Isothermal Chemical Denaturation and Peptide Stability

Heating a peptide until its structure fails tells you where it sits on the temperature axis. Chemical denaturation asks the same question along a different coordinate: hold temperature fixed, titrate a denaturant, and follow a structural signal as the folded population disappears. Differential scanning calorimetry and the analysis of solvent denaturation curves by the linear extrapolation method are the two most common ways of measuring conformational stability ^[1], and they are not interchangeable readouts of one quantity. The isothermal experiment never moves the sample off the temperature you care about, which matters because heating carries its own artefacts: in a formulation case study on a model monoclonal antibody, differential scanning fluorimetry produced misleading physical stability rankings because of buffer-specific pH shifts during heating, and isothermal chemical denaturation was brought in to work around that ^[2].

The trade is throughput and sensitivity. Thermal shift assays remain the dominant format for stability monitoring, but they can give misleading rankings of protein stability and may fail to reflect stability under the conditions of interest; current isothermal chemical denaturation techniques give more precise information about stability and interactions while offering less throughput, and are typically limited to micromolar protein concentrations ^[3].

What the denaturant is actually doing

Urea does not act the way the phrase "hydrophobic effect" would lead you to expect. Group transfer free energies of side-chain and backbone units from water to 1 M urea, analysed through Tanford's transfer model, show that urea does not denature proteins through favourable interactions with nonpolar side chains; what drives urea-induced unfolding is the large favourable interaction of urea with the peptide backbone. Of the surface newly exposed on denaturation, only about 25% contributes favourably, because of newly exposed backbone units, while about 75% modestly opposes urea-induced denaturation and originates from side-chain exposure ^[4].

Guanidinium chloride is stronger and less clean. Measured on a homologous series of peptides with the repeating sequence Ala-Glu-Ala-Ala-Lys-Ala at chain lengths from 7 to 50 residues, GdmCl was, under many conditions, roughly twice as effective a denaturant as urea on a molar basis, in agreement with many studies on proteins. The GdmCl m-value for those peptides was strongly dependent on the total ionic strength of the solution, and controlling ionic strength with added NaCl separated it into two components: one identical to that found for urea, and a second depending only on the molar concentration of chloride. For those peptides an equimolar mixture of urea and NaCl was nearly as effective as GdmCl at unfolding the helical conformation ^[5].

The consequence for a charged peptide is immediate: a GdmCl curve mixes denaturation with electrostatic screening, and the m you extract from it is not directly comparable with a urea m unless the ionic contribution is accounted for. Salt is not inert in urea curves either. KCl affects urea-induced denaturation when the native protein contains charge–charge interactions and/or an anion binding site, in a manner consistent with crystal structure data ^[6].

The test that two denaturants report the same underlying quantity is that they return the same free energy. For phenylmethanesulfonyl alpha-chymotrypsin followed by difference spectroscopy at 293 nm, guanidinium chloride, urea and 1,3-dimethylurea gave unfolding free energy changes identical within error, and identical native-minus-unfolded extinction coefficient differences in the limit of zero denaturant, supporting the view that the free energy change is a property of the protein system and independent of the denaturant used ^[7].

The linear extrapolation model and what m reports

The working equation of the linear extrapolation method is that the free energy of unfolding falls linearly with denaturant concentration, ΔG = ΔG(H₂O) − m[denaturant]. The parameter m measures the dependence of ΔG on denaturant concentration, and much of the recent interest in it turns on what it is taken to say about the change in accessible surface area when a chain unfolds and about the denatured state itself ^[8].

The empirical basis for that reading is a survey of m values collected across a large set of proteins. The m value correlates very strongly with the amount of protein surface exposed to solvent upon unfolding, with linear correlation coefficients of R = 0.84 for urea and R = 0.87 for guanidine hydrochloride, improving to R = 0.90 once the effect of disulfide bonds on the accessible area of the unfolded protein is included. For proteins that undergo a simple two-state unfolding mechanism, the surface exposed on unfolding is a main structural determinant of both the m value and the heat capacity change ^[9].

A peptide-scale calibration exists for the urea case. For alanine-based helical peptides of the same repeating sequence at chain lengths from 14 to 50 residues, the logarithm of the Zimm–Bragg helix propagation parameter fell linearly with urea molarity at 0 °C, and multiplying the m value determined for helix unfolding by the number of helical residues in partly helical protein molecules gave values agreeing within a factor of two with the m values observed for those proteins — evidence that the interaction between urea and peptide groups accounts for a major part of the denaturing action of urea on proteins ^[10].

Never read m as a fixed property of a sequence. For ribonuclease A the dependence of ΔG on urea concentration rises from about 1200 cal mol⁻¹ M⁻¹ at high pH to about 2400 cal mol⁻¹ M⁻¹ at low pH, which suggests that the unfolded conformations become more accessible to urea as the net charge on the molecule increases; for ribonuclease T1 the same quantity is minimal near pH 6 and increases at both higher and lower pH. On that evidence the unfolded states of proteins in urea and GdnHCl may differ significantly in the extent of their interaction with denaturants, so the conformations assumed by the unfolded state may depend to some extent on sequence ^[11].

What the extrapolation to zero denaturant assumes

Three assumptions carry ΔG(H₂O). The first is two-state behaviour, with no measurable population of intermediates. The second is that the pre- and post-transition baselines can be extended linearly into the transition zone; in the chymotrypsin study cited above, the independence of the native-minus-unfolded extinction difference from the choice of denaturant was what supported that linear extension and licensed evaluation of unfolding equilibrium constants on the two-state assumption ^[7]. The third is that the linear relation between ΔG and denaturant concentration holds all the way down to zero, far outside the window in which it was observed.

That third assumption is the contested one, and it has been tested head-on. The relation can be measured directly over only a small range, approximately ±0.8 M urea around the midpoint of the unfolding transition. Calorimetric measurement of barnase unfolding at urea concentrations between 0 and 4.5 M gave a ΔG at 298 K that deviates systematically from linearity and fits ΔG = 10.5 − 2.65 × [urea] + 0.08 × [urea]² kcal mol⁻¹. The curvature makes apparent m values increase when measurements are made at lower urea concentrations, which could account for increases in m at low pH or in destabilised mutants, since those denature at lower denaturant concentrations. The work was prompted by a persistent 1.6 kcal mol⁻¹ discrepancy between the extrapolated value for barnase and the equivalent value from thermal unfolding measurements. The same analysis notes that small curvature leads to negligible errors in the change in unfolding free energy on mutation, provided the curvature is similar for all mutants ^[12].

The opposing result is equally direct. Using GdnHCl to shift the urea transition into an accessible window and correcting for the GdnHCl contribution, the urea dependence of ΔG for lysozyme, ribonuclease A and cytochrome c was found to be linear over the full concentration range, with the effects of the two denaturants additive ^[6].

Neither result settles the general case, which is exactly why the reporting convention matters. Report C_m, the denaturant concentration needed to unfold half the population, and m as the measured quantities, and ΔG(H₂O) as a model-dependent extrapolate. Under the linear model ΔG(H₂O) is the product of m and C_m, so any bias in m propagates undiminished into the extrapolated free energy — and the error in the free energy obtained by linear extrapolation has been argued to be significantly larger than commonly reported in the literature ^[7]. In the antibody formulation study noted earlier, the concentration dependence of the Gibbs free energy of unfolding gave better predictions of physical stability than either the melting temperature from differential scanning fluorimetry or C_m itself ^[2].

Reversibility is a precondition, not a formality

Free energy is a state function and therefore path independent, so extrapolating one from a denaturation curve is defensible only when the unfolding and refolding processes are themselves path independent and reversible, and can give erroneous results when they are not. Reversibility was demonstrated in several early studies on smaller proteins but was assumed in some reports on large proteins with complex structures. A systematic survey using an automated isothermal chemical denaturation instrument found that the apparent unfolding mechanism and conformational stability of large proteins can be denaturant-dependent, that equilibration times for large proteins are non-trivial and may introduce significant error into calculations of ΔG, and that irreversible unfolding and hysteresis can occur in the absence of aggregation ^[13].

That last finding is the one that catches people out: a clear, non-scattering solution is not evidence of reversibility. Demonstrating reversibility takes a deliberate experiment. Unfold the peptide at a denaturant concentration well past the transition, dilute into matched buffer to land on a series of points spanning the transition, incubate, then check that the refolding titration superimposes on the unfolding titration point for point and that the plateau signal is fully recovered. The recommendation drawn from that survey is explicit: confirm the state functions by, for example, performing refolding experiments or using additional denaturants ^[13]. Equilibration time needs the same scrutiny — re-read the whole series after a longer incubation and confirm the curve has stopped moving before fitting it.

Probes disagree, and the disagreement is data

A denaturation curve is a curve in whichever signal you chose to record. Solvent denaturation curves can reveal useful structural information such as the existence of domains or the presence of stable folding intermediates ^[8], but only when more than one probe is used, because one probe on its own cannot separate a genuine two-state transition from coincidentally overlapping ones. Two probes returning the same C_m and the same m are evidence for two-state behaviour; non-coincidence is evidence against it. In the equilibrium unfolding of a cyclophilin carrying a lone partially exposed tryptophan, the C_m values from fluorescence and from circular dichroism did not coincide, indicating a non-two-state transition; the conclusion was corroborated by bis-ANS binding and by the lack of an isochromatic point in the far-UV CD ^[14]. The instrument survey cited above likewise reported that fluorescence emission spectroscopy may not correspond to other methods such as circular dichroism when used to follow unfolding ^[13].

Where a peptide offers no usable intrinsic fluorophore, an external reporter is the alternative. A Förster resonance energy transfer probe has been applied as an external probe for isothermal chemical denaturation with three chemical denaturants at nanomolar protein concentration, and the same probe is applicable to the conventional thermal shift format ^[3]. Any added reporter is one more component in the equilibrium, so run it against a peptide-free denaturant series across the full titration before reading anything into the curve shape.

Short peptides: where the two-state assumption is least safe

A short helix does not unfold in two states, and the peptide work never claimed it did. The urea unfolding of the alanine-based helices was analysed with helix–coil transition theory rather than a two-state equilibrium, and both a binding-site model and a solvent-exchange model for the interaction of urea with the peptides described the data well enough that the two could not be told apart ^[10]. A helix–coil transition is broad and multi-state by construction; forcing a two-state expression onto it returns numbers, but not the numbers the equation is supposed to mean.

Cooperative peptides do exist. The 33-residue leucine zipper segment GCN4-p1, in its M2V form, was mapped by CD-monitored urea unfolding at a series of temperatures and temperature-induced unfolding at a series of urea concentrations; a global analysis of the combined equilibrium data gave strong support for a two-state mechanism, with no detectable population of intermediate states, consistent with differential scanning calorimetry and with thermal CD melts run as a function of peptide concentration. The relative magnitudes of the urea dependence of the unfolding and refolding rate constants indicated that 55–65% of the surface area is buried in the transition state ^[15].

That cooperativity is fragile. Moving the single Asn residue of the GCN4 leucine zipper from the a position at residue 16 to the a position at residue 9 turned the thermal unfolding three-state: the variant unfolds first from its N terminus and changes its oligomerisation specificity, passing through a partially unfolded intermediate containing a mixture of dimers and trimers before reaching unstructured monomers — a pathway established by thermal denaturation, limited proteinase K digestion and sedimentation equilibrium analysis ^[16].

Two operational corollaries follow. First, when the folded species is an oligomer, unfolding is coupled to dissociation, so the transition midpoint shifts with peptide concentration and the association state has to be established independently rather than assumed; sedimentation equilibrium served that purpose in the leucine zipper work above ^[16]. Second, marginal stability destroys the baseline. If a peptide is already substantially disordered at zero denaturant there is no pre-transition baseline left to extend, and the observation that justified extending baselines into the transition zone ^[7] has nothing to act on. In that regime the honest output is a C_m and a transition width, with the extrapolated free energy flagged as unreliable rather than quoted to two decimal places.

What the measurement licenses

An isothermal chemical denaturation experiment licenses a statement about the relative stability of samples measured under identical conditions — the same temperature, pH, ionic strength, peptide concentration, probe and incubation time — provided each was shown to refold. It licenses a C_m and an m for each. It does not on its own license a free energy in water, unless the two-state model is supported by agreement between probes, the reversibility check has actually been run, and you are prepared to defend the linear extrapolation across the gap between the transition and zero denaturant. It reports a conformational equilibrium and nothing about covalent integrity. The defensible position is the one the formulation work arrived at: combine the isothermal and thermal readouts rather than choosing between them ^[2], use agreement between denaturants as the consistency test on the free energy ^[7], and use agreement between probes as the test of the two-state assumption ^[14].