The Second Virial Coefficient by Static Light Scattering

Two molecules of the same solute in the same buffer either find each other more often than random placement would predict, or less often. The osmotic second virial coefficient collapses that preference into one number by integrating the pair interaction over separations and orientations, and static light scattering gets at it not from the intensity of any single sample but from how the intensity changes as you dilute. That is the whole reason the technique earns its place: the interaction term lives in a slope, not in an absolute reading, so it survives a great deal that would corrupt an absolute reading.

What the Debye plot gives you

For a concentration series the working relation is K*c/R(c,θ) = 1/[M·P(r_g,θ)] + 2B₂₂c, with R the excess Rayleigh ratio at scattering angle θ, c the mass concentration, K* the optical constant and P the particle form factor ^[2]. Plot the left-hand side against c and the intercept returns the reciprocal molar mass while the slope returns twice the virial coefficient. One dilution series therefore delivers a mass and an interaction term from two different features of the same fit, which is convenient and also the source of most of the trouble, because the two features fail independently and a plot can be wrong in one without looking wrong in the other.

The sign convention is fixed: positive values of B₂₂ indicate a predominant mutual repulsive interaction, and negative values a predominant attractive interaction ^[2]. The load-bearing word is predominant. What the slope reports is the net of everything the pair does to each other, weighted over every relative orientation the pair can adopt, so a coefficient near zero is not an absence of interaction. It is a cancellation, and it licenses you to say only that the solution behaves close to ideally over the concentration window you sampled — the coefficient is, in the end, a parameter of solution non-ideality ^[11].

The intercept deserves the same suspicion as the slope. A Kirkwood–Buff re-analysis of Rayleigh scattering in multi-component solutions establishes that for dilute solutions light scattering can be used to obtain B₂₂ rigorously and unambiguously, and in the same treatment shows that the apparent molecular weight obtained from light scattering is generally not equal to the true molecular weight, the difference arising from a combination of protein–solute and protein–cosolute interactions ^[6].

A coefficient without its buffer is not a result

Static light scattering and small-angle neutron scattering on lysozyme and chymotrypsinogen gave virial coefficients in very good agreement without the use of adjustable parameters, and the same study mapped how far those coefficients move with solution conditions: at low electrolyte concentration they depend strongly on pH and change from positive to negative as pH increases, while at high salt concentration all coefficients are slightly negative and depend weakly on pH ^[3]. The salt dependence is not one trend you can memorise and reuse. For lysozyme the coefficients always decrease with increasing electrolyte concentration, whereas for chymotrypsinogen there is a cross-over point around pH 5.2, above which the coefficients decrease with increasing ionic strength, which the authors read as attractive electrostatic interactions ^[3].

Lysozyme measured instead by self-interaction chromatography reproduces that picture from the other direction: under acidic solution conditions B₂₂ is positive, favouring repulsion, and with increasing NaCl concentration it decreases and becomes negative; the coefficient also reflects, in general, the salting-in and salting-out ranking of the Hofmeister series ^[11]. For bovine serum albumin measured by composition-gradient multi-angle static light scattering, the systematic variation of B₂₂ with pH and NaCl concentration reveals the net charge change and the isoelectric point, the magnitude falling to 1.13 × 10⁻⁵ mL·mol/g² near the isoelectric point of pH 4.6 at 25 mM NaCl ^[2].

DLVO-type modelling that accounts for repulsive and attractive electrostatic, van der Waals and excluded-volume interactions of equivalent colloid spheres agreed with the lysozyme and chymotrypsinogen data, but was unable to resolve the complex short-ranged orientational interactions ^[3]. Since the same protein moves from repulsive to attractive across a pH unit or a salt step, a coefficient reported without its buffer composition, pH, salt identity and concentration, and temperature is not transferable to anyone else's bench; temperature belongs in that list because the correlations built on these coefficients are fitted across variations in protein concentration, salt type and concentration, temperature and pH ^[12].

The concentration series is the experiment

Static light scattering determinations of the coefficient typically use protein concentrations of less than 5 mg/mL ^[16]. How many points and how well you know them matters more than the instrument. In a fully Bayesian treatment of static light scattering that jointly infers concentration, refractive index, oligomer size and the second virial coefficient, hen egg-white lysozyme was diluted sequentially to fourteen nominal concentration levels from 2.5 to 50 mg/mL; the accompanying simulation study identified precision of the concentration measurement as the primary driver of uncertainty, with discrepancies between measured and actual concentrations producing substantial bias, particularly where the coefficient is small in absolute terms ^[15]. Ordinary least-squares fitting of a Debye plot treats concentration as though it were known exactly, and it is not.

The upper end of the series is bounded by the linear form itself. Once the solution is no longer dilute, conventional analysis overestimates the magnitude of B₂₂ for significantly repulsive conditions and underestimates it for significantly attractive ones, the protein–protein Kirkwood–Buff integral G₂₂ becomes the more relevant quantity obtainable from scattering, and a simple approach is available both for deciding whether the protein concentration is dilute enough for B₂₂ to apply and for correcting values regressed at higher concentration ^[6]. Curvature in the plot is information, not noise to be fitted through.

Dust is the other half of the problem, and it follows directly from the working equation: in the dilute limit the excess Rayleigh ratio scales with the product of molar mass and mass concentration ^[2], so a contaminant a thousand times the mass of the analyte contributes the same scattering at a thousandth of the mass concentration. Nothing in the fit flags it. In the composition-gradient study of bovine serum albumin, samples were filtered through a 0.1 µm syringe filter followed by a 0.02 µm syringe filter to remove aggregates and possible contaminants prior to analysis ^[2]. Run the buffer through the same filtration train, because the baseline is subtracted from every point and a dirty baseline shifts the intercept and tilts the slope at once.

Composition-gradient methods

Composition-gradient multi-angle static light scattering replaces manual dilution with automated mixing, delivery and measurement, giving high-speed, continuous, fluctuation-free sample delivery ^[2]. The practical consequence is throughput and reduced handling error rather than a different physical quantity: in the bovine serum albumin work, a protein concentration of 1.5–35 mg/mL was reported as appropriate for the chemical conditions studied, with one gradient stepping through six sample concentrations from 17.6 down to 4.4 mg/mL, and five measurements of the coefficient could be completed in three hours ^[2]. Since concentration precision dominates the uncertainty budget, in-line gradient generation attacks the dominant error term directly rather than the instrument noise.

Self-interaction chromatography

The chromatographic route immobilises protein on chromatographic particles and measures the retention of the same protein in isocratic elution; the relative retention reflects the average protein interactions, which are related to the second virial coefficient via statistical mechanics ^[7]. Quantitative agreement between coefficients measured this way and by traditional characterisation methods was obtained for both lysozyme and chymotrypsinogen over a wide range of pH and ionic strengths, for at least an order of magnitude less time and protein than the alternatives ^[7].

The accuracy sits in the peak analysis, not the column. Non-Gaussian peaks are commonly encountered in self-interaction chromatography, with typical peak asymmetries of 10% for lysozyme reflecting a non-linear chromatographic retention process, so centre-of-mass calculations are needed to determine accurate retention volumes; the empirical peak-maximum analysis often reported in the literature can result in errors of up to 50% in the coefficient ^[8].

For short chains the chromatographic method is not an alternative but the only option of the two. Peptides do not scatter enough light to allow static light scattering measurement of the coefficient, whereas self-interaction chromatography does not carry the molecular size limitation ^[10]. In the first such application, the coefficient for enfuvirtide, a 36-amino-acid peptide, was measured as a function of salt concentration, salt type and pH; it correlated strongly with solubility and apparent molecular weight, and the peptide concentration itself was shown to have a significant effect on the coefficient, but only at concentrations above 80 mg/mL ^[10]. That last figure is worth carrying: a chromatographic measurement made on a dilute mobile phase and a determination made in a concentrated solution are not automatically the same number.

Solubility and the crystallisation slot

The oldest practical use of the coefficient is a phase-behaviour screen. Coefficients measured by static light scattering for a variety of proteins in solvents known to promote crystallisation were found to lie within a fairly narrow range, referred to as a crystallisation slot, while solution conditions known not to favour crystallisation gave values well outside that slot ^[1]. The relation to equilibrium solubility was later put on a thermodynamic footing: a correlation derived from classical thermodynamics, based on equality of fugacities of the protein in the equilibrium phases, was fitted mainly to lysozyme data and describes the experimental data both qualitatively and quantitatively, with very good agreement for protein solubilities up to 30 mg/mL; above that value the model underpredicts the data, probably as a result of multibody interactions the pairwise treatment omits ^[12]. For lysozyme, the coefficient measured chromatographically was likewise found to correlate with protein solubility as well as with physical protein stability ^[11].

Viscosity and aggregation propensity

The link to concentrated-solution behaviour is real but qualitative. For three monoclonal antibodies whose coefficients were estimated by sedimentation-equilibrium analytical ultracentrifugation at initial loading concentrations below 10 mg/mL, and whose viscosity and stability were assessed above 100 mg/mL, the coefficient depended on pH and correlated qualitatively with both viscosity and aggregation propensity: the more negative the value, the more viscous the solution and the greater the propensity to aggregate ^[13]. The detail that matters for interpretation is what happened on either side of zero. Solutions with coefficients around 2 × 10⁻⁵ mL·mol/g² acquired similar viscosity and aggregation propensity regardless of amino acid sequence, whereas for solutions with negative values the viscosity and aggregation propensity differed depending on sequence, which the authors suggest reflects the influence of the surface properties of the individual antibodies; their conclusion is that the coefficient can be used, within certain limitations, as an effective indicator of the viscosity and aggregation propensity of highly concentrated antibody solutions ^[13].

A chromatographic study across four proteins — lysozyme, recombinant human lactoferrin, concanavalin A and catalase — in solutions of varying salt type, concentration and pH found a linear correlation between the coefficient and the initial hydrodynamic size of the species present measured by dynamic light scattering, for all proteins and almost all solution conditions, with aggregate sizes below roughly 10 nm consistently accompanied by positive coefficients ^[9]. The trends themselves were strongly protein-dependent: catalase and concanavalin A showed attractive interactions at low ionic strengths, while lactoferrin stayed highly positive across a wide range of conditions ^[9].

The average that hides the interaction

The most important limitation is structural rather than instrumental. Calculations that account for steric, electrostatic and short-range interactions with the structural and functional anisotropy of the protein explicitly represented show that the orientational dependence has a pronounced effect: the relatively few protein–protein configurations in which the apposing surfaces display geometric complementarity contribute disproportionately strongly to the coefficient, and the importance of electrostatic interactions is amplified in those high-complementarity configurations ^[4]. A single number that is dominated by a handful of orientations cannot tell you which orientations they were.

Whether such patch interactions dominate at all is itself condition-dependent. A series of exactly solvable models shows a delicate balance between the strength of attractive patch–patch interactions and the patch size, with repulsive patch–patch interactions contributing significantly to the coefficient only where the repulsions are long-ranged; strong patch–patch attraction will make anisotropic interactions dominate only if the unavoidable entropic penalty is overcome, which cannot occur if the patches are too small ^[5]. The same analysis suggests a cheap experimental probe: the temperature dependence of the coefficient may be a simple means of assessing whether a small number of strongly attractive configurations dominate the dilute-solution behaviour ^[5].

Reversible self-association is the concrete case where the average misleads. Light scattering at low concentration is frequently analysed on the assumption that the protein exists either as a single non-associating species or as a fixed combination of assembly states; comparing coefficients calculated with and without accounting for weak reversible dimerisation shows that equilibrium dimer systems with dissociation constants below 20 mM for a 20 kDa protein, and below 1 mM for a 150 kDa protein, significantly affect the apparent coefficient, and that correct interpretation requires an expanded model separating monomer–monomer, monomer–dimer and dimer–dimer terms ^[16]. Heterogeneity shows up chromatographically too: the variance in the coefficient due to chemical effects can be quantified statistically as a measure of the heterogeneity of protein–protein interactions in solution, and for lysozyme a wide range of values is measured that can vary significantly from the average ^[8].

What this is not: the DLS interaction parameter, and viscosity

The concentration dependence of the diffusion coefficient measured by dynamic light scattering gives k_D, which is routinely treated as a stand-in for the virial coefficient. It is not one. Accurate determination of the coefficient from transport data requires independent measurement of k_D and the sedimentation interaction parameter k_s by orthogonal techniques; using sedimentation velocity together with dynamic light scattering on hen egg-white lysozyme and a monoclonal antibody in different salt solutions, the two parameters were of comparable magnitude, and solution conditions were found under which they assumed opposite signs and under which k_D was greater than or equal to k_s ^[14]. In that work the aggregation of the antibody correlated well with the coefficient and with both transport parameters, which the authors present as establishing the potential for k_D to serve as a high-throughput predictor of aggregation ^[14] — but a k_D alone is a composite of thermodynamic and hydrodynamic terms, and reporting it as a virial coefficient overstates what the dilution series established.

The same caution runs the other way toward the viscometer. Viscosity at high concentration is a many-body property; the correlation reported for the pairwise coefficient was explicitly a qualitative one, with no quantitative prediction established ^[13].

What to record

Report the buffer composition, pH, salt identity and molarity, and temperature alongside every value, in stated units. Report the concentration range, the number of points, and how concentration was assayed, because that assay carries the dominant uncertainty ^[15]. Report the filtration train. State whether the plot was linear across the whole range, and give the apparent molar mass from the intercept next to the coefficient from the slope: an apparent mass well above the sequence mass is the fastest indication that what you fitted was not a monodisperse monomer, and the intercept is in any case not guaranteed to equal the true molecular weight ^{[6, 16]}.