Biolayer interferometry does not measure binding either. It measures the optical thickness of a layer on the end of a glass fibre. A bait is immobilised on a matrix at the tip of a fibre-optic sensor, and binding between that ligand and a molecule in the analyte solution produces a change in optical thickness at the tip and a wavelength shift proportional to binding [1]. Everything downstream — the two rate constants and the affinity they define — is inferred from that trace under a model you chose, exactly as in surface plasmon resonance. What differs between the two techniques is not the kinetics. It is the geometry, and the geometry decides which artefacts you inherit.

The readout, and what it is not

Visible light travels down the fibre and reflects from two places: an internal optical layer and the outer face of the biolayer containing the proteins of interest. The two reflections interfere, and BLI measures the change in that interference pattern; the sensor is a disposable probe immersed in a 96-well or 384-well plate [2]. In a four-platform comparison the authors noted that, unlike the surface plasmon resonance instruments, the BLI system does not rely on continuous flow fluidics — the sensor tips collect readings while immersed in analyte solutions during orbital agitation [3].

The first consequence is a unit problem. Response is a wavelength shift in nanometres, and loading is quoted the same way: in an assay-design study, a monoclonal antibody capture density of less than 0.6 nm was found to be optimal to measure a wide range of binding affinities, and 0.1 nm surfaces were used to reduce mass transport limitation [4]. A nanometre of shift is not a response unit. The RU-to-surface-mass scaling that makes the SPR Rmax formula numerically usable has no equivalent number here, so carry the physics across and leave the arithmetic behind.

The physics does carry. In a platform-independent analysis of 1:1 binding, the measured signal is taken as proportional to the amount of ligand–target complex formed, with a maximal proportional response corresponding to the signal at full occupancy of the immobilised partner's binding sites [5]. A small analyte therefore still produces a small ceiling — a bound 50 kDa protein thickens the layer more than a bound 2 kDa peptide does, which is geometry rather than a finding, and is flagged as such.

The second consequence is the one most often overstated. Using BLI to follow inhibitory interactions within an ATP synthase complex, one group reported that the signal is not overly sensitive to solution composition, and used that property to monitor allosteric effects of catalytic-site ligands [6]. Read the hedge as written: not overly sensitive is not insensitive. A buffer mismatch between the loading well and the association well still produces a step at the transition, and referencing remains obligatory.

Dip-and-read: what the missing flow cell buys and costs

BLI is described as a platform enabling label-free monitoring of biomolecular interactions without the use of flow cells [7], using disposable fibre-optic biosensors that address samples from an open shaking microplate without any microfluidics [8]. Tips run in parallel, there is nothing to clog, and sensors are consumable — so regeneration becomes an efficiency choice rather than a requirement, as in a teaching protocol describing optimised timing and regeneration methods specifically to increase throughput and reduce cost [9].

It costs three things. First, transport to the surface is governed by orbital agitation and well geometry rather than by a volumetric flow rate you can set and compute. The SPR lever — raise the flow rate and gain delivery — has no direct analogue, which leaves ligand density as the transport control you actually have.

Second, each phase happens in a different well. Baseline, loading, second baseline, association and dissociation are separate dips, so every phase boundary is a physical transfer of the probe rather than a valve switch, and any dissociation occurring during that transfer is unobserved. The step sequence is standard instrument workflow and is stated without citation.

Third, every sensor is a separate piece of hardware. In SPR the reference is another channel on the same chip; in BLI it is another tip, with its own loading and its own surface. Frenzel and Willbold applied kinetic titration series to BLI in part because doing so circumvents problems in data evaluation caused by unavoidable sensor differences [10].

Loading, and which partner belongs on the tip

The workhorse is a streptavidin tip and a biotinylated ligand. A protocol using streptavidin-based BLI reports quantitative equilibrium binding affinities together with rates of association and dissociation measured in minutes using nanomole quantities of sample [1]; the same chemistry carries a teaching experiment in which biotinylated protein A is loaded onto a commercial sensor and kon, koff and KD are extracted from an IgG dilution series [9]. That streptavidin–biotin capture is effectively irreversible over an experiment is textbook, and uncited here.

Peptides get an advantage here that recombinant proteins do not. Traditional methods of protein immobilisation often result in activity loss due to random coupling; a SpyCatcher variant achieving over 99% efficient site-specific immobilisation was applied to BLI sensors and yielded reproducible kinetic data with lower variability than conventional methods [11]. That work used human growth hormone and the Z domain of protein A, so the size of the improvement should not be transferred elsewhere; the diagnosis does transfer. A synthetic peptide sidesteps it, because the biotin is installed at a chosen terminus during synthesis rather than wherever a lysine happens to react — routine peptide chemistry, stated without citation.

Direction follows from the signal argument. Because the response tracks the amount of complex and saturates at full occupancy [5], and because a heavier bound partner thickens the layer more, the peptide should generally be the immobilised ligand and the protein the analyte. That is the configuration Abdiche and colleagues used: a monoclonal antibody against a panel of peptides, compared directly with Biacore's 3000 and Bio-Rad's ProteOn XPR36 [8]. It is also enough to discriminate among candidate binders of a short sequence — BLI showed an α1D-adrenergic receptor C-terminal PDZ ligand binding the PDZ protein scribble with more than eightfold higher affinity than syntrophin, CASK and DLG1 [12], a ranking specific to PDZ recognition of a defined C-terminal motif.

The tether is the price. If the binding determinants of your peptide include the terminus you biotinylated, no assay-internal test will tell you — the check is external, either a solution method or the same measurement with the opposite terminus tethered.

Density is the other loading decision, and the one with published guidance. Capture densities below 0.6 nm were optimal across a wide affinity range, and 0.1 nm surfaces significantly reduced mass transport limitation and enabled characterisation of picomolar-affinity antibodies [4]. Those numbers came from antibody capture on one instrument generation; take them as an order of magnitude to bracket, not a specification for a peptide surface.

The traces, and what a fit is worth

The rate constants are the same quantities as in SPR: ka is second order, kd is first order and concentration-independent, and KD = kd ÷ ka. Only two are independent. These definitions are standard and are not cited.

Because kd is first order, the observable dissociation window is fixed by arithmetic. A complex with kd = 1 × 10−4 per second has a half-life of 0.693 ÷ 0.0001 = 6,931 s, about 1.9 hours, and any practical dissociation step samples only its first sliver. A complex with kd = 0.5 per second has a half-life of 1.39 s and is 95% dissociated after ln(20) ÷ 0.5 = 6.0 s; at kd = 2 per second the half-life is 0.35 s and 95% is gone in 1.5 s. Those are suppositions worked through a first-order rate law, not measurements. Set them against a format in which the probe must be lifted from one well into another between association and dissociation: for the fast case, much of the decay you need to fit is gone before the tip is reading buffer.

A recent deterministic analysis of error propagation reported a triphasic dependence of the relative error in KD on the ratio of total analyte concentration to KD, including a low-ratio regime in which KD is intrinsically robust to systematic error, and showed that accurate constants can be obtained far below the true KD when the maximal response is established independently — subject to the stated limitation that kon and koff be resolvable from the measured traces [5]. That framework assumes reversible 1:1 binding under pseudo-first-order conditions: it constrains what your numbers mean once the model is granted, and says nothing about whether your tip presents one class of sites.

Mass transport and rebinding

Both distortions are named explicitly in the BLI literature: the utility of the biosensors for accurate and reproducible measurement of monoclonal antibody binding rate constants was described as limited by challenges such as analyte rebinding and mass transport limitation [4]. Rebinding is the sharper problem in a dipped-tip format, because no flow sweeps released analyte away from a densely loaded surface. Abdiche and colleagues addressed it directly: a sink method was used to prevent analyte from rebinding the ligand-coated tips, and this enabled extraction of accurate kinetic rate constants, as judged by their close agreement with those determined by SPR [8].

Leaving rebinding in place produces a directional error. Suppose half the analyte released from a site is recaptured before it escapes the surface layer: the apparent kd is then roughly half the true value, the apparent half-life doubles, and the apparent KD tightens by the same factor. That shows the sign, not a magnitude. Rebinding never makes a complex look weaker or shorter-lived than it is, so a suspiciously slow off-rate from a heavily loaded tip is the result to distrust.

The general theory came from evanescent wave biosensors, where conventional analyses were shown to yield arbitrary apparent reaction rate constants lying between the mass transport rate constant and the true intrinsic chemical binding rate constants, depending on the ligand concentrations used; the two-compartment approximation introduced to fix this was described by its authors as independent of the details of the transport process [13]. That last clause is why the warning survives the change of geometry while the numbers do not: no transport rate constant fitted in a rectangular flow cell applies to a dipped tip in a shaken well.

The diagnostic needs no theory: run the same analyte series over tips loaded to two or three levels, and if the fitted rate constants move with loading level, they are reporting the surface, not the complex.

Nonspecific binding and reference sensors

Quantitative analysis by BLI requires effective suppression of nonspecific binding between analytes and biosensors, and the study of weak interactions — KD above 1 μM — requires high concentrations of analyte, which substantially increases that nonspecific binding; only a few blockers are compatible with biomolecules, and saccharides — sucrose in particular — were found to be potent ones [14]. That was demonstrated on one viral protein and kinase pair, so sucrose is a candidate to test, not a fix to assume. It is also the conflict that defines weak peptide work: the series that brackets a micromolar KD is the series that maximises adsorption to the sensor matrix.

Referencing is the other lever, and not the same operation as in SPR. A reference channel there carries the same chemistry on the same chip under the same flow; a reference tip is a different tip. The usual construction — one loaded sensor in buffer alone to capture drift, one unloaded or blocked sensor in the identical analyte series to capture matrix binding — is standard practice, stated without citation. The failure mode deserves the emphasis: because sensor differences are unavoidable [10], subtraction works only to the extent that the two tips were equivalent, and nothing in the trace tells you whether they were.

Small analytes, and where the floor actually sits

The published position has moved, and both ends are worth knowing. In 2008 the instrument was reported as not sensitive enough to detect the binding of small molecules directly, though their affinities could be accessed indirectly via solution competition experiments [8]. By 2011, BLI had been validated for small molecule detection and fragment screening with model systems and well-characterised targets, where affinity constants and binding profiles were generally similar to those obtained with surface plasmon resonance, and screens against BCL-2, JNK1 and eIF4E were run with a 6,500-compound fragment library [2]. Each holds for the instrument generation it describes; neither licenses a general claim about your compound.

Detection is also not kinetics. A BLI biosensor assessed with a low-molecular-weight model analyte gave replicate repeatability of 0.1 to 6.5% CV across six simultaneous calibration curves [15] — precision in a quantitation assay, not a determination of ka and kd.

A 1–5 kDa peptide sits between the regimes: far larger than a fragment, far smaller than the antibodies where BLI has shown SPR concordance. Kinetic analysis of 150 antibodies against 10 antigens of 21 to 105 kDa gave R2 above 0.90 against two SPR platforms [4], and ten high-affinity monoclonal antibodies against PCSK9 were characterised on four biosensors using the same assay format and reagents [3]. Those benchmarks used large analytes fitting a simple 1:1 model; they do not extend to a small, fast peptide by continuity.

Why fast-off peptide interactions are the hard case

Four penalties land on the same experiment, and each fix worsens another.

  • A short-lived complex leaves little decay to fit, and the well-to-well transfer consumes part of what there is.
  • Weak affinity forces high analyte concentrations, which substantially increase nonspecific binding [14].
  • If the peptide is the analyte rather than the ligand, the response ceiling is set by the mass it adds per binding event, and that is small.
  • Loading densely to recover signal promotes rebinding, which biases the fitted off-rate slow [4, 8].

The compounding is the trap. Lowering the loading level suppresses rebinding and mass transport limitation [4] but lowers a signal that was already marginal; putting the peptide on the tip restores signal, but only if the tether leaves it competent. No configuration removes all four, which is why a fast-off peptide interaction measured on one surface at one loading level should be reported as a bound rather than a value.

Three things remain available. Kinetic titration — sequential analyte injections over one sensor without waiting for complete dissociation — has been shown to be straightforward in BLI, saving resources and increasing throughput while avoiding the sensor-difference problem [10]; it was demonstrated on two high-affinity pairs, so its behaviour on a fast-dissociating system is untested there. A sink in the dissociation well suppresses rebinding [8]. And when kd cannot be resolved at all, equilibrium analysis of plateau response against concentration keeps the affinity and abandons the residence time — a convention, stated without citation, that should be declared rather than implied.

What transfers from the SPR guide, and what does not

  • Transfers: the rate-constant definitions; bracketing the fitted KD with the concentration series; the principle that a good 1:1 fit is weak evidence for a 1:1 surface; and the warning that transport-limited curves yield apparent constants lying between transport and chemistry [13].
  • Does not transfer: response units and the RU-based Rmax arithmetic; flow rate as a transport lever, since there is no flow cell [7, 8]; the on-chip reference channel, replaced by a separate tip [10]; regeneration as a necessity, since sensors are disposable [8, 9]; and the assumption that referencing is the only defence against bulk effects, which BLI reduces without removing [6].

What to report

  • the sensor chemistry, the loading level in nm, and which partner was immobilised, with the reason
  • plate temperature and agitation speed, which set transport here in place of flow rate
  • the reference design, the loading levels tested, and whether fitted constants moved between them
  • the analyte series as multiples of the fitted KD, whether kd was resolved or the analysis was equilibrium-only, and whether rebinding was suppressed

A KD reported without the loading level is a number without a surface, and in this format the surface is most of the experiment.