IVIVE: In Vitro-In Vivo Extrapolation of Hepatic Clearance

In vitro-in vivo extrapolation (IVIVE) is the set of methods pharmacokineticists use to convert an intrinsic clearance value measured in a test tube — from human liver microsomes or suspended/cryopreserved hepatocytes — into a predicted whole-body hepatic clearance in a living human, before a single dose has ever been given to a person. It sits at the center of early drug development: a compound’s predicted human clearance from IVIVE feeds candidate triage, first-in-human dose selection alongside NOAEL-based safety margins, and PBPK model parameterization submitted to regulators. The method is genuinely useful and widely used — but it is also a genuinely imperfect one, with a documented, actively researched tendency to systematically under-predict clearance for some drug classes. Treating an IVIVE number as a precise forecast rather than a directional estimate with known failure modes is one of the more common ways the method gets misapplied.

Scaling In Vitro Intrinsic Clearance to a Whole Liver

Liver microsome and hepatocyte incubations measure intrinsic clearance (CL_int) in units tied to the incubation itself — microliters per minute per milligram of microsomal protein (μL/min/mg protein) for microsomes, or per million cells (μL/min/10⁶ cells) for hepatocytes. Neither unit means anything at the level of a whole organ until it is scaled up by physiological quantities that convert “per milligram of protein” or “per million cells” into “per whole liver.” Three scaling factors do that work:

• Microsomal protein per gram liver (MPPGL) converts a microsomal CL_int into a per-gram-liver value. Reported MPPGL values vary considerably across studies and individual donor livers — commonly cited averages sit roughly in the 30–45 mg protein/g liver range, though widely cited cross-study compilations of the literature document meaningfully wider spread than any single-number summary implies, and MPPGL is known to decline with donor age.
• Hepatocellularity plays the same role for hepatocyte-derived CL_int, converting a per-million-cells value into a per-gram-liver value. Literature values are typically quoted in the broad vicinity of 100–120 × 10⁶ hepatocytes per gram of liver, again with real inter-study and inter-donor variability rather than a single fixed constant.
• Liver weight then converts the per-gram value to a whole-organ value. A commonly used reference figure is on the order of 20–25 g of liver per kilogram of body weight — roughly 1.5–1.8 kg of liver for a 70 kg adult — though actual liver weight scales with body size and shifts with age and hepatic disease.

Multiplying an in vitro CL_int (per mg protein or per 10⁶ cells) by the relevant protein/cellularity factor, then by total liver weight, yields a whole-liver intrinsic clearance — the input the liver models below actually use. Because each scaling factor carries real biological variability, and different laboratories and published compilations report somewhat different central values, the choice of scaling factor is itself a documented source of variability between IVIVE predictions generated for the same compound in different labs. That is worth stating explicitly whenever a predicted clearance number is reported, rather than treating the scaled CL_int as one unambiguous quantity.

Well-Stirred vs Parallel-Tube: Two Different Liver Models

A scaled whole-liver CL_int still is not a hepatic clearance — it describes only the liver’s intrinsic metabolic capacity, independent of how much drug-carrying blood actually reaches the hepatocytes. Converting it into a hepatic clearance, CL_H, that can be compared against a clinical value requires a model of how blood flows through, and equilibrates with, the liver. Two models dominate the IVIVE literature, and they encode genuinely different physiological assumptions about that flow.

The well-stirred (venous equilibrium) model treats the liver as a single, instantaneously and completely mixed compartment, so drug concentration everywhere inside the liver equals the concentration leaving in the hepatic vein:

CL_H = Q_H × f_u,b × CL_int / (Q_H + f_u,b × CL_int)

where Q_H is hepatic blood flow and f_u,b is the unbound fraction of drug in blood (see blood-to-plasma ratio for how f_u,b is derived from a plasma-measured unbound fraction).

The parallel-tube (sinusoidal perfusion) model instead treats the liver as a set of parallel tubes with no back-mixing along their length, so drug concentration declines continuously — more like plug flow — as blood traverses the sinusoid:

CL_H = Q_H × (1 − exp(−f_u,b × CL_int / Q_H))

At low intrinsic clearance relative to hepatic blood flow — low-extraction drugs — the two models converge and give very similar predictions. As intrinsic clearance rises toward and past hepatic blood flow — high-extraction drugs — they diverge, with the parallel-tube model generally predicting a somewhat higher CL_H for the same inputs, because it does not allow already-depleted blood near the sinusoid outlet to dilute back with fresher blood the way an instantaneously mixed compartment implicitly does. The well-stirred model remains the default across most published IVIVE work, largely for its computational simplicity and long track record, even though the parallel-tube model is often considered more mechanistically faithful to the liver’s actual sinusoidal architecture; a third, dispersion model sits between the two and is used less often in practice. Which model is used is not a minor implementation detail — for a high-extraction compound, the choice can materially shift the predicted clearance and, downstream, the predicted oral bioavailability.

The Well-Documented Problem: IVIVE Often Under-Predicts Clearance

None of the scaling or modeling steps above fixes what is probably the single most consequential limitation of hepatic IVIVE: for a substantial subset of compounds — particularly those metabolized extensively by CYP3A4 and other high-turnover enzymes — IVIVE-predicted clearance from liver microsomes systematically under-predicts the clearance actually observed in vivo, sometimes by several-fold. This is not a fringe observation; it is one of the most heavily cited limitations in the drug metabolism and pharmacokinetics literature, traced back to influential late-1990s and early-2000s comparisons of predicted versus observed human clearance across panels of marketed drugs, and it remains an active area of methodological research rather than a solved problem.

Several mechanisms are discussed in the literature as contributors, and more than one typically applies to a given compound:

• Nonspecific binding inside the incubation itself. Microsomal and hepatocyte suspensions contain lipid and protein that can bind drug nonspecifically, much as plasma proteins do. If the unbound fraction in the incubation (f_u,inc) is not measured and corrected for, the calculated intrinsic clearance understates the true unbound intrinsic clearance, and that error propagates directly into an under-predicted CL_H.
• Loss of active hepatic uptake transporters. Isolated microsomes contain no functioning uptake transporters at all, and even suspended hepatocytes can under-represent sustained transporter activity (for example, OATP1B1/1B3-mediated uptake) relative to more physiologically intact systems such as sandwich-cultured or plated hepatocytes. For transporter-dependent compounds, that gap alone can account for meaningful under-prediction.
• Enzyme and cofactor stability over the incubation. Some metabolic activity, and the cofactor systems supporting it, can degrade over the timescale of a standard incubation, particularly for low-turnover compounds that need longer incubations to generate a measurable signal in the first place.
• Incomplete representation of the relevant pathway. A microsomal system captures CYP-mediated, and with added cofactors some UGT-mediated, metabolism, but not every contributing pathway — and it captures nothing happening outside the liver.

The field’s response has been pragmatic rather than a single fix: empirical scaling factors — numeric correction multipliers, sometimes derived per CYP isoform, proposed by several DMPK research groups working on this specific gap — are widely used as a post-hoc correction rather than a mechanistic solution, and improved in vitro systems (extended or “relay” hepatocyte incubations for low-clearance compounds, co-cultured or sandwich-cultured hepatocyte formats intended to better preserve transporter function) remain active areas of ongoing methods development aimed specifically at this gap. None of these fully closes it for every compound class, and the degree of under-prediction is compound- and pathway-dependent rather than correctable with one universal factor. Any IVIVE-derived clearance prediction — especially one used to inform a first-in-human dose or a drug-drug interaction risk assessment — should be read with this limitation stated explicitly, not silently assumed away.

Where IVIVE Sits in Drug Development

In early discovery, IVIVE is used less for its absolute numeric accuracy than for its relative ranking value — comparing predicted metabolic stability across a series of candidate compounds to triage which are worth advancing, well before the precision of any single predicted clearance value matters much. Later in development, IVIVE-derived clearance becomes one input, typically alongside allometric scaling from animal PK and increasingly full physiologically based pharmacokinetic (PBPK) modeling, feeding first-in-human starting dose selection — a decision that also has to respect the safety margin defined by NOAEL-based dosing from nonclinical toxicology, and that draws on the same toxicokinetic exposure data used to bridge animal and human predictions. IVIVE-based reaction-phenotyping and inhibition data are also a standard part of the in vitro drug-drug interaction risk assessments included in regulatory submissions, and the same clearance and unbound-fraction terms are downstream inputs to population PK and dose-optimization modeling once clinical data exist.

Because oral bioavailability for a high-extraction drug is dominated by hepatic first-pass metabolism, an IVIVE-predicted extraction ratio also connects directly to how a compound’s absorption behavior gets classified — see the Biopharmaceutics Classification System for how permeability and solubility fit alongside metabolic clearance in that picture.

Interpreting an IVIVE Prediction Responsibly

A predicted clearance value on its own, without the context behind it, is not fully interpretable. Before relying on one, it is worth confirming: which liver model (well-stirred, parallel-tube, or dispersion) generated it, since the choice materially affects high-extraction compounds; which scaling factors (MPPGL or hepatocellularity, liver weight) were used, and whether they reflect a documented literature source rather than an unstated default; whether an unbound-fraction correction was applied correctly on a blood basis rather than left as a plasma value; whether an empirical scaling correction was applied and on what basis; and whether the prediction was cross-checked against an orthogonal method — allometric scaling from animal data, or an independent in vitro system — rather than relied on alone. Given that under-prediction is a documented, compound-class-dependent phenomenon rather than a rare edge case, a single IVIVE-derived clearance number, however carefully generated, is best treated as an estimate with a known, directionally biased failure mode for certain compound classes, not a substitute for clinical PK data once it becomes available.

Frequently Asked Questions

What is IVIVE in pharmacokinetics?

In vitro-in vivo extrapolation is the process of converting an intrinsic clearance measured in liver microsomes or hepatocytes into a predicted in vivo hepatic clearance, using physiological scaling factors and a liver model such as the well-stirred or parallel-tube model.

What is intrinsic clearance (CL_int)?

CL_int measures the liver’s inherent metabolic capacity to eliminate a compound, independent of blood flow or protein binding — it is what is measured directly in a microsomal or hepatocyte incubation, before any scaling or liver-model conversion is applied.

Should I use liver microsomes or hepatocytes for IVIVE?

Both are standard systems with real trade-offs: microsomes capture CYP-mediated (and, with added cofactors, some UGT-mediated) metabolism cheaply and reproducibly but contain no transporters and no cytosolic enzymes, while hepatocytes retain a fuller complement of metabolic pathways and some transporter function, at greater cost and variability. Many programs run both and compare the resulting predictions.

Well-stirred or parallel-tube model — which should I use?

For low-extraction compounds the choice barely matters, since the two models converge. For high-extraction compounds it matters more; the well-stirred model remains the default in most published work for its simplicity, while the parallel-tube model is often considered more physiologically realistic. Reporting which model was used matters more than declaring one universally “correct.”

Why does IVIVE tend to under-predict clearance?

Several documented mechanisms contribute — most commonly incomplete correction for nonspecific binding in the in vitro incubation, loss of active transporter function in isolated systems, and incomplete representation of every contributing metabolic pathway — and the degree of under-prediction varies by compound and enzyme pathway rather than following one universal correction factor.

Is an IVIVE prediction reliable enough to set a human dose on its own?

Not on its own. It is one input among several — typically alongside allometric scaling and PBPK modeling — feeding a first-in-human dose decision that also has to respect nonclinical safety margins. Given the documented under-prediction issue, treating a single IVIVE number as a precise, standalone forecast is a known way the method gets misused.

For the wider clinical-research operations and regulatory context this fits into, see CASRAI’s clinical research hub.