Beyond Simple Allometric Scaling: Why Time-Course Methods Win for First-in-Human Predictions

A mouse and a human do not run on the same clock. Give both a bolus dose of methotrexate, and the drug’s half-life is roughly five times longer in the human than in the mouse. The chronological time is wildly different. The pharmacokinetic time, once you know how to normalize for it, is not.

That observation, first described by Dedrick, and now formalized in Dedrick plots, is the starting point for empirical time-course scaling: a family of techniques, built on allometric scaling principles, that let translational scientists do something remarkable — take concentration-time data from a handful of preclinical species and predict what the full human concentration-time profile will look like, well before a single dose is given to a person.

The problem with simple allometric scaling

Ask most pharmacometricians what allometric scaling means in practice, and you’ll hear some version of “plotting PK parameters against body weight.” That’s not wrong, but it’s a narrow slice of a much older idea. Allometric scaling describes how biological size and biological function relate to one another, and in pharmacokinetics, it usually shows up as two numbers: an exponent for clearance, typically close to 0.75, and an exponent for volume of distribution, typically close to 1.

The trouble is that scaling clearance and volume only tells you about two parameters. It doesn’t tell you what the concentration-time profile actually looks like in a human. To go from those two scaled parameters to something usable, you’re often forced to assume the simplest possible shape: a mono-exponential decline. That assumption holds up reasonably well for IV-administered drugs. It holds up much less well for the extravascular drugs that make up most of today’s pipeline, where absorption, distribution, and elimination all interact to shape the curve.

In other words: simple allometric scaling answers “how fast does the body clear this drug?” It doesn’t answer “what will the concentration look like at 6 hours, or 24, or at trough?” And for early dose-selection decisions, that second question is usually the one that matters.

Where the idea comes from

The underlying science predates pharmacokinetics by decades. In the 1930s, Huxley and Tessier described a power-law relationship between the size of an organism and its biological function: Y = b·X^k, where k is the allometric exponent. Around the same time Kleiber found that basal metabolic rate across birds and mammals followed an exponent of about 0.74, remarkably close to the exponent later observed for clearance, which makes sense given that clearance is, fundamentally, a biological rate process. West later extended the relationship from single-celled organisms up through the largest land mammals, and it continued to hold.

By the time these principles made their way into pharmacokinetics, the pattern was clear: clearance tends to scale with roughly the same exponent as metabolic rate, and volume of distribution tends to scale close to 1, i.e., proportional to body weight. Decades of literature, and a lot of day-to-day translational practice, have built on that foundation – the working basis for allometric scaling in PK as it’s applied today. It remains a starting point even inside more mechanistic frameworks: you use what you have, and your models grow in sophistication as more mechanistic information becomes available.

Two ways to scale the full time course using allometric scaling principles

Scaling parameters is one thing. Scaling the entire concentration-time profile is another, and this is where things get genuinely useful for dose prediction.

Dedrick plots. Dedrick’s original insight, using methotrexate across species, was that if you normalize the time axis by body weight raised to some power (a “species-invariant time”) and normalize concentration by body weight, profiles from very different species can become superimposable. Boxenbaum later described two variants of this transformation. The kallynochron assumes a clearance exponent of 0.75 and a volume exponent of 1. The apolysichron instead uses the exponents actually estimated from your data, which matters because volume doesn’t always scale with an exponent of exactly 1 in practice.

Dedrick plots are elegant, but they carry a built-in limitation worth stating plainly: they rely entirely on allometric scaling. Their predictive performance can never exceed the predictive performance of the allometric relationship underneath them.

The Wajima superposition method. Published in 2004, the Wajima approach extends the same normalization logic but removes the dependency on strict allometric exponents. Instead of assuming clearance and volume must scale at 0.75 and 1, Wajima normalizes concentration by steady-state concentration (dose divided by volume of distribution) and normalizes time by mean residence time (volume divided by clearance). Once you have that normalized, superimposed curve, you can back-transform to a human profile using clearance and volume estimates from any source: simple allometric scaling, PBPK modeling, in vitro-in vivo extrapolation, or some blend of methods. The scaling itself is still empirical, and you still need at least one animal species, but you’re no longer locked into IV-only data or a fixed exponent.

Choosing between them. In practice, the two methods are worth running side by side. When Dedrick and Wajima superposition predictions agree, that’s a reassuring signal. When they diverge, usually in the terminal phase, it’s a prompt to dig into why, rather than a reason to distrust the whole exercise.

The workflow in Phoenix

Certara’s translational science team built a repeatable workflow inside Phoenix that takes this from theory to a working human prediction in a handful of steps.

1. Start with the right dataset. At minimum, you need concentration-time data from two preclinical species, ideally three, dosed IV. Total or unbound concentrations both work, depending on how protein binding varies across species. Each animal or profile needs a unique ID, an observed body weight, and dose recorded in consistent units.
2. Explore the data first. Before any scaling happens, by-animal plots and descriptive statistics (both raw and dose-normalized) surface the issues that would otherwise quietly undermine the rest of the workflow, for example missing samples, an outlier species, non-dose-proportional exposure.
3. Run an NCA to get your allometric exponents. A non-compartmental analysis on each animal (or profile, if pooling data) yields clearance and volume estimates. Log-transforming those and regressing them against log-transformed body weight gives you the allometric exponents directly, visually confirmed on an XY plot with a fitted regression line. In a representative dataset, this comes out to roughly 0.75 for clearance and closer to 0.9–1 for volume, right where theory predicts.
4. Build the Dedrick overlay. Using the exponents from step 3, normalize concentration and time (via equivalent time, kallynochron, or apolysichron definitions) and check whether the resulting profiles superimpose across species. When they do, back-transforming to a chosen human body weight and dose is a matter of a few data-wizard calculations.
5. Build the Wajima superposition overlay. Normalize concentration and time by steady-state concentration and mean residence time instead, then merge in a predicted human clearance and volume, whether from allometric scaling or another method, and back-calculate the human profile the same way.
6. Compare the two. Overlaying both sources of human predictions on one plot is often the most useful single output of the exercise. Agreement builds confidence; disagreement tells you where to look closer.
7. Extend beyond a single IV dose. Real first-in-human studies are rarely single-dose IV. Fit the predicted profile to a compartmental model, add an absorption compartment with a rate constant and bioavailability for extravascular dosing (often available in other preclinical species), then simulate across a range of doses and schedules. Overlaying safety and efficacy reference lines — an in vitro IC50/EC50 for efficacy, a NOAEL-derived Cmax for safety — turns the exercise from an academic curve-fitting demonstration into something that directly informs a first-in-human dose decision.

Know the limits

None of this replaces judgment. Dedrick plots are exactly as good as the allometric relationship feeding them, no better. Superimposition doesn’t always come out clean, and when it doesn’t, that’s information, not failure. And these are, by design, empirical methods: useful precisely because mechanistic and PBPK approaches, however fast they’re advancing, still take time to validate against every new chemical modification. When a chemistry backbone shifts to improve half-life or reduce off-target binding, in silico predictions need supporting data to catch up. Empirical time-course scaling remains one of the fastest, most practical bridges available while that validation work happens.

Where this fits

Mechanistic and physiologically based approaches are advancing quickly, and for good reason: they offer a path toward fewer animal studies and a more direct read on human physiology. But traditional time-course scaling isn’t a relic waiting to be replaced. For novel chemotypes moving faster than in silico platforms can be validated against them, it’s often the most defensible tool available for an early dose-selection decision, and it’s built directly into the Phoenix workflow translational scientists already use.