A number from a peptide assay carries an implicit claim: that the response measured was produced by the molecule added, acting the way the experimenter thinks it acts. Every control on the plate exists to close off one specific alternative explanation for that number. Controls are not a quota to be filled; each is a competing hypothesis you are attempting to kill, and a plate can be crowded with controls and still leave the decisive artifact untested.

This guide walks the standard control set used in peptide assays and states, for each, the artifact it excludes and the artifact it leaves untouched.

A control is an answer to a question about causation

Before choosing controls, write down the ways the well could have produced the observed signal without the peptide engaging its intended target. The usual list: the solvent did it; a residue from synthesis did it; a contaminant in the preparation did it; the peptide did something physical rather than specific, such as partitioning into membranes, adsorbing to plastic, aggregating, or quenching a fluorophore; the cells would have done it anyway; the detection chemistry did it; or the plate is drifting. Each entry maps to a different control, and a control matching no entry is decoration.

Two consequences follow, both conventions of experimental design rather than findings from any single study. A control is informative only when run in the same plate, at the same passage, on the same day, with the same reagent lots as the samples it explains; a historical control excludes almost nothing. And the absence of a control is not neutral: if nothing on the plate addresses membrane damage, membrane damage remains a live explanation no matter how small the p-value is.

Vehicle: rules out the solvent and the handling

The vehicle control is the complete assay minus the peptide: identical buffer, solvent, final solvent percentage, pipetting and incubation. It excludes one class of explanation only, that the response came from the carrier or the act of dosing rather than from the test article.

The common failure is treating the solvent as inert. In three-dimensional cardiac and hepatic microtissues exposed to medium with or without 0.1% dimethyl sulfoxide, transcriptome analysis detected more than 2,000 differentially expressed genes in both tissue types, affecting similar biological processes, which the authors read as indicating consistent cross-organ actions of the solvent; they also reported large-scale deregulation of cardiac microRNAs and concluded that DMSO is not inert [1]. Whether comparable effects appear in another cell system, at another percentage, over another exposure window is an empirical question for that system; the study does not settle it. The methodological consequence is narrower and firm: the vehicle well must carry the same solvent percentage as the highest-concentration peptide well, not the average across the series.

What the vehicle control cannot rule out is anything that travels with the peptide. Residual counterions, a co-purifying impurity, or endotoxin are absent from the vehicle well by construction, so the vehicle can look perfectly clean while the entire effect is caused by something it never contained.

Scrambled sequence: rules out sequence-independent action

A scrambled control is the same amino acids in a permuted order, ideally from the same synthesis route, at the same purity, with the same counterion. It shares molecular weight, net charge at a given pH, and gross hydrophobic content with the test peptide. What it excludes is the possibility that the response is a function of bulk composition rather than of sequence. If the scramble reproduces the effect, the effect belongs to charge, amphipathicity, or hydrophobic surface area, not to a binding sequence.

Two failure modes recur, both matters of design practice rather than of published measurement. A permutation can accidentally reconstitute a short recognition motif, so a scramble should be checked residue by residue against the motifs the parent sequence is thought to use rather than generated once at random and trusted. And for peptides whose behavior is dominated by cationic charge or amphipathicity, scrambling changes very little that matters, so the control is close to guaranteed to be active and reports only that composition matters. There, a single-residue alanine substitution at a position believed to be essential is the sharper instrument, because it changes almost nothing physical while removing a specific contact.

Inactive analog: rules out the chemotype rather than the sequence

An inactive analog is a close structural relative that has lost target engagement while retaining the rest of the molecule's physical behavior. It controls for effects that depend on the compound being that kind of compound: nonspecific membrane partitioning, protein adsorption, aggregation, interference with the detection chemistry. A scramble does this less reliably, because permutation can alter folding and amphipathic character, which an inactive analog is designed to preserve.

The failures of the equivalent practice in small-molecule pharmacology are instructive. Arrowsmith and colleagues, writing as a large multi-institution group, noted that chemical probes of poor quality or used incorrectly generate misleading results, and proposed a community-driven wiki resource to convey current best practice [2]. Blagg and Workman described evidence of widespread continuing misuse and promulgation of poor-quality and insufficiently selective probes, and argued that biologists should be more discriminating in the reagents they choose [3]. Both are commentary rather than measurement, and should be read that way.

What an inactive analog does not rule out is whether it is genuinely inactive at the concentration used. An analog with 100-fold lower affinity behaves as a negative control at 10 nM and as a weak agonist at 10 µM. Its inactivity has to be demonstrated in the same assay, in the same run, over the same concentration range, not inherited from a binding constant measured elsewhere.

Receptor-null cells: rules out everything the cell can do without the receptor

Genetic controls are the strongest of the set because they remove the proposed mediator instead of modifying the ligand. An international working group convened on antibody validation proposed five conceptual pillars to be used in an application-specific manner, the first being genetic strategies: measuring the relevant signal in control cells or tissues in which the target gene or epitope has been knocked out or knocked down. Their stated logic is that any signal observed after substantial reduction of the target protein indicates crossreactivity, and they suggested at least one pillar as a minimum criterion for claiming that a reagent has been adequately validated for a specific application [4]. That group wrote about immunoreagents, not about ligands, but the inference structure is the same one a receptor-null cell line supplies for a peptide response.

The premise that a parental or null line is a blank slate deserves checking. A microarray survey of HEK293, AtT20, BV2 and N18 cells profiled a comprehensive set of non-chemosensory G protein-coupled receptors and more than a hundred related signaling gene products; the authors observed that such lines are often treated as blank slates yet endogenously express GPCRs and related signaling proteins, so that the outcome of a given GPCR study can be profoundly influenced by this largely unknown complement [5]. That work measured mRNA rather than receptor protein or function, so it constrains expectations rather than settling what any particular line will do.

Two practical limits. A knockout clone differs from its parent in more than the deleted gene, having passed through single-cell cloning and selection, which is why several independent clones, or rescue by re-expression, carry more weight than one. And knockout cannot distinguish direct engagement from a response that merely requires the receptor somewhere upstream.

Positive reference agonist: rules out a dead plate and fixes the denominator

A reference agonist of established behavior in the assay answers a question no negative control can: would this plate have detected a real response today? Without it, a flat concentration-response curve is uninterpretable, because an inactive peptide, a degraded peptide, an exhausted detection reagent, an unresponsive passage and a pipetting error all produce the same flat line.

The reference also supplies the assay-quality statistic. Zhang, Chung and Oldenburg defined a screening window coefficient, the Z-factor, that is reflective of both the assay signal dynamic range and the data variation associated with the signal measurements, and is dimensionless [6]. Two related coefficients circulate under that heading, and which one is in hand depends on which wells entered the calculation. The coefficient computed from the two control groups alone is conventionally written Z-prime: it equals one minus three times the sum of the positive-control and negative-control standard deviations, divided by the absolute difference of the two control means. The plain Z-factor is the corresponding statistic in which the test-sample wells stand in place of the positive control. A number obtained from a reference agonist and a vehicle is therefore a Z-prime, and calling it a Z-factor on the test article names a quantity that was not the one computed.

Reagent lots are a concrete failure mode the positive control is built to catch. In an evaluation of a rapid particle gel immunoassay across 1,376 plasma samples, the incidence of false-negative results was strikingly variable depending on the polymer lot employed, ranging from none up to 82% (95% CI 66 to 98%); the authors concluded that laboratories should implement stringent internal quality controls, noting that faulty lots can be recognized by titrating commercial positive controls and stored known-positive samples [7]. Titration, not a single-point positive, carried the information there.

Controls for the preparation, not for the molecule

Some artifacts belong to the vial rather than the sequence, and no ligand-side control reaches them. Endotoxin is the standing example. Reviewing the reported cytokine effects of heat shock proteins, Tsan and Gao wrote that recent evidence suggests those effects may be due to contaminating lipopolysaccharide and LPS-associated molecules, and listed among the reasons the contaminants went unrecognized a failure to use highly purified, low-LPS preparations, a failure to recognize the heat sensitivity of LPS, and a failure to consider contaminants other than LPS; they argued it is essential to determine conclusively whether such effects come from the protein or the contaminant [8]. That hedge is the authors' own and should not be hardened.

The sensitivity of the contamination test is itself part of the control. Four cement formulations found free from endotoxin by standard gel-clot testing nonetheless stimulated different levels of cytokine release from macrophages; a more sensitive kinetic endotoxin assay showed that release correlated with minor contamination, and in a repeat using particles with low or undetectable endotoxin by that kinetic assay, the differences between formulations were not seen [9]. That was a different material and format, but the transferable point is that a negative on an insufficiently sensitive contamination test is not a control at all.

The corresponding control is preparation-matched rather than sequence-matched: a mock preparation carried through the same cleavage, workup, lyophilization and reconstitution without the target sequence. Because that review lists the heat sensitivity of LPS among the things investigators missed, treating a heat-denaturation step as a clean discriminator between peptide and contaminant is not a reading the review supports.

Controls that catch the readout rather than the biology

Some controls exist because the detection method lies. In a reevaluation of cellular uptake of cationic cell-penetrating peptides, cell fixation even under mild conditions was shown to cause artifactual uptake of Tat 48-60 and nonaarginine, and the authors reported that flow cytometry cannot validly be used to evaluate uptake unless a step of trypsin digestion of membrane-adsorbed peptide is included [10]. The controls that matter there are a live-cell comparison and a trypsin wash; neither is a control on the peptide.

Chemistry-side interference has an analogous literature in screening. A sulfhydryl-scavenging high-throughput screen targeting the histone acetyltransferase Rtt109 returned hits that were nearly all artifacts or promiscuous bioactive compounds, and protein mass spectrometry with ALARM NMR confirmed that several chemotypes reacted covalently with cysteines on multiple proteins [11]. Separately, colloidal aggregation has been proposed as a single mechanism accounting for many promiscuous inhibitors, and two rapid assays for detecting such aggregates were reported and tested against 1,030 drug-like molecules [12]; promiscuous inhibitors can also be distinguished from classical one-to-one inhibitors by judicious use of detergent [13]. Those studies concern small molecules, so a detergent counter-screen on a peptide is an extension by analogy, not a result carried over.

A worked plate, with the arithmetic shown

The values below are illustrative and invented for the calculation; they are not measurements. Suppose a luminescent reporter plate gives, in relative light units: reference agonist wells, mean 42,000, standard deviation 2,100; vehicle wells, mean 3,000, standard deviation 450; the test peptide at its top concentration, 28,500; and the same peptide on receptor-null cells, 7,200, with vehicle on those null cells also 3,000.

  • Assay quality. Z-prime equals 1 minus 3 times (2,100 plus 450) divided by the absolute difference between 42,000 and 3,000. That is 1 minus 7,650 divided by 39,000, which is 1 minus 0.196, or 0.80. Both inputs are control wells, so 0.80 is a Z-prime and should be reported as one rather than as a Z-factor on the test peptide. The plate separates its controls.
  • Response as a fraction of reference. The denominator is the background-subtracted reference response, 42,000 minus 3,000, which is 39,000. The numerator is the background-subtracted test response, 28,500 minus 3,000, which is 25,500. The ratio is 25,500 divided by 39,000, or 65.4%.
  • The denominator error. Dividing raw by raw gives 28,500 divided by 42,000, or 67.9%. The difference is 2.5 percentage points, which grows as the vehicle background grows relative to the window.
  • What the null cells remove. On receptor-null cells the peptide gives 7,200 minus 3,000, or 4,200 above vehicle. That receptor-independent component is 4,200 divided by 25,500, the background-subtracted test response, or 16.5%. The receptor-dependent remainder is 25,500 minus 4,200, or 21,300, which is 21,300 divided by 39,000, or 54.6% of reference.

The headline moves from 65.4% to 54.6% of reference once the null-cell control is applied, and only the second figure is a statement about the receptor. Which value sits in the denominator is not presentational; it changes what the percentage means.

What controls cannot do, and where practice is convention

No control set makes a single plate conclusive. Controls narrow the space of explanations; they do not establish mechanism, and each is silent outside the artifact it was designed for. A vehicle control says nothing about contaminants, a scramble nothing about the readout, receptor-null cells nothing about whether the concentration used is physically reasonable. A positive control confirms the assay was alive, not that the test article's number is right.

Several widespread habits are conventions with no settled evidence base behind the specific threshold: numeric cutoffs for acceptable assay quality, the number of replicate wells treated as sufficient, and the acceptance of a single knockout clone are inherited defaults, and should be stated as such rather than presented as validated criteria. The full impurity profile of a given peptide preparation is also generally unknown, so the safest reading of a well-controlled experiment is that the listed alternatives were excluded, not that all alternatives were.