Radioligand design, receptor occupancy, and compartmental modeling — from kon/koff to clinical BPND
Before a single atom is made radioactive, a PET tracer must first exist as a "cold" (non-radioactive) chemical entity that binds its target receptor with extraordinary precision. Robert Innis and colleagues formalized the empirical rules that separate a useful clinical radioligand from the thousands of candidate compounds that fail — rules built from decades of trial and error across dopamine, serotonin, opioid, and amyloid imaging programs.
A PET radioligand is injected in tracer-dose quantities — typically less than 10 µg of mass for the entire scan, occupying far less than 1% of available receptor sites so that the biology being imaged is not perturbed by the imaging agent itself. This "tracer principle" only holds if the compound binds with very high affinity: at sub-nanomolar Kd, picomolar-to-low-nanomolar tracer concentrations are still sufficient to generate detectable, receptor-saturable signal above the noise floor of a PET camera.
If Kd is too high (low affinity, e.g. > 10 nM), the injected mass needed to see specific binding becomes large enough to occupy a pharmacologically significant fraction of receptors, violating the tracer principle and biasing the very occupancy or density measurement the scan is meant to provide.
Affinity is measured in vitro long before any radiochemistry: competitive radioligand displacement assays against a reference tracer, saturation binding assays fitting a Scatchard/Rosenthal plot, or surface plasmon resonance on recombinant receptor. Only compounds clearing the sub-nanomolar bar advance to selectivity profiling.
Mintun, Raichle and colleagues (1984) formalized the tracer-kinetic definition of "binding potential" (BP = Bmax/Kd) that still underlies every modern PET occupancy study — the ratio of receptor density to dissociation constant, not either quantity alone, is what a PET scan actually measures.
A candidate is counter-screened against a panel of 40–80 unrelated receptors, transporters, and ion channels (CEREP/Eurofins-style broad panels) to rule out off-target binding that would appear as confounding signal anywhere it occurs in the body. For CNS tracers, closely related receptor subtypes are the most dangerous off-targets:
• Dopamine D2/D3 tracers ([¹¹C]raclopride, [¹⁸F]fallypride) must not cross-react with D1-family receptors or serotonin 5-HT2A, which share overlapping pharmacophores • Serotonin transporter (SERT) tracers ([¹¹C]DASB) must discriminate against the closely related norepinephrine (NET) and dopamine (DAT) transporters, all of which co-localize in some brain regions • Amyloid tracers ([¹⁸F]florbetapir, [¹¹C]PiB) must avoid binding tau, alpha-synuclein, and white-matter myelin, historically a major source of off-target signal
A selectivity ratio > 100-fold over the nearest off-target is the conventional threshold; many clinically successful tracers exceed 1,000-fold.
Passive diffusion across the blood-brain barrier requires enough lipophilicity to cross lipid membranes, but not so much that the compound gets trapped by non-specific, low-affinity interactions with membrane lipids and proteins throughout the brain — a phenomenon that raises background signal and lowers the specific-to-non-specific binding ratio.
Empirically, calculated logP (or logD at physiological pH 7.4) between 1 and 3 balances these opposing constraints:
• logP < 1: poor membrane permeability, low brain uptake, K1 too low for adequate signal • logP 1–3: the productive window — sufficient permeability, manageable non-specific binding • logP > 3: high non-specific ("non-displaceable") binding swamps the specific signal, and P-glycoprotein efflux at the BBB often increases, further reducing net brain uptake
Molecular weight under ~450 Da and fewer than 3–4 hydrogen-bond donors (Lipinski-like heuristics adapted for CNS penetration) further constrain the chemical space explored during lead optimization.
Once a chemical lead clears the affinity, selectivity, and lipophilicity filters, it must be radiolabeled with a positron-emitting isotope, synthesized fast enough to outrun radioactive decay, and formulated for intravenous injection — all within a time window measured in minutes for carbon-11 and closer to an hour for fluorine-18.
Positron-emitting isotopes decay by emitting a positron that annihilates with a nearby electron, producing two 511 keV photons in near-opposite directions that the PET scanner detects in coincidence. Two isotopes dominate CNS receptor imaging:
Carbon-11 (t½ = 20.4 min): • Can label a methyl or methoxy group directly on the native drug-like scaffold via [¹¹C]methyl iodide or [¹¹C]CO2, often preserving the exact pharmacophore validated in vitro • Short half-life forces synthesis, purification, quality control, and injection within roughly 45 minutes, demanding an on-site cyclotron and rapid automated synthesis modules • Ideal for same-day test-retest or multiple-tracer studies in one subject, since the isotope decays away before the next injection
Fluorine-18 (t½ = 109.8 min): • Longer half-life allows synthesis, quality control, and shipping to satellite imaging sites tens to hundreds of kilometers from the cyclotron • Requires incorporating a fluorine atom, sometimes via a fluoroethyl or fluoropropyl linker not present in the original lead compound — occasionally altering affinity or lipophilicity enough to require re-optimization • Dominates commercial and multi-site clinical trial use (e.g., [¹⁸F]FDG, [¹⁸F]florbetapir, [¹⁸F]fallypride) because of its logistical flexibility
Radiochemical yield, radiochemical purity (>95% required for human use), and molar activity are verified by HPLC before every injection — a synthesis that misses the purity or activity specification is discarded, since the short isotope half-life leaves no time to reformulate.
Molar activity (specific activity) is the radioactivity per mole of compound — a mixture of radiolabeled ("hot") and unlabeled ("cold") molecules of identical chemical structure. Low molar activity means more cold carrier molecules must be injected to reach a target radioactivity dose, and those cold molecules compete with the hot tracer for the same finite pool of receptor sites.
High molar activity (>1,000 Ci/mmol, ideally >5,000 Ci/mmol for very high-affinity, low-density targets) keeps injected mass in the low-microgram range, satisfying the tracer principle: receptor occupancy by the tracer itself stays below ~1–5%, so the measured signal reflects the biological system at rest rather than a pharmacological perturbation caused by the scan.
Injected radioactivity is typically 5–20 mCi (185–740 MBq), chosen to balance count statistics (image quality) against radiation dosimetry limits for human research subjects, governed by regulatory dose limits (e.g., 21 CFR 361.1 in the US for research use).
The purified radiotracer is formulated in a sterile, pyrogen-free, isotonic solution (typically saline with a small ethanol co-solvent fraction) and passed through a sterilizing filter immediately before injection. Quality control confirms:
• Radiochemical identity and purity by analytical HPLC (co-injection with authentic cold standard) • Residual solvent levels below pharmacopeial limits • Sterility and endotoxin testing (the latter via a rapid limulus amebocyte lysate assay, since full sterility culture takes days — longer than the isotope survives) • Molar activity calculated from the ratio of measured radioactivity to UV-quantified total mass
The dose is administered as a bolus or bolus-plus-infusion intravenous injection, timed to the start of dynamic PET acquisition so the arterial input function and brain time-activity curves are captured from time zero.
After intravenous injection, the tracer must survive plasma protein binding and peripheral metabolism, cross the blood-brain barrier as free (unbound) drug, and reach brain tissue in high enough concentration to generate a measurable signal — all while an arterial or image-derived input function is sampled to anchor the kinetic model.
Immediately after injection, a fraction of the tracer binds reversibly to plasma proteins (albumin, α1-acid glycoprotein) and is transiently unavailable for tissue uptake. Only free, unbound tracer is assumed capable of crossing the blood-brain barrier and participating in receptor binding — a foundational assumption of the standard compartmental model.
The plasma free fraction, fP, is measured ex vivo by ultrafiltration or equilibrium dialysis of a plasma sample spiked with the radiotracer, and is typically 2–20% for lipophilic CNS radioligands. fP enters directly into the calculation of the total distribution volume (VT) and, for some analyses, into deriving Bavail/Kd from K1/k2 ratios — small errors in fP measurement propagate directly into quantification error, making this an important and sometimes underappreciated source of test-retest variability.
Unlike glucose (which requires GLUT transporters) or many drugs (which require active transport), most CNS PET tracers cross the blood-brain barrier by simple passive diffusion, governed by the Renkin-Crone capillary permeability formalism:
K1 = F · (1 − e^(−PS/F))
Where F is regional cerebral blood flow, and PS is the permeability-surface-area product, itself driven by lipophilicity (logP), molecular weight, hydrogen-bonding capacity, and — for some tracers — susceptibility to active efflux by P-glycoprotein (P-gp) at the luminal endothelial membrane.
For small, moderately lipophilic tracers (logP 1–3, MW < 450), PS is large relative to flow, so K1 approaches flow-limited behavior (K1 ≈ F) — meaning the tracer reaches equilibrium with tissue quickly enough that regional differences in blood flow, not permeability, dominate early uptake. Tracers that are P-gp substrates show markedly reduced brain uptake despite favorable logP, a common and sometimes late-discovered failure mode in tracer development.
P-glycoprotein efflux was a major, initially unrecognized reason several promising opioid and antidepressant-target tracers failed in humans after performing well in rodents — P-gp expression and substrate specificity differ enough between species that primate or human-tissue validation is now considered essential before a first-in-human PET study.
Quantitative kinetic modeling requires knowing the time course of parent (unmetabolized) tracer concentration in arterial plasma — the "input function" that drives tissue uptake. This is measured by:
• Frequent arterial blood sampling (often every 5–15 seconds initially, spacing out to every several minutes later in the scan) via a radial arterial line • Rapid automated well-counting of whole blood and plasma radioactivity • HPLC separation of parent tracer from radiolabeled metabolites at selected time points, since many tracers are metabolized in the liver into polar metabolites that do not cross the BBB but do circulate in plasma and would otherwise be mistaken for active tracer
The measured total plasma curve is multiplied by the HPLC-derived parent fraction (fit with a sigmoid or Hill-type function of time) to yield the true metabolite-corrected arterial input function. Reference-tissue methods (covered in Stage 5) were developed largely to avoid the discomfort, cost, and complexity of arterial cannulation and metabolite analysis when a receptor-free reference region is anatomically available.
At the molecular level, PET receptor imaging is a real-time, whole-body radioligand binding assay. Free tracer molecules diffusing through the synaptic neuropil associate with unoccupied receptor binding sites at a rate governed by kon, and previously bound molecules dissociate back into the free pool at a rate governed by koff — the same second-order/first-order kinetics measured in vitro decades earlier by Solomon Snyder, Candace Pert, and the founders of modern receptor pharmacology.
Receptor binding is modeled as a reversible bimolecular reaction between free tracer (F) and unoccupied receptor sites (Bavail − B, where B is already-bound receptor and Bavail is total available density):
d[B]/dt = kon · [F] · ([Bavail] − [B]) − koff · [B]
At equilibrium, d[B]/dt = 0, giving the classic Michaelis-Menten-like saturation relationship:
B = (Bavail · F) / (Kd + F), where Kd = koff / kon
This is the same mathematics underlying enzyme-substrate kinetics, antibody-antigen binding, and every other reversible bimolecular interaction in biochemistry — PET imaging is distinctive only in that the "assay" is running inside a living human brain, non-invasively, in real time.
Because tracer concentration [F] is kept far below Kd (the tracer principle again), the system operates in the linear region of the saturation curve, where bound tracer is approximately proportional to Bavail/Kd — the quantity PET actually measures, called binding potential.
Total tissue radioactivity measured by PET is the sum of three pools:
• Free tracer (F): unbound tracer in tissue water, in rapid exchange with plasma • Specifically bound tracer (B): tracer bound to the target receptor with saturable, displaceable, high-affinity kinetics • Non-specifically bound tracer (NS): low-affinity, non-saturable association with lipid membranes and non-target proteins, proportional to tracer concentration but not to receptor density
Only the specific binding compartment carries information about the receptor of interest. Non-specific binding is estimated using a receptor-free reference region (e.g., cerebellum for many dopamine and serotonin tracers, which is nearly devoid of the target) or by pharmacological blocking studies with a large dose of unlabeled, high-affinity competitor that displaces essentially all specific binding, leaving only F + NS.
A tracer's "signal-to-noise" for receptor imaging is largely set by its specific-to-non-specific binding ratio at equilibrium — the reason logP optimization (Stage 1) and non-specific binding characterization are inseparable parts of tracer development.
Displacement studies — injecting a pharmacological dose of unlabeled competitor mid-scan and watching bound tracer washout accelerate — directly demonstrate that binding is specific, saturable, and reversible, and are a standard validation experiment for any new radioligand.
The dissociation rate koff determines how quickly a tracer can "sense" a change in receptor occupancy caused by an administered drug, and constrains scan duration and study design:
• Fast koff (short receptor residence time, t½ = ln2/koff of a few minutes): tracer reaches transient equilibrium quickly, enabling shorter scans and same-day test-retest studies, but the tracer is more vulnerable to being displaced by fast plasma clearance, reducing peak specific signal • Slow koff (long residence time, t½ of tens of minutes to hours): higher retained specific signal and better image contrast late in the scan, but the system may not reach true equilibrium within a practical scan duration, requiring full kinetic modeling (rather than simple equilibrium ratio methods) to extract accurate Kd and Bavail
This tradeoff is why kinetic rate constants — not just equilibrium Kd — are reported and optimized during tracer characterization: two tracers can share an identical Kd while having very different kon/koff pairs and therefore very different practical imaging properties.
A dynamic PET scan produces a time series of regional radioactivity concentration — the tissue time-activity curve (TAC) — for every voxel or region of interest. Converting these curves into physiologically meaningful, comparable numbers (K1, k2, k3, k4, VT, BPND) requires fitting compartmental models or applying graphical (linearized) analyses validated against them.
The standard kinetic model divides the system into linked compartments connected by first-order rate constants, fit by nonlinear regression to the measured tissue TAC given the arterial input function Cp(t):
One-tissue compartment model (1TCM): • Plasma → Free+NS tissue compartment, rate K1 in, k2 out • Used for tracers with fast equilibration and no resolvable specific binding compartment (e.g., [¹⁸F]FDG uses a related but distinct model for glucose metabolism)
Two-tissue compartment model (2TCM): • Plasma → Free+NS compartment (K1, k2) → Specifically bound compartment (k3, k4) • k3 reflects the pseudo-first-order association rate (kon · Bavail, since [F] is far below Kd) and k4 = koff • The gold-standard model for most receptor tracers when arterial input is available; fit by weighted nonlinear least squares
From the fitted rate constants, total distribution volume VT = (K1/k2)(1 + k3/k4), and binding potential relative to plasma BPP = VT − VND (where VND is the non-displaceable distribution volume, measured in a receptor-free region or by blocking).
Arterial cannulation is invasive, uncomfortable, and logistically demanding. When a brain region devoid of the target receptor exists (a reference region), its TAC can substitute for the arterial input function, because it shares the same free+non-specific kinetics (K1/k2 ratio, i.e. VND) as target regions but lacks the specific-binding compartment.
The Simplified Reference Tissue Model (SRTM, Lammertsma & Hume 1996) assumes the target region can be adequately described by a single tissue compartment relative to the reference region TAC CR(t), fit for three parameters: R1 (= K1/K1′, relative delivery), k2, and BPND directly — without any blood sampling:
CT(t) = R1·CR(t) + [k2 − R1·k2/(1+BPND)] · CR(t) ⊗ e^(−k2t/(1+BPND))
SRTM (and its computationally faster linearized cousin, SRTM2) is the dominant analysis method for clinical dopamine, serotonin, and opioid receptor PET studies precisely because it eliminates arterial sampling while recovering BPND with accuracy comparable to full 2TCM in validated systems.
Binding potential relative to non-displaceable uptake, BPND = Bavail/Kd, is the single most widely reported outcome measure in receptor PET — a dimensionless ratio directly proportional to receptor density and inversely proportional to affinity, robust to many sources of measurement noise that would otherwise confound absolute Bmax or Kd estimation individually.
Graphical methods linearize the kinetic equations after an initial equilibration period, trading some statistical efficiency for computational simplicity, robustness to noise, and model-independence:
Logan plot (reversible binding, Logan et al. 1990): • Plots ∫₀ᵗCT(τ)dτ / CT(t) against ∫₀ᵗCp(τ)dτ / CT(t) • Becomes linear after transient equilibrium is reached; the slope of the linear portion directly equals the total distribution volume VT • A reference-tissue version (Logan et al. 1996) substitutes CR(t) for the plasma input, yielding a slope equal to (1 + BPND) without any blood sampling
Patlak plot (irreversible/trapped tracers, e.g. [¹⁸F]FDG, [¹⁸F]FDOPA): • Appropriate for tracers with negligible k4 (essentially irreversible trapping over the scan duration) • Plots CT(t)/Cp(t) against ∫₀ᵗCp(τ)dτ / Cp(t); the slope gives the net influx constant Ki
Both methods require the system to have reached "transient equilibrium" (constant apparent distribution volume ratio) before the plotted points become linear — using data before this point biases the fitted slope, a common analytic pitfall in early-phase tracer studies.
| Product | Indication | Trial Design | Key Result |
|---|---|---|---|
| [¹¹C]Raclopride | Dopamine D2/D3 receptor | Moderate affinity (Kd ≈ 1–3 nM), fast kinetics; sensitive to endogenous dopamine competition | Gold standard for measuring dopamine release (drug challenge paradigms) |
| [¹⁸F]Fallypride | Dopamine D2/D3 receptor | High affinity (Kd ≈ 0.03 nM), slow kinetics; quantifies low-density extrastriatal D2/D3 | Extends D2/D3 imaging beyond striatum into cortex and thalamus |
| [¹¹C]DASB | Serotonin transporter (SERT) | High affinity, high selectivity vs. NET/DAT; robust reference-tissue kinetics | Standard tracer for antidepressant SERT-occupancy studies |
| [¹⁸F]Florbetapir | Amyloid-β plaques | Binds fibrillar amyloid; non-receptor target but same BPND/SUVR quantification framework | FDA-approved for Alzheimer's disease amyloid status determination |
The clinical payoff of a validated PET radioligand is the receptor occupancy study: measuring how much of a candidate drug's target is engaged at a given dose in living human brain, directly informing the dose selected for pivotal efficacy trials — a translational bridge from radiochemistry to central nervous system drug development that few other technologies can provide.
Fractional receptor occupancy by an unlabeled drug is calculated by comparing BPND measured at baseline (drug-free) to BPND measured after drug administration, in the same subject or matched cohort:
Occupancy (%) = [1 − BPND(drug) / BPND(baseline)] × 100
This rests on the assumption that the drug and radioligand compete for the same binding site (or that the drug reduces Bavail available to the tracer through an equivalent mechanism), and that non-displaceable binding (VND) is unchanged by the drug — an assumption tested during radioligand validation.
Occupancy measured across a dose range is fit with an Emax (sigmoidal) model:
Occupancy(Dose) = Emax · Dose / (ED50 + Dose)
where ED50 is the dose producing half-maximal occupancy. This dose-occupancy curve, not the dose-plasma-concentration curve alone, is what regulators and clinical teams use to select doses expected to achieve a target occupancy window associated with efficacy while avoiding occupancy levels associated with dose-limiting side effects.
Decades of PET occupancy studies across drug classes have established empirical occupancy windows associated with clinical efficacy and tolerability — arguably the single most consequential quantitative contribution of receptor PET to drug development:
• Antipsychotics (D2/D3 antagonists): therapeutic efficacy generally requires ~65% striatal D2 occupancy; extrapyramidal side effects (parkinsonism, akathisia) become common above ~80% occupancy — a narrow ~15-percentage-point therapeutic window that PET directly visualizes and that guided the development of "atypical" antipsychotics with faster D2 dissociation (higher koff) to reduce sustained peak occupancy • SSRIs (SERT inhibitors): clinical antidepressant efficacy is associated with SERT occupancy > 80%, information derived directly from [¹¹C]DASB occupancy studies that helped establish minimum effective doses • Opioid antagonists (naltrexone-class): mu-opioid receptor occupancy above ~90% is associated with blockade of exogenous opioid effects, informing dosing for relapse-prevention indications
Because occupancy PET directly measures target engagement in the organ of interest, it lets a drug development program distinguish "the drug failed because it didn't reach the target at an adequate dose" from "the drug reached the target but the target hypothesis itself was wrong" — a distinction that has redirected or terminated numerous CNS drug programs earlier and more cheaply than waiting for a failed Phase 2 efficacy trial.
The full translational path from a novel PET tracer idea to routine clinical or drug-development use typically spans 8–15 years:
1. Target validation and medicinal chemistry lead optimization (1–3 years): affinity, selectivity, logP screening as in Stage 1 2. Radiolabeling feasibility and non-human primate PET (1–2 years): confirms brain penetration, specific binding, reversibility, and absence of confounding radiometabolites entering brain 3. Radiation dosimetry and IND/IMPD-enabling studies (6–12 months): estimates human radiation exposure from biodistribution in animals, sets maximum allowable injected activity 4. First-in-human PET (6–12 months): confirms brain kinetics, test-retest reproducibility, and optimal scan duration/model in a small healthy cohort 5. Validation against gold-standard kinetic modeling and reference-tissue methods (ongoing): establishes which simplified analysis (SRTM, Logan) is appropriate for routine use 6. Deployment in occupancy and disease-biomarker studies (multi-year, ongoing): the tracer becomes a working tool for CNS drug development programs and, occasionally, an FDA/EMA-recognized diagnostic (as with amyloid and tau PET tracers in Alzheimer's disease)
Of the many hundreds of PET radioligands synthesized and tested in academic laboratories, only a small fraction ever reach routine multi-site clinical or pharmaceutical-industry use — the tracers profiled throughout this simulation ([¹¹C]raclopride, [¹⁸F]fallypride, [¹¹C]DASB) represent the rare successes that cleared every stage of this pipeline.