Receptor Selectivity Panels: What a Clean Profile Does and Does Not Show

A receptor selectivity panel produces one of the most persuasive-looking documents in pharmacology: a grid of targets, a column of numbers, and a great many entries reading "no significant activity." The grid invites a conclusion it cannot actually support. This guide is about the distance between what the panel measured and what the reader takes away.

Selectivity and specificity are not interchangeable words

In analytical method validation, specificity is the assertion that the response being measured comes from the intended analyte and nothing else. Receptor pharmacology borrows the same instinct but applies it to the compound rather than to the detector, and the two terms then part company.

Selectivity is a ratio. It compares the compound's potency at one target against its potency at others, and it is always a number with a denominator. Specificity, used strictly, is an absolute claim — that a compound acts at one target and no other. Specificity in that strict sense is unfalsifiable by a panel, because a panel is finite and the proteome is not. Every panel result is therefore a selectivity statement, however it is labeled. This distinction is terminological convention rather than a finding, and usage in the literature is loose enough that a reader should check which sense an author means rather than assume it.

Panel composition is a design decision, not a standard

There is no canonical panel. A review of secondary pharmacology practice notes that target composition, screening technologies, assay formats, interpretation and scheduling of panels can vary significantly between companies in the absence of dedicated guidelines ^[1]. An industry survey of the same field found variation in the choice of assay technology, in whether compounds are tested at a single concentration or several, in the rationale for target and panel selection, and in the threshold at which an interaction is called significant — the authors concluded that there is only loose consensus on the testing paradigm ^[2].

Panels are built by pruning. One published approach began from IC50 values across 123 in vitro pharmacological assays, treated an IC50 at or below 1 µM as a hit, and removed targets whose hit patterns duplicated those of a target with a higher hit rate; the authors reported that 95% of the compounds from the source database were identified within the top 47 ranked assays, and assembled a 50-target panel from that ranking plus three targets added for established safety concerns ^[3]. Redundancy removal is efficient, but it means a surviving target stands in for the ones deleted only to the extent that the historical correlation holds for your compound.

Membership is also partly empirical. One retrospective analysis of measured and predicted in vitro bioactivities against two outcome databases reported 45 protein targets with statistically significant associations, of which 30 were already included on existing safety panels; the remaining 15 included 9 carbonic anhydrases ^[4]. The same analysis reported that when an individual target's bioactivity was used as a predictor, there was a trade-off between positive predictive value and the fraction of the associated compounds that could be detected at all ^[4].

The practical consequence is blunt: a target not on the panel returns no information whatsoever, and a blank cell in a grid of 50 says nothing about the thousands of proteins that were never plated.

Single-point screening: what a percent-inhibition number can carry

Most panels screen at one fixed concentration and report percent displacement or percent inhibition. Under a simple one-site competitive model, percent inhibition relates to IC50 as

percent inhibition = 100 × C / (C + IC50)

which rearranges to IC50 = C × (100 − percent inhibition) / percent inhibition. This is algebra, not an experimental result, and it holds only if the model does.

Take a 10 µM test concentration, the concentration at which the 50-target panel described above was routinely screened ^[3]. A reading of 50% inhibition implies IC50 = 10 × 50/50 = 10 µM. A reading of 40% implies 10 × 60/40 = 15 µM; a reading of 60% implies 10 × 40/60 = 6.67 µM. So a 20-percentage-point band around the midpoint corresponds to a 15/6.67 = 2.25-fold span in IC50. Near the top of the curve the leverage is worse: 85% implies 150/85 = 1.76 µM and 95% implies 50/95 = 0.53 µM, so a 10-point band spans 1.76/0.53 = 3.35-fold. A single-point reading compresses hardest exactly where the compound is most potent.

The same arithmetic sets the detection edge. A 50%-at-10 µM threshold maps precisely onto IC50 ≤ 10 µM under this model, so a target with IC50 = 12 µM reads 100 × 10/22 = 45.5% and is recorded as a miss. Nothing in the report distinguishes that near-miss from a genuine blank.

Whether a blank is real is structurally hard to check. In a screening study where every compound was tested 42 times across 14 concentrations so that true activity could be assigned from a full curve, the authors observed that false-negative rates are ordinarily hard to verify because only compounds selected in the primary screen are followed up, and reported that error rates estimated under a normal-noise assumption did not agree with the actual error rates, because the tails of the noise distribution departed from normality ^[5]. That work concerned high-throughput screening rather than receptor selectivity panels specifically, but the sampling problem it describes applies wherever blanks go unconfirmed.

Whether the assay could have seen a hit at all

Before interpreting a blank, ask whether the well had the dynamic range to produce a signal. The Z-factor is a dimensionless screening-window coefficient that reflects both the assay's signal dynamic range and the variation in the signal measurements, and it was proposed as a tool for comparing and evaluating the quality of screening assays ^[6]. The common working rule that a Z-factor at or above 0.5 marks an acceptable screening assay is a widely used convention rather than a validated cut point, and panel reports frequently omit per-target Z-factors altogether. A clean row from a poorly performing assay is not evidence of selectivity; it is evidence of nothing.

Full curves, and what a Ki is

A concentration series converts a percent into a curve with a fitted midpoint and a slope. The slope matters. The relationship used to convert a 50%-inhibition concentration into an inhibition constant was originally derived for enzyme inhibition ^[7], and its radioligand form — dividing IC50 by one plus the ratio of labeled ligand concentration to that ligand's dissociation constant — carries the same assumptions: competitive displacement, a single site, equilibrium, and unit slope. Where those assumptions fail, corrected forms are required. One analysis treats the slope parameters of the concentration-response and inhibition curves as cooperativity indices and derives power equations that reduce to the standard relationship only when both indices equal unity ^[8].

The assumptions have arithmetic consequences. Suppose two laboratories measure the same compound with a true Ki of 10 nM. Laboratory A runs the radioligand at 1.0 nM against a Kd of 0.50 nM, so the correction factor is 1 + 1.0/0.50 = 3.0 and the observed IC50 is 30 nM. Laboratory B runs the same radioligand at 0.20 nM, so the factor is 1 + 0.20/0.50 = 1.4 and the observed IC50 is 14 nM. The reported IC50 values differ by 30/14 = 2.14-fold with no disagreement about the compound at all. IC50 values are only comparable across laboratories when the radioligand conditions are stated.

Off-target thresholds are chosen, not discovered

A hit threshold is an operational convention. The survey cited above found the threshold for calling an interaction significant to be one of the parameters that varies between organizations ^[2], and the panel-construction study used IC50 ≤ 1 µM as its hit definition while screening routinely at 10 µM ^[3]. Two groups can screen identical compounds against identical targets and publish different hit counts because they drew the line in different places.

Hit rates are also not uniform across a panel. In that 50-target panel, screening approximately 1,200 small molecules at 10 µM, adenosine A3 and 5-HT2B receptors showed the highest hit rates at around 30%, while half the targets had hit rates at or below 4% ^[3]. If those two high-rate targets behaved independently — which promiscuity almost certainly does not, so treat this as intuition rather than prediction — a compound drawn from that chemical space would clear both only 0.70 × 0.70 = 0.49 of the time. Which targets you include changes how impressive "clean" looks.

Those hit-rate figures come from small-molecule libraries ^[3]. Whether comparable hit rates apply to peptides is not established by that work and should not be assumed.

What the assay system contributes to the answer

A number from a cell line is a property of the compound and the system jointly. In one study using an inducible expression system in transfected Chinese hamster ovary cells, both the potency and the efficacy of agonists at the recombinant human mGlu1α receptor were influenced by receptor expression level, while competitive antagonist activities at that receptor were independent of expression level ^[9]. That was a single receptor in a single engineered system and it should not be generalized into a rule; it does establish that an apparent potency can move without the molecule changing.

Assay format decides what class of interaction is visible at all. A review of ion channel profiling states that for ion channels the best data come from functional assays, because binding assays cannot detect all interactions and do not provide information on whether an interaction is that of an agonist, antagonist, or allosteric modulator ^[10]. The scope of that statement is ion channels; the underlying point — that displacement of a labeled orthosteric ligand is silent about the direction of an effect — is why a binding panel and a functional panel are not substitutes for one another.

Artifacts in both directions

Apparent activity can arise from a compound's physical behavior rather than from receptor engagement. Some organic molecules form large colloid-like aggregates that sequester and thereby inhibit enzymes; the authors of one study reported that screening hits, discovery leads and even several drugs appear to act through this mechanism at micromolar concentrations, and described two rapid assays for detecting it, tested against 1,030 drug-like molecules ^[11]. That evidence concerns enzyme inhibition rather than receptor binding, so it does not transfer automatically to a displacement panel — but it establishes that a micromolar hit deserves a mechanism check before it is called an off-target interaction.

Losses run the other way too. Compound precipitation, adsorption to plasticware, and degradation over the course of an incubation all reduce the effective concentration and produce blanks that look like selectivity. Confirming the compound's identity and concentration in the assay buffer is standard bench practice; it is rarely reported alongside panel data.

Selectivity ratios and the denominator

A selectivity ratio is only interpretable when you know which value sits underneath. If the intended target has Ki = 2.0 nM and an off-target has Ki = 400 nM, the ratio is 400/2.0 = 200-fold. Invert the denominator and you get 2.0/400 = 0.005, which is the same information rendered meaningless. When the off-target produced no measurable displacement at the top concentration, the honest figure is a bound: with a 10 µM ceiling, the ratio is greater than 10,000/2.0 = 5,000-fold, and reporting it as an equality overstates what was measured.

Ratios also fail when the numerator and denominator come from different assay formats — a binding Ki over a functional IC50 is a number without a coherent unit of meaning, even when both are quoted in nanomolar.

Aggregating many targets into a single selectivity score is an active methodological area rather than a settled one. Two metrics proposed for kinase profiling, a window score and a ranking score, were reported by their authors to bring additional information relative to already published selectivity metrics when compared across several data sets ^[12]. A separate group argued that existing metrics quantify a compound's overall selectivity but fall short of quantifying how selective it is against a particular target, proposed a target-specific decomposition into absolute and relative potency, and reported the resulting score to be relatively robust against missing bioactivity values and against dataset size, with permutation-based empirical p-values for significance ^[13]. Both were developed and tested on kinase data; neither has been shown to transfer to receptor panels.

What a clean profile does and does not show

A clean panel shows that, in the specific systems used, at the specific concentration used, against the specific targets included, the compound did not cross the threshold the authors chose. That is a real and useful result. It does not show:

• Absence of activity at any target that was not on the panel.
• Absence of activity above the tested concentration, or below the assay's detection window.
• The direction of any interaction that was detected, if the format was binding rather than functional ^[10].
• Allosteric interactions that an orthosteric displacement assay is not configured to see ^[10].
• That the compound was actually present in the well at the nominal concentration.
• Anything about behavior in an intact organism, which no in vitro panel measures.

What remains convention, and what is unknown

The 10 µM screening concentration, the 50%-inhibition hit call, the Z-factor floor of 0.5, and the practice of running a reference compound on every plate are conventions that have hardened through use rather than positions established by a decisive experiment. The field's own reviews describe an absence of dedicated guidelines and only loose consensus on interpretation and follow-up ^{[1, 2]}. False-negative rates in screening are hard to verify at all, because in most assays only the compounds selected in the primary screen are followed up ^[5]; the same limitation applies to panel blanks, but it has not been quantified for receptor panels specifically. And the translation from any in vitro hit to consequences in a living system sits outside what these assays measure. The correct posture toward a clean panel is to read it as a bounded negative with stated coordinates — and to ask, first, which targets were left off the plate.