The number you report from a label-free binding experiment is set at the immobilisation step, before any analyte is injected. A surface that presents its ligand in three orientations, at a density that outruns delivery, referenced against a cell that does not match it, will still produce a clean-looking sensorgram and a global fit with small residuals.
The coupling reaction decides what you can measure
Amine coupling is the default chemistry on carboxymethylated dextran, and its mechanism explains most of what goes wrong there. It is a zero-length route: a carbodiimide such as EDC, acting in the presence of N-hydroxysuccinimide, converts carboxyl groups into succinimidyl esters, and the crosslink forms when lysine ε-amino groups displace the succinimidyl moiety, leaving no spacer between the partners [1]. Run under flow on a sensor chip, the sequence is: activate the carboxymethyl groups with an NHS/carbodiimide mixture, inject the ligand, then cap the surviving esters with ethanolamine [2].
Nothing in that sequence selects which amine reacts. A ligand carrying several accessible primary amines — the N-terminal α-amine among them — couples through whichever one first meets an activated ester, so the immobilised population is a mixture of orientations, some of which bury the binding site. The effect is quantifiable. Single-domain VHH antibodies were immobilised by amine coupling and their conserved lysines mutated to alanine, with the fraction of immobilised antibody still able to bind antigen measured by SPR; binding activity tended to be higher when K72, the lysine close to the antigen-binding site, was mutated, and adding a Lys-tag to the C-terminus also raised it [3]. Removing K72 was the more effective of the two interventions, and coupling through the N-terminus — itself close to the binding site — did not depress activity as much as coupling through K72 [3].
The cost is not only lost signal. Heterogeneity of the surface sites and mass transport limitation are the two most common sources of deviation from ideal pseudo-first-order binding kinetics in SPR, and both are intrinsic to the use of a biosensor surface at all [4]. A surface carrying several orientations of ligand is not a 1:1 system, however cleanly the sensorgram subtracts.
Preconcentration, and the pH scout that finds it
Amine coupling works at low ligand concentration for an electrostatic reason rather than a chemical one. Carboxymethyl dextran is polyanionic; a ligand passed over the surface in low-ionic-strength solution at a pH below its isoelectric point carries net positive charge, is drawn into the matrix, and accumulates there far above the injected concentration while active esters are still available, so coupling proceeds from within a concentrated layer [2]. Immobilisation of proteins by this route completes in under 30 minutes and reaches typical surface densities near 70 fmol mm-2 [2].
A pH scout is the experiment that locates that condition: the ligand is injected in a series of low-ionic-strength buffers spanning pH values below its expected pI, across an unactivated surface, and the buffer giving the largest and fastest accumulation is the one to couple in. Read it as a diagnostic rather than as a tuning step. If nothing accumulates on the unactivated surface, activating the chip wastes both the chip and the ligand.
The boundary of the method is sharply drawn. Electrostatic accumulation at slightly acidic pH and low ionic strength is used to load neutral and basic ligands efficiently, but it is not suitable for acidic proteins or other ligands with low isoelectric points, such as nucleic acids, because the charge density of the polycarboxylate is greatly reduced at acidic pH or because the electrostatic attraction cannot be achieved at all [5]. Reversing the surface charge with a precisely controlled amount of tertiary amine, while keeping the surface compatible with carbodiimide chemistry, allowed acidic proteins to be preconcentrated above their isoelectric points, and allowed DNA — which had previously proved impossible to preconcentrate electrostatically on a polyanionic chip — to be coupled in good yield [5]. A failed pH scout on an acidic ligand is a statement about the matrix, not about the ligand.
Thiol, aldehyde, and a comparison that cuts the other way
Three further chemistries were developed on the same carboxymethyl dextran support: thiol/disulfide exchange, aldehyde coupling and biotin–avidin coupling [6]. Thiol coupling routes attachment through a cysteine rather than a lysine, which in a synthetic peptide is a position you choose. It is also reversible: haptens attached by a surface-thiol method can be removed completely from the sensor surface again by a reducing step, which is how a single physical chip was used for successive immobilisations of different ligands and for building a calibration series [7].
The four-way comparison of those chemistries cuts against the reflex to reach for oriented coupling. When each method was optimised — varying activation time, protein concentration, ionic strength and degree of modification — the four gave rise to only minor differences in the specific activity of an immobilised monoclonal antibody, and the conclusion drawn was that with this flexible dextran matrix a site-directed immobilisation strategy for monoclonal antibodies has no advantage [6]. Optimised systems reached roughly 75% specific activity for the binding of β2-microglobulin to an immobilised antibody directed against it, and the reduced specific activities produced by varying the coupling conditions were attributed to deterioration of the antibody's active site [6].
That result and the VHH result are differently scoped rather than contradictory. One compares chemistries applied to intact monoclonal antibodies inside a flexible three-dimensional matrix [6]; the other compares coupling positions within a small single-domain protein whose binding site sits close to a reactive lysine [3]. What survives both is that orientation matters in proportion to how near the reactive residue is to the binding determinant, and that a flexible matrix can carry a large ligand whose binding site is distal.
Capture instead of coupling
Capture substitutes a high-affinity non-covalent bond, made at a defined point on the ligand, for the covalent one. Covalent immobilisation and streptavidin–biotin are both irreversible in nature: they can afford excellent baseline stability, but they impose limits on throughput for slowly dissociating analytes or unstable targets [8]. His-tag capture on Ni-NTA is reversible, and pays for it — reversible immobilisation typically precludes accurate quantification of slow dissociation kinetics, because of baseline drift [8].
The NTA surface is a mixed monolayer: an NTA-terminated alkanethiol that forms a tetravalent chelate with Ni(II), diluted with a tri(ethylene glycol)-terminated thiol chosen because it resists protein adsorption, with the hexahistidine tag binding to the two sites Ni(II) leaves vacant [9]. The immobilised His-tagged proteins were kinetically stable in buffered saline at pH 7.2 but could be desorbed by treatment with 200 mM imidazole [9]. In SPR studies of two model systems, His-tagged proteins adsorbed on the NTA monolayer retained a greater ability to participate in binding interactions with proteins in solution than protein immobilised in a thin dextran gel layer by covalent coupling [9].
The most direct head-to-head data come from single-domain antibodies compared across immobilisation routes by SPR in a multiplexed format. The construct providing the highest density of active molecules, by at least a factor of two, was a single-domain-antibody–streptavidin core tetramer fusion; next came a fusion with alkaline phosphatase, then a site-specifically biotinylated monomer [10]. The two poorest performers were also the two most common: direct covalent attachment, and randomly biotinylated antibody captured via NeutrAvidin [10]. Random biotinylation is not oriented capture. It reproduces the amine-coupling problem with an extra reagent.
Where a tag can be built into the construct, site-specific covalent capture gets both properties at once. Cyclophilin D expressed as a fusion with a mutant of the human DNA repair protein O6-alkylguanine-DNA-alkyltransferase, then contacted with a surface bearing an amine-coupled alkylguanine derivative, gave high densities of covalently linked protein with relative binding activities close to 100%; the constants recovered for cyclosporin A were KD 12 nM, kon 3 × 105 M-1 s-1 and koff 4 × 10-3 s-1, with the KD in agreement with a value determined in solution by fluorescence titration [11]. Agreement with a solution measurement is the strongest evidence available that a surface is not distorting the interaction.
The peptide case
For a peptide the argument sharpens, because there is nowhere distal to attach. A short sequence typically carries one or two primary amines — a lysine side chain and the N-terminal α-amine — so amine coupling has few options, and where the binding determinant sits within a few residues of either, a large fraction of the immobilised material is inactive. The flexible-matrix result that rescues an intact antibody [6] was obtained on a ligand far larger than a synthetic peptide, and should not be assumed to transfer. The practical response is to design the attachment point into the synthesis: a cysteine at the terminus away from the determinant for thiol coupling [7], or a biotin or bio-orthogonal handle placed there instead. Peptide epitopes immobilised on an SPR chip through a chemoselective copper-catalysed azide–alkyne cycloaddition demonstrated that fine tuning of the surface peptide density plays a crucial role in fully exploiting the potential of oriented and multiple peptide display, and by SPR imaging a tandem construct gave approximately twofold increased binding capacity relative to the linear and branched forms, attributed to a lower rate of dissociation [12].
Choosing a ligand density
Ligand density is a design parameter, not a yield to be maximised, because the constants you extract depend on it: the values of ka, kd and KD determined by label-free interaction analysis are strongly affected by the ligand density at the sensor surface [13]. Two published strategies confront that dependence. One applies a ligand density gradient across an SPR-imaging surface and extrapolates the analyte response to Rmax = 0, which avoids trial-and-error tuning of ligand density and minimises surface-dependent effects such as rebinding and strong biphasic behaviour [13]. The other globally fits sets of coupled surfaces spanning a range of ligand densities and analyte concentrations, then interpolates the result to a common reference level, reporting KD at an Rmax of 100 RU [14]. Both concede the same point: a measurement at one density is one point on a curve you have not mapped.
If a gradient is not available, choose the density from the constant you intend to report. A rate-constant fit wants the lowest density that still yields a signal you can fit. A concentration measurement wants the opposite: there, the mass transport limited region is used to determine the analyte concentration [13], and the transport-limited regime is the measurement principle rather than an artefact.
Mass transport limitation is what too much ligand buys you
When ligand density is high, analyte is captured at the surface faster than flow and diffusion replace it, and the observed rate reports delivery rather than chemistry. A simple two-compartment model of transport and binding was tested against a computer model that calculated laminar flow of analyte within the flow cell, its diffusion both perpendicular and parallel to the sensor surface, and the reversible reaction with immobilised reactant; analysing the simulated data showed that good estimates of the intrinsic reaction rate constants are recovered even when mass transport influences the binding reaction, significantly extending the range of association rate constants that can be accurately determined [15]. The condition is that the transport term is actually in the model.
Leave it out and the failure mode is quiet. Simulations of binding to receptor immobilised in a polymer matrix found that, under conditions that may apply to many published experimental studies, diffusion within the gel significantly decreased the overall transport of the soluble partner, and that extremely transport-limited reactions can in some cases be fitted with apparently ideal binding progress curves, although with apparent reaction rates much lower than the true rates [16]. The same work found that the ratio of the apparent rate constants can be semiquantitatively consistent with the true equilibrium constant, that apparently cooperative binding can result from high chemical on-rates at high receptor saturation, and that a thinner polymer matrix could significantly increase the range of detectable rate constants [16]. That is the kinetics-versus-affinity split in one sentence: a transport-compromised surface may still return approximately the right equilibrium constant while returning two wrong rate constants.
The standard diagnostic is flow rate. The observed binding rate changes with flow rate at high immobilised ligand levels, where transport from bulk to the surface becomes partially rate limiting, and a simple analytical expression for the mass transport coefficient derived for fully transport-limited conditions is applicable under partial limitation as well [17]. Run one analyte concentration at two or three flow rates over the same surface: if the initial rate tracks flow, the surface is reporting delivery. Lowering ligand density is the standard first response; a thinner polymer matrix could significantly increase the range of detectable rate constants [16]. Surface-site heterogeneity, the other main source of non-ideal kinetics, has to be addressed at the coupling step instead [4].
Regeneration scouting and what it costs the surface
Regeneration must remove analyte completely without removing or damaging ligand, and those two requirements conflict. A general regeneration identification and optimisation protocol was built from six multi-ingredient stock solutions representing the six most common chemical properties employed as regeneration agents, with cocktails of those stocks tested iteratively; across 13 antibody–antigen systems, the screening session identified cocktails that removed more than 90% of the bound analyte in a 30 s pulse for 9 of them, and for 5 systems identified cocktails that regenerated the surface completely and were more gentle than the conditions previously used [18]. For every model system tested, at least one cocktail was significantly better than all the diluted stock solutions, which is the argument for scouting combinations rather than titrating one reagent upward [18].
Surface degradation is answered empirically, by repeating the cycle and asking whether the ligand baseline and the analyte response are still what they were. A cocktail that clears analyte while shaving a few per cent of capacity per cycle looks acceptable across three injections and invalidates a ten-point concentration series. For captured ligands the calculus differs: regeneration can be designed to strip the ligand deliberately and rebuild the surface — as when imidazole desorbs a His-tagged protein from an NTA monolayer [9] — trading the baseline stability of an irreversible attachment for a fresh ligand layer every cycle [8].
Reference surfaces
Every sensorgram contains signal that is not binding. Use of a reference surface made it possible to identify conditions where a response related to changes in matrix conformation could be ignored, and improved data quality by separating signals related to binding events from signals due to differences in refractive index between sample and running buffer [19]. In the same work, binding curves obtained on surfaces with and without a dextran matrix were virtually superimposable, so no binding artefacts associated with the dextran matrix were observed [19].
The reference cell should therefore differ from the active cell in one respect only, the ligand, and should have been carried through the same activation and ethanolamine deactivation steps [2], since it is that chemistry rather than the ligand that sets the surface's residual charge. An underivatised cell references bulk refractive index but not the surface you actually built. In a capture format the reference should carry the capture reagent without the captured ligand, so that the capture layer subtracts out along with the buffer.
What none of this buys is mechanism. Referencing does not convert a heterogeneous, overloaded surface into a 1:1 interaction [4], and a good fit is not a proof: sets of binding curves inconsistent with a one-to-one reaction could be fitted well by several other schemes including parallel, competitive and two-state reactions, so data analysis alone was not sufficient to discriminate between them, and several schemes were ruled out only by further experiments — varying the duration of the injection, and reanalysing analyte recovered from the surface [19]. Most of that discrimination is won or lost at the immobilisation. A surface built on a scouted pH, an attachment point chosen against the binding determinant, a density set by the constant you intend to report, and a reference cell that differs in one variable licenses you to interpret the fit. Without them, the fit describes the surface.