A buffer is usually treated as a background condition, chosen once and forgotten. It is closer to a reagent present in large molar excess: it fixes the protonation state of every ionizable group on a peptide, supplies most of the ionic strength, can bind the metal you are titrating, can form a covalent adduct with the analyte, and in calorimetry enters the reported number directly.
What follows covers what the pKa on the bottle does not tell you, how much capacity there actually is, why ionic strength is a variable you set by accident, which buffers break which readouts, and the ionization enthalpy calorimetry cannot ignore.
What the number on the bottle is
The Henderson-Hasselbalch expression, pH = pKa + log10([A-]/[HA]), is the definition of the acid dissociation constant rearranged rather than a finding, uncited here. At pH equal to the pKa the two forms are equimolar; one unit above, the ratio is 10:1, so 10/11 of the buffer — 90.9% — is basic and 9.1% remains acid.
Much of what sits on a modern bench shares one ancestry: Good and five colleagues published a set of hydrogen ion buffers intended for biological research in 1966 [1], and buffers of that lineage still carry his name, TES and bicine among them [2]. The name marks that shared origin rather than a chemical class, and a buffer chosen well for one problem is not thereby chosen well for another — a general caution stated here as convention, not as a finding of either paper.
The pKa is a function of temperature, and so is the pH you set
A pKa is not a fixed property of a compound. Its temperature dependence is set by the enthalpy of deprotonation: in van 't Hoff form, dpKa/dT = -ΔH_ion / (2.303 R T^2), standard thermodynamics, uncited. A buffer whose deprotonation absorbs a lot of heat moves a lot.
Fukada and Takahashi determined enthalpy and heat capacity changes for the deprotonation of 18 buffers calorimetrically in 0.1 M potassium chloride from 5 to 45 degrees Celsius. For every buffer except phosphate and glycerol 2-phosphate the enthalpy change was linear in temperature; for those two, where a divalent anion forms on dissociation, a second-order function fitted better. Buffer pH calculated from their values was in good agreement with pH actually measured between 0 and 50 degrees Celsius for all buffers studied, and they publish a table of pK against temperature for the 18 [3].
Run the arithmetic with invented enthalpies, to show the size of the effect. With R = 8.314 J per mole per kelvin and T = 298.15 K, 2.303 R T^2 = 2.303 x 8.314 x 88,893, about 1.70 x 10^6 J K per mole. A supposed ΔH_ion of 45 kJ per mole gives dpKa/dT = -45,000 / (1.70 x 10^6) = -0.0264 per kelvin. Cooling from 25 to 4 degrees Celsius is -21 K, so the pKa rises by 0.0264 x 21 = 0.55 units, and if the base-to-acid ratio is unchanged so does the pH: a solution titrated to 7.4 on the bench sits near 7.95 in the cold room. A supposed 5 kJ per mole gives 0.06 units.
That treats ΔH_ion as constant across the interval, which is exactly what a nonzero heat capacity change says it is not [3]. Calibrating the electrode at the working temperature — convention, uncited — corrects the meter, not the pKa shift.
Buffer capacity, and how much of it there is
Buffer capacity is the strong base needed to move the pH by one unit. For a monoprotic buffer at total concentration C, neglecting water's own autoprotolysis, beta = 2.303 x C x Ka x [H+] / (Ka + [H+])^2 — textbook, uncited. At pH = pKa, [H+] = Ka, so the fraction is Ka^2 / (2 Ka)^2 = 1/4 and beta = 0.576 C. At pH = pKa + 1, [H+] = Ka/10, so it is (Ka^2 / 10) / (1.1 Ka)^2 = 0.1 / 1.21 = 0.0826 and beta = 0.190 C. Dividing by the peak value of 0.25 — the peak is the denominator — gives 0.33: one pH unit out, a third of the capacity remains.
For a 50 mmol per liter buffer at its pKa, beta = 0.576 x 50 = 28.8 mmol per liter per pH unit; one unit off, 9.5. Let 1 mmol per liter of strong acid enter the tube: at the pKa the pH moves by roughly 1 / 28.8 = 0.035 units, one unit away by 1 / 9.5 = 0.105, three times as far. Both are linearized estimates: beta is a derivative, the true displacement an integral.
Ionic strength is something you set by accident
Ionic strength is I = 0.5 x sum over ions of c z^2, a standard definition, uncited. Two buffers at the same molarity and pH are usually not at the same ionic strength: a divalent ion counts four times as much and brings more counterions.
Compare 50 mM sodium phosphate with 50 mM Tris-chloride, both nominally pH 7.4. Suppose the second pKa of phosphate is 7.20 and the pKa of Tris is 8.06 at 25 degrees Celsius — commonly tabulated values, used as suppositions here rather than results from any source cited below. Phosphate: pH - pKa = 0.20, so [HPO4 2-]/[H2PO4 -] = 10^0.20 = 1.585, and 1.585/2.585 = 61.3% of the 50 mM is divalent, 30.7 mM, leaving 19.3 mM monovalent. Charge balance needs 80.7 mM sodium, so I = 0.5 x (80.7 + 122.8 + 19.3) = 111 mM, 2.2 times the nominal concentration.
For Tris-chloride the pair is the cation TrisH+ and neutral Tris, so only the protonated fraction contributes, matched by chloride: 1 / (1 + 10^(7.4 - 8.06)) = 1 / 1.219 = 0.820, giving 41.0 mM each and I = 41 mM. Swapping one 50 mM buffer for the other changes ionic strength by a factor of 111 / 41 = 2.7 while changing nothing on the label.
In work on the model amphiphilic peptide RADA 16-I, fibril-fibril aggregation at pH 2.0 was followed by dynamic light scattering and analyzed as an activation energy barrier; the authors report that the barrier's dependence on ionic strength could be described quantitatively as charge screening within DLVO theory, although specific anion and cation effects were also observed [4].
The opposite result exists too. Tyler-Cross and Schirch measured nonenzymatic asparagine deamidation in two series of pentapeptides, VSNXV and VXNSV, with X taken from 10 different amino acids. They report that the deamidation rates of the peptides studied were not affected by ionic strength and were not specific base catalyzed, while general base catalysis was observed for small bases such as ammonia [5]. Ionic strength is a variable to control and report, not automatically one that moves a readout.
Buffers that are also reagents
The clearest failure mode is metal binding. Nakon and Krishnamoorthy determined formation constants for TES and bicine and found the resulting chelates undergo loss of an internal alcohol proton for bicine, hydrolysis for both, and dimerization for TES; they conclude these two buffers buffer not only hydrogen ions but also metal ions, and that their metal complexes in turn buffer hydrogen ions by secondary reactions [2]. A free metal concentration calculated from what was weighed in is not the one in the tube.
A methods account of isothermal titration calorimetry with metal ions makes the same point: such measurements are susceptible to competing reactions — oxidation, precipitation, hydrolysis — and to coupled reactions involving the buffer and protons, and delivering the metal as a well-defined buffer complex can suppress unwanted chemistry, at the cost of having to know that complex's thermodynamics [6].
Buffers can generate chemistry as well as bind it. Morris and Zinser measured hydrogen peroxide production in illuminated seawater media containing a range of zwitterionic buffers; rates varied extensively between them, HEPES highest and TAPS among the lowest, and accumulation increased linearly with buffer concentration and light level. HEPES at 1 to 10 mM, concentrations they describe as commonly used in culture media, generated enough peroxide to kill the axenic Prochlorococcus strain VOL1 in enriched seawater media at lower, environmentally realistic cell concentrations or under high light [7]. That concerns illuminated seawater and one organism, not HEPES in the dark. It does show a buffer can be a slow oxidant source.
Covalent adduct formation is the least expected mode. Richter and colleagues found that the model compound clerocidin forms adducts with both Tris and phosphate by two routes: a kinetically controlled reaction of its carbonyls, especially the C15 aldehyde, with the Tris amino group to give a Schiff base, and a thermodynamically controlled attack on an oxirane ring by the buffer's nucleophiles, the amino nitrogen in Tris and an oxygen in phosphate [8]. They conclude that reactivity toward buffers cannot be directly predicted from the functional groups potentially involved in the initial reaction.
Which buffers break which readouts
Interference is assay-specific. Lleu and Rebel studied low concentrations of HEPES and other common buffers in protein determination and found that some interfere, to differing degrees, with the Lowry method; a structure-interference comparison led them to suggest the ethanolamine group is involved. They observed no interference when protein was measured with bicinchoninic acid at the same concentration as the Lowry reagent [9].
Three further exclusions are convention, not findings from any study cited: primary-amine buffers are kept out of amine-reactive chemistry, far-ultraviolet absorbance limits how far down a circular dichroism spectrum can be collected, and nonvolatile salts are exchanged out before electrospray mass spectrometry.
The ionization enthalpy that matters for calorimetry
In isothermal titration calorimetry the buffer is not background. If binding takes up or releases protons, they come from or go to the buffer, and the heat of the buffer's own protonation is measured along with the heat of binding: ΔH_obs = ΔH_intrinsic + n_H x ΔH_ion, where n_H is the protons taken up per binding event.
Baker and Murphy set out the theory for binding linked to protonation of an ionizable group on a protein. They show that running experiments as a function of pH in buffers with varying ionization enthalpy yields the pKa values of the responsible group in the free and the liganded states, its protonation enthalpy in those states, and the intrinsic energetics of binding; and they state that the observed binding enthalpy and heat capacity change can undergo extreme deviations from their intrinsic values depending on pH and buffer conditions [10].
The design is a straight-line fit. Suppose a titration in a buffer with ΔH_ion supposed to be 5 kJ per mole gives ΔH_obs = -20.0 kJ per mole, and the same titration in one supposed to be 45 kJ per mole gives -8.0 kJ per mole. All four numbers are inventions for the arithmetic. The slope is (-8.0 - (-20.0)) / (45 - 5) = 12.0 / 40 = 0.30 protons per binding event; the intercept is the intrinsic enthalpy, -20.0 - (0.30 x 5) = -21.5, confirmed from the other point as -8.0 - 13.5 = -21.5 kJ per mole. The two observed values differ by 12.0 kJ per mole, which is 12.0 / 21.5 = 56% of the magnitude of the intrinsic enthalpy — the intrinsic value is the denominator here and below. That spread is not the error of either measurement, though. Each observed value sits above the intrinsic one by n_H x ΔH_ion: by 0.30 x 5 = 1.5 kJ per mole in the low-enthalpy buffer, which is 1.5 / 21.5 = 7.0%, and by 0.30 x 45 = 13.5 kJ per mole in the high-enthalpy one, 13.5 / 21.5 = 63%. So the error of reporting an observed value as the binding enthalpy scales with the buffer's ionization enthalpy, and its sign is the sign of the product n_H x ΔH_ion, so it flips only if the proton stoichiometry or the buffer's ionization enthalpy changes sign, not merely because the buffer was swapped: here both terms are positive, so both observed values are less negative than the intrinsic enthalpy, one barely and one by nearly two thirds of it.
Kaul and colleagues used buffer-dependent calorimetry on three aminoglycosides binding a 16S rRNA A-site model oligomer: at pH 5.5 binding was independent of drug protonation effects for all three, while at pH 9.0 the binding of paromomycin I and neomycin B was coupled to the uptake of 3.25 and 3.80 protons respectively, and lividomycin A at pH 8.8 to 3.25 protons [11].
How many buffers, at what pH values and temperatures, is itself a design question. Coussens and colleagues simulated families of calorimetry data sets embedding different strategies for the number of experiments, the pH range, the buffer ionization enthalpy and the temperature, then reanalyzed them by global analysis with a proton linkage model. They report that the global analysis can yield reliable estimates of the intrinsic binding and protonation thermodynamics, and that comparing strategies illustrates the benefit of experiments at a range of temperatures [12]. That is a simulation study of design, not a measurement of a system. The ionization enthalpies it depends on are themselves measured quantities [3], at an ionic strength that is not automatically yours.
Freezing changes the pH of a phosphate buffer
A buffer can stop behaving as it freezes, because one component crystallizes out and leaves its partner behind. Gomez and colleagues froze sodium phosphate buffers of 8 to 100 mM at initial pH values between 5.7 and 7.4, cooling to -10 degrees Celsius while monitoring pH in situ. Precipitation of the dodecahydrate of disodium hydrogen phosphate caused abrupt pH decreases beginning between -0.5 and -4.0 degrees Celsius. From an initial pH of 7.4 the 50 and 100 mM solutions reached 4.2 plus or minus 0.1 at -10 degrees Celsius, while the 8 mM solutions reached 5.2 plus or minus 0.2 [13].
Note the direction: the more concentrated buffer gave the larger acidification, the opposite of the intuition that more buffer means more protection. That study measured pH, not peptide integrity, so nothing in it says a peptide is damaged by an excursion to pH 4.2. It does say a freeze-thaw step is a pH experiment whether or not one was intended.
What is convention, and what is unknown
What is not known, in general, is whether your buffer interferes with the system in front of you. Interference is a property of the pair, and the published catalog of artifacts is a list of cases somebody happened to investigate. Absence from it is not evidence of inertness — phosphate's reputation as a largely inert physiological buffer is what made the clerocidin adducts worth reporting [8].
The response is not to find the one safe buffer but to record enough to diagnose a discrepancy later: buffer identity and lot, total concentration, the temperature at which the pH was set, the titrant, the final ionic strength including counterions, and whether the material was frozen in it. Where a result should be sensitive to any of this, the strongest check is to repeat it in a chemically unrelated buffer at matched ionic strength. If the number survives, the buffer was a background condition. If it does not, the buffer was part of the experiment, and always had been.