response units and $EC_{50}=4$ concentration units, but species A has $h_A=1$ while species B has $h_B=2$. At the midpoint $C=4$, both predict $$ E_A(4)=100\frac{4}{4+4}=50, \qquad E_B(4)=100\frac{4^2}{4^2+4^2}=50\text{ response units}. $$ Thus, agreement at the midpoint alone cannot establish a shared Hill coefficient.

At $C=2$, species A predicts $100(2)/(4+2)=33.33$ response units. Species B predicts $$ E_B(2)=100\frac{2^2}{4^2+2^2} =100\frac{4}{20}=20.00\text{ response units}. $$ At $C=8$, species A predicts $100(8)/(4+8)=66.67$ response units, whereas species B predicts $$ E_B(8)=100\frac{8^2}{4^2+8^2} =100\frac{64}{80}=80.00\text{ response units}. $$

The maxima and midpoints are identical, yet the curves differ on both sides of the midpoint. These predictions follow from the specified Hill model and parameters; they do not establish that a two-fold difference in Hill coefficient is biologically plausible for any particular compound. In an application, that question depends on the data, the measurement scale, and the mechanistic interpretation available.

A Practical Strategy for Parameter Constraints

Make the response and concentration comparable.
Confirm that species are analyzed on a common response scale and that concentrations have a comparable meaning, such as a consistent unbound concentration measure. Check units before comparing $EC_{50}$ values.

Assess each curve feature separately.
Ask whether the data inform the plateau, the midpoint, and the slope. Do not infer a shared maximum from similar mid-range responses or infer a shared slope from a shared midpoint.

Fit scientifically motivated constraints.
Compare a fully shared curve with partial-sharing alternatives and a species-specific curve when the data support those fits. State which parameter is constrained and which remains species-specific.

Check predictions against the observations.
Inspect whether each model captures the observed response range in each species, especially at low and high concentrations. A constraint that forces systematic overprediction or underprediction in one species is not supported just because it reduces the number of parameters.

Limit the transport claim to the evidence.
Report the assumed equalities and their rationale. If the data do not identify a species-specific parameter, describe it as weakly informed rather than concluding that the parameter is equal across species.

The alternatives make different commitments. A fully shared curve is parsimonious but asserts equal maximum, potency, and slope. A partially shared model can express a more selective biological hypothesis, such as shared $E_{\max}$ with species-specific $EC_{50}$. A fully species-specific model avoids equality constraints but requires enough information within each species to estimate all parameters reliably. External mechanistic evidence can motivate a constraint, but should not be confused with evidence that the observed concentration-response data themselves identify it.

Useful diagnostics include plotting observations and fitted curves by species, checking whether residuals show a consistent species-specific pattern across concentration, and comparing parameter estimates under plausible constraints. Sparse coverage, implausible fitted slopes, or maxima driven by extrapolation indicate that conclusions depend strongly on the model assumptions. A simple model comparison statistic cannot by itself establish biological equivalence: evidence for equivalence requires an appropriate definition of how close the parameters or predictions must be, as well as data precise enough to assess that definition.

Common Mistakes and Limitations

Sharing every parameter because one parameter appears conserved. A common $E_{\max}$ does not entail a common $EC_{50}$ or $h$. Specify constraints parameter by parameter.
Calling an observed response the maximum. A high observed response may still be below the plateau. Without concentrations that approach saturation, $E_{\max}$ can be weakly determined.
Reading $EC_{50}$ as a universal potency constant. It is the half-maximum concentration for the specified model, endpoint, and concentration scale. It can differ across species and can be difficult to interpret if exposures or assay conditions are not comparable.
Interpreting a fitted Hill coefficient as a direct mechanistic count. The coefficient describes curve steepness in this model; by itself, it does not identify the number of binding sites or prove a particular molecular mechanism.
Treating a successful constrained fit as proof of equality. Limited data may be unable to reveal a true species difference. A compact fit is useful only to the extent that its assumptions are plausible and its predictions are adequate.

The Emax-Hill model is a useful summary, not a guarantee that the response is governed by a single instantaneous concentration-response relationship. Its parameters may be hard to interpret if the endpoint, exposure measure, or response timing differs across species. The practical conclusion should therefore be conditional: a parameter may be transported under stated assumptions and supported by available evidence, rather than presumed universal.

Check Your Understanding

Answer each question, then compare with the brief solutions.

In the Emax-Hill model, what response does $EC_{50}$ predict?
For $E_{\max}=80$ response units, $EC_{50}=4$ concentration units, $h=1$, and $C=4$, calculate the predicted response.
Two species have a shared maximum and shared $EC_{50}$, but different Hill coefficients. Can they have the same predicted response at the midpoint and different responses away from it?
Why is it weak evidence for a shared $E_{\max}$ if neither species has observations near a plateau?
What does a shared $EC_{50}$ assumption mean, and why should concentration units be checked before making it?

Solutions. (1) At $C=EC_{50}$, the Hill fraction is $1/2$, so the predicted response is $E_{\max}/2$. (2) $E(4)=80\times4/(4+4)=40.00$ response units. (3) Yes. Both curves are at half their maximum at the shared midpoint, while different $h$ values can give different predictions elsewhere, as the worked example demonstrates. (4) Without near-plateau observations, the maximum may be extrapolated and poorly distinguished from other combinations of maximum, potency, and slope. Similar fitted maxima under that condition do not establish equality. (5) It sets the species' half-maximum concentrations equal. The comparison is interpretable only when concentration is measured on comparable scales and expressed in consistent units; otherwise, apparent equality or difference may be an artifact of the scale.