Microscale Thermophoresis and the Limits of an MST Binding Constant

Focus an infrared laser onto a spot inside a glass capillary holding a small volume of buffer, and the fluorescent molecules in that spot begin to leave it. That drift is thermophoresis, and Duhr and Braun's measurements on DNA and polystyrene beads across a large range of particle sizes, salt concentrations and temperatures supported a unifying theory based on solvation entropy, in which the Soret coefficient is given by the negative solvation entropy divided by kT — a theory that predicted the thermodiffusion of both species without any free parameters ^[1]. The number an MST instrument hands you is therefore not a property of your molecule alone. It is a property of your molecule's interface with the solvent, which puts your salt, your detergent, your label and your matrix inside the measurement alongside the binding event you meant to study.

What follows is about separating those contributions: which parts of an MST titration describe the interaction, which describe the assay, and what an MST dissociation constant is and is not entitled to claim next to one from calorimetry or a biosensor surface.

What the gradient is sensitive to

The solvation-entropy account assumes a local thermodynamic equilibrium of the solvent molecules around the molecule, an assumption the authors state is fulfilled for moderate temperature gradients below a fluctuation criterion ^[1]. Inside that regime the direction of drift is not fixed. For both DNA and polystyrene beads, thermophoretic motion changes sign at lower temperatures; the authors attribute this thermophilicity toward lower temperatures to an increasing positive entropy of hydration, whereas the generally dominating thermophobicity is explained by the negative entropy of ionic shielding ^[1]. A trace that moves one way at one instrument temperature and the other way at a lower temperature is not automatically a fault to be tuned out.

For binding work the useful property is breadth of sensitivity rather than specificity. Thermophoresis is highly sensitive to all types of binding-induced changes of molecular properties, be it in size, charge, hydration shell or conformation ^[2]. That breadth is what lets a small ligand register against a large target: despite the high molecular weight ratio, the interaction of small molecules and peptides with proteins is readily accessible by MST ^[2]. It is also why every co-solvent, aggregate and second fluorescent species enters the signal through exactly the same door as the ligand does.

Reading the trace, not just the ratio

The compact description of the experiment is this: a capillary tube filled with a solution containing a dye-labelled solute is illuminated with an infrared laser, rapidly creating a temperature gradient; molecules migrate along that gradient, causing changes in the observed fluorescence; and because the net migration of the labelled molecules depends on their liganded state, a binding curve as a function of ligand concentration can be constructed from MST data and analysed to determine K_D ^[3]. What that summary compresses is that the fluorescence-versus-time record has internal structure, and the structure is not all thermophoretic.

Before the laser fires, the trace reports only how much labelled material sits in the illuminated volume. When the laser turns on, the first thing to change is the dye, not its position. Temperature-related intensity change describes the change in fluorescence intensity of a fluorescently labelled molecule upon a change in temperature; that intensity variation is dependent on the physicochemical environment in the vicinity of the dye and is strongly affected by binding events, which makes detection of binding events independent of mass ^[4]. Directed migration develops after that. When the laser switches off, the sample re-equilibrates, and that phase carries information too — Stein and colleagues derived kinetic relaxation fingerprints from the fluorescence changes during thermodynamic re-equilibration of the sample after local heating ^[5].

Handle the name of the technique carefully, because it is doing more than one job. Microscale thermophoresis and the closely related temperature-related intensity change are described as synonyms for the same measurement technique ^[6], and several assumptions about it — including the extent of thermal deviations generated by the infrared laser and the thermophoretic foundation of the measured signal — have been revised during the last decade ^[7]. The bench consequence is that the time window you integrate is part of your method, not a display preference. A constant taken from an early window and one taken from a late window are under no obligation to agree, and a reader who does not know which you used cannot repeat your experiment.

Labelled and label-free modes

In standard MST one binding partner is fluorescently labelled, but the measurement can also be performed label-free by exploiting intrinsic protein UV fluorescence ^[2]. Two instrument classes follow. One requires labelling with a fluorophore in the blue or red spectrum; the second relies on the intrinsic fluorescence of aromatic amino acids in the UV range; both detect the movement of molecules in response to localised induction of heat from an infrared laser ^[8].

Each mode buys something and pays elsewhere. Label-free MST can use untagged native proteins; however, many analytes fluoresce in the UV range, which can interfere with determination of accurate K_D values ^[8]. If your titrant emits anywhere near tryptophan, label-free is the wrong mode, and the interference is not something a fitting model can be relied on to remove. Labelled MST allows a greater diversity of measurable pairwise interactions by using probes with measurable absorbances in the visible range as opposed to the UV, limiting the potential for interfering signals from analytes ^[8].

Covalent labelling carries its own cost, since amine-reactive chemistry does not choose its attachment site. Bartoschik and colleagues addressed that with a near-native, site-specific in situ labelling strategy exploiting the high-affinity interaction of tris-NTA with oligohistidine tags, which they report has distinct advantages over routine covalent labelling techniques; among the conjugates tested, RED-tris-NTA was identified as optimal on the basis of high affinity towards oligohistidine tags, high fluorescence signal and an optimal signal-to-noise ratio in MST binding experiments, and its emission in the red region of the spectrum also enabled reliable measurements in complex biological matrices such as cell lysates ^[9].

Whichever route you take, labelling is a chemical modification of the target and needs a functional control rather than an assumption. Nasreddine and colleagues labelled hyaluronidase with an ATTO-647 NHS ester and reported efficient labelling in a phosphate-based buffer without loss of catalytic activity ^[10]. That is the shape of the control: an orthogonal activity readout run on the labelled material itself.

Buffer, detergent, and the capillary wall

MST assays are highly adaptable to the requirements of different biomolecules, including membrane proteins that must be stabilised in solution, and the type of buffer and additives can be chosen freely ^[2]. Free choice is not the same as free of consequence.

Surfactant is rarely optional. Working with lipases from crude porcine pancreatic extracts, Al Hamoui Dit Banni and colleagues composed an MST buffer of 20 mM NaH₂PO₄ plus 77 mM NaCl at pH 6.6 with 0.05% Triton X added, which they report efficiently prevented protein aggregation and adsorption even when using only standard, uncoated MST capillaries ^[11]. Read that in reverse and it names the two failure modes waiting for any sample loaded into a glass capillary: aggregation in the tube and adsorption onto the wall. Coated capillaries exist as an alternative consumable for samples where an uncoated surface does not suffice.

Additives are not spectators either. Testing Tween-20, glycerol and trehalose directly against the MST signal of a labelled enzyme and, in parallel, against catalytic activity measured by capillary electrophoresis, Nasreddine and colleagues found that Tween-20 showed no impact on MST signals nor on enzymatic performance when used between 0.005 and 0.05% (v/v); that glycerol also did not interfere with MST measurements but significantly reduced catalytic activity at concentrations above 2% (v/v); and that trehalose affected the labelled enzyme's fluorescence in a concentration-dependent manner and enhanced catalytic activity even at 0.02% (v/v) ^[10]. Their own framing is the one to design against: such additives can affect the enzymatic activity and the enzyme's affinity for active compounds, and must be considered during assay design since a slight shift in enzyme behaviour may compromise the reliability of the results ^[10]. An excipient carried in with a ligand stock, at a concentration that climbs across the dilution series, produces a concentration-dependent fluorescence change that owes nothing to binding.

The same logic scales to the matrix. Wienken and colleagues determined the affinity of the small-molecule inhibitor quercetin for the kinase PKA in buffer and in human blood serum as an assay matrix, revealing a 400-fold reduced affinity in serum ^[12]. A complex matrix is not background; it is a competing pool of binders and surfaces. An MST constant is a number attached to a named solution.

Aggregation, adsorption, and the traces that should stop the fit

Two artefacts dominate, and they share a signature. Aggregation removes labelled material from solution; adsorption removes it onto the wall. Both change how much fluorophore occupies the illuminated volume, and both mimic a concentration-dependent signal whenever the ligand is what triggers them. A non-ionic surfactant addressed both at once in the lipase work above ^[11], which is also why a binding curve that appears only when detergent is left out deserves suspicion rather than a figure panel.

Against that, MST can be informative about its own failures, provided you look at raw data. Screening fragments against the kinase MEK1 by automated MST, Linke and colleagues reported that MST identified multiple hits that were confirmed by X-ray crystallography but not detected by orthogonal methods, and that MST also provided information about ligand-induced aggregation and protein denaturation ^[13]. Both halves matter. The technique recovered real binders that other assays missed, and it separately flagged compounds whose effect on the protein was not binding at all.

So fit nothing before inspecting every capillary's initial fluorescence and every raw trace. A titration whose starting fluorescence drifts with ligand concentration has at least two live explanations: a genuine change in the physicochemical environment in the vicinity of the dye, which is itself a binding-sensitive observable ^[4], or straightforward loss of labelled protein to a wall or a pellet ^[11]. A binding isotherm fitted to the amplitude cannot distinguish them, because both produce a monotonic, saturable-looking series.

Making the constant identifiable

Scheuermann and colleagues put the analysis question directly, reporting simulations that demonstrate the limits of K_D that can be measured in current instrumentation and showing that binding kinetics is a major concern in planning and executing MST experiments; two protein-protein interactions were used to illustrate challenges encountered in acquiring and analysing MST data, and the work sets out a set of best practices and introduces software for rigorous data analysis ^[3]. Two operational rules follow. The instrument has a floor set by how far the labelled species can be diluted and still be seen, and below that floor a titration reports the labelled concentration rather than the affinity. And every capillary must have reached equilibrium before the laser fires, which for a slow off-rate means a timed incubation rather than a mixing step.

How much of the residual scatter belongs to the technique and how much to the operator is a question that has been measured. A European working group ran a benchmark across 32 scientific groups in Europe and two in the US, carrying out experiments on 40 Monolith instruments using a standard operation procedure and centrally prepared samples, with a protein-small molecule interaction, a newly developed protein-protein interaction system and a pure dye as test systems; they characterised instrument properties and evaluated instrument performance, reproducibility, the effect of different analysis tools, and the influence of the experimenter during data analysis ^[6]. Their stated motivation is worth rehearsing before presenting any MST number: MST and the related intensity-change measurement have not enjoyed the unambiguous acceptance from biophysicists afforded to other biophysical techniques like isothermal titration calorimetry or surface plasmon resonance, which the authors suggest might be attributed to several facts, among them that various — not fully understood — effects are contributing to the signal ^[6].

What the constant licenses, and what ITC and SPR give instead

An MST titration licenses an equilibrium dissociation constant, obtained in free solution without immobilisation ^[5], for one labelled partner, in one named buffer, read over one declared time window. Three things are routinely taken from it that it does not support.

It does not give rate constants. Extracting those from thermophoresis required a modified experiment: Stein and colleagues extended MST to measure binding kinetics as well as binding affinity in a single experiment by increasing the thermal dissipation of the sample, and used it to measure DNA hybridisation on-rates in the range 10⁴–10⁶ M⁻¹ s⁻¹ and off-rates in the range 10⁻⁴–10⁻¹ s⁻¹; their claim is explicitly conditional, that for biomolecular interactions with large enthalpic contributions the kinetic MST technique offers a robust, cost-effective and immobilisation-free determination of kinetic rates and binding affinity simultaneously ^[5]. That is a property of a modified instrument configuration, not of a standard run.

It does not give thermodynamics. Isothermal titration calorimetry monitors heat changes, released and/or absorbed, caused by macromolecular interactions with no restrictions of buffer and molecular weight of the macromolecules, and allows biophysical characterisation of binding between label-free, non-immobilised, in-solution biomolecules by providing the stoichiometry of the interaction, the equilibrium binding constants and the thermodynamic parameters ^[14]. If the question is an enthalpy, or a stoichiometry that does not rest on interpreting an amplitude, that is the experiment — at a much higher material cost.

And it does not compete with a biosensor on kinetics, though biosensor rates carry a caveat of their own. Schuck and Minton showed that the methods then in use for analysing surface plasmon resonance or resonant mirror biosensor data do not adequately take into account the effects of mass transport on the kinetics of ligand association and dissociation, and that conventional analyses may yield arbitrary apparent reaction rate constants lying between the mass transport rate constant and the true intrinsic chemical binding rate constants, depending on the choice of ligand concentrations used in the experiments ^[15]. The three techniques do not rank; their failure modes sit in different places. MST's are optical and interfacial — the dye's local environment, the wall, the aggregate. The biosensor's are at the immobilised surface and in transport to it. Calorimetry's are material consumption and interactions too enthalpically quiet to detect.

Report an MST result the way it was actually measured. Name the labelling chemistry and, where it is defined, the site, together with the functional control run on the labelled protein ^{[9, 10]}. Name the buffer, the surfactant and its concentration, the capillary type, and the final co-solvent percentage held constant across the series ^{[10, 11]}. State the concentration of the labelled partner relative to the fitted constant, and the incubation used to reach equilibrium ^[3]. Declare the time window from which the ratio was taken, given that the physical basis of the signal in that window has been revised over the technique's lifetime ^[7]. Show the capillary fluorescence scan and the raw traces beside the isotherm, because the artefacts live there and not in the fit ^[13]. And read disagreement with an orthogonal method as information rather than as a problem to be resolved by choosing the number you prefer — in fragment work, MST has recovered crystallographically confirmed binders that orthogonal assays missed ^[13].