Isoelectric Point and Net Charge in Peptides

A peptide's isoelectric point is not a value you look up. It is the root of an equation you assemble yourself, and the equation is only as good as its inputs. Paste a sequence into a pI calculator and it returns a number with two or three decimal places; that precision is manufactured by the arithmetic, not by the molecule. What the calculation does is sum the fractional charge on every ionizable group across trial pH values and find where the sum crosses zero. For peptides and proteins that means the dissociation constants of the charged groups of seven amino acids plus the amino and carboxyl groups at the polypeptide termini ^[1]. Everything downstream follows from how well those pKa values describe your particular chain: how much confidence a computed pI deserves, what a focused band in a gel entitles you to claim, and which pH to load an ion exchanger at.

The pKa values you feed in were measured on other molecules

Tabulated pKa sets come from model compounds chosen so that the ionizable group sits in a defined, minimal environment. Thurlkill and colleagues used potentiometric titration of alanine pentapeptides with appropriately blocked termini and reported, in 0.1 M KCl at 25 °C: 3.67 ± 0.03 for the α-carboxyl, 3.67 ± 0.04 for Asp, 4.25 ± 0.05 for Glu, 6.54 ± 0.04 for His, 8.00 ± 0.03 for the α-amino group, 8.55 ± 0.03 for Cys, 9.84 ± 0.11 for Tyr and 10.40 ± 0.08 for Lys ^[2]. Several of these depart from the Nozaki and Tanford values the authors note are often used in the literature: Asp 3.67 against 4.0, His 6.54 against 6.3, the α-amino group 8.00 against 7.5, Cys 8.55 against 9.5 and Tyr 9.84 against 9.6 ^[2]. Half a pH unit on the N-terminal amine and nearly a full unit on cysteine are not rounding differences — they move the computed titration curve, and with it the crossing point.

So the choice of table is a choice about the answer. A benchmark of five pI prediction methods against the PIP-DB database and a publicly available dataset concluded that the methods vary in their accuracy and are, in the authors' words, highly sensitive to the choice of basis set; machine-learning approaches, and the support-vector-machine algorithm in particular, performed best on peptide mixtures ^[3]. That sensitivity is the practical argument for carrying a computed pI as an interval rather than a point.

How wide the interval is

Two benchmarks put numbers on it. Using computationally optimized pKa sets, the Isoelectric Point Calculator reported an average error in pI estimation of 0.87 pH units for proteins and 0.25 pH units for peptides ^[4]. The deep-learning successor reported root-mean-square deviations of 0.848 for proteins against 0.868 for the previous best, and 0.222 for peptides against 0.405 ^[1]. Note which number belongs to which class: in both benchmarks peptides are predicted several times more tightly than proteins. The same work found that predicting individual pKa values from sequence alone outperformed structure-based methods, 0.576 against 0.826 ^[1] — a result worth remembering before you invest in a structural model to refine a charge estimate.

A cleaner experimental check comes from peptides designed for the purpose. Sixteen synthetic trimers to hexamers, each carrying one tryptophan for detection at 280 nm plus ionizable side chains, were focused in slab-gel IEF with the pH gradient read directly by electrode; their measured pI values spanned 3.38 to 10.17 and agreed with values calculated from amino acid composition to a root-mean-square difference of 0.15 pH unit ^[5]. That is close to a best case: very short chains, deliberately simple charge complements, no folded structure. Treat 0.15 as a floor on the disagreement, not a typical value.

The ceiling is higher. When ribonuclease Sa (pI 3.5) was re-engineered by replacing surface Asp and Glu residues with Lys to give variants of pI 6.4 and 10.2, the authors showed that pI values estimated using pK values based on model compound data can be in error by more than 1 pH unit ^[6]. That result is on a folded protein, where burial and tertiary contacts are available to perturb pKa in ways a short unstructured peptide has fewer opportunities to reproduce — but it sets the scale of what structure can do.

What moves an effective pKa

Three mechanisms are worth separating, because they respond differently to your buffer.

Neighbouring charge. NMR titration of all 26 sites in α-synuclein that ionize below pH 7 — an intrinsically unfolded 140-residue chain, so the closest protein analogue to a long flexible peptide — found the N-terminal domain with systematically lowered pKa values, attributed to weak electrostatic interaction between acidic and basic residues in its sequence repeats, and the acidic C-terminal domain with elevated pKa values from repulsion between like charges. The effects were smaller at physiological salt but did not vanish, and the authors concluded that the pKa values of the uncomplexed protein differ significantly from random coil model peptides even though the chain is unfolded ^[7]. Disorder does not buy you model-compound behaviour, and a run of consecutive Glu residues should not be expected to titrate like an isolated Glu.

Structure and burial. In folded RNase Sa, the buried Asp33 carboxyl, which forms three intramolecular hydrogen bonds, has a pK of 2.4, while Asp79 — also buried but not hydrogen bonded — has a pK of 7.4 ^[8]. Two aspartates in one protein, five pH units apart. For surface groups the picture was simpler: comparing the wild type with the charge-reversed 5K variant, all measured pK values in 5K were lowered, the largest shift being 2.2 pH units for Glu14, and the differences were reproduced by a Coulomb's law model with a dielectric constant of 45, with pK values calculated from that model correlating with experiment at R = 0.83. Unexpectedly, a finite-difference Poisson-Boltzmann treatment that also included desolvation and charge-dipole terms correlated more poorly with experiment (R = 0.65) ^[8]. The authors read this as evidence that charge-charge interaction is the chief perturbant for surface groups; it is also a caution that a more elaborate electrostatic model is not automatically a better one.

The interface. Charge is not a fixed property carried into every environment. In the α-synuclein work, the pKa of His50 was raised by 1.4 pH units in the SDS micelle-bound state, and titrations of variants showed the shift came primarily from interaction with the sulfate groups of the detergent rather than from a neighbouring glutamate ^[7]. A molecular theory of silica-binding peptide adsorption makes the same point structurally: it integrates charge regulation explicitly, and finds that local environmental changes near the surface modulate peptide charge, which in turn modulates adsorption; residue composition and sequence order both matter to binding affinity ^[9]. A peptide approaching a charged stationary phase, a membrane or a glass wall is not carrying the charge your calculator computed for bulk solution.

Measuring pI: what a focused band is and is not

Isoelectric focusing sets up a pH gradient in carrier ampholytes and lets the analyte migrate until its mobility falls to zero. The gradient itself is less stable than the textbook picture suggests. Reviewing the theory from Svensson's 1961 steady-state model onward, Vigh and Gas describe the currently held view of open-system IEF as a two-stage process: a rapid first phase in which a linear pH gradient develops, followed by a slow second stage in which the gradient decays as isotachophoretic processes carry the extreme-pI carrier ampholytes into the electrode vessels ^[10]. They present that two-stage model as the field's consensus rather than a settled physical description, and flag an alternative — transient bidirectional isotachophoresis — developed in companion papers. The operational consequence is the same either way: a focused band is a snapshot of a gradient that is still evolving, and focusing time is a method parameter, not a formality.

In slab-gel IEF the pH gradient can be read directly, by measuring the pH of the gel with a membrane electrode ^[5]. In a capillary the studies cited here supplied it instead by co-focusing peptide pI markers. Using a set of synthetic peptide markers spanning 3.38 to 10.17, cIEF returned pI values for nine proteins and one peptide of known pI (3.55 to 9.60) that agreed with the reference values within an average difference range of 0.04–0.08 pH unit, on 10–100 ng of sample in 30–60 minutes ^[11]. Two caveats from the same study travel with that figure. Detection time plotted against pH was not linear in most cases, so fitting a single calibration line across the whole gradient is erroneous; the authors instead interpolated linearly between the two flanking marker peaks. And because focusing may not be complete, they recommend running two different focusing times and confirming the answer does not move ^[11]. Carrier ampholytes from two suppliers and in different pH ranges were compared, and some were preferentially more effective at either the acidic or the basic side of the gradient, so the ampholyte choice is part of the method.

Why the markers focus sharply is itself instructive. A set of nineteen fluorescent pI markers covering pH 3.64–10.12 was built from tetra- to tridecapeptides designed to include amino acids whose charged side-chain pKa values lie close to the corresponding pI ^[12]. Placing a titrating group near the focusing pH makes mobility fall steeply as the band approaches its pI, which is what produces a narrow zone. A peptide whose ionizable groups all sit far from its pI titrates shallowly through that region and focuses into a broader band — the same pI, a worse measurement.

Resolution can be very fine when the gradient is well behaved. An in-line cIEF-ESI interface that desalts ampholytes before the electrospray achieved baseline resolution of peptides from a tryptic digest of bovine serum albumin differing by as little as 0.02 pI unit, with an average full width at half maximum of 7.1 s; separately, the interface preserved pH gradient linearity with R² of 0.99 and, in proof-of-principle experiments, gave more than a 90% increase in electropherogram area under the curve relative to running without it ^[13].

What a focused band licenses is narrow: it gives the pI of that species in that ampholyte mixture, at that ionic strength, temperature and focusing time, read against those markers. It does not give you the net charge of the peptide in your assay buffer, and it does not certify that the band is a single species — anything that co-focuses is indistinguishable by position alone.

Charge at working pH is the quantity you actually use

The pI is one point on a titration curve. Every separation and ionization decision depends on the ordinate somewhere else on that curve.

Capillary electrophoresis. Fitting mobility data for 21 peptides of 350–1850 Da measured at varying acidic pH, Castagnola and colleagues extracted the dissociation constants of the C-terminus and of the Asp and Glu side chains alongside the Stokes radius at each protonation stage. The radius scaled with molecular mass to a fractional exponent of 0.68 at pH 2.25 — close to the M^2/3 proportionality Offord proposed — falling to 0.58 at pH 4.25, which they interpret as a mean conformational transition toward more compact structures following C-terminal dissociation. Having each peptide's own constants let them quantify the error incurred by predicting charge from mean dissociation constants; charge prediction was best at the most acidic edge of the range studied, mainly at pH 2.25, and they placed the best performance of their mobility model in the very acidic range, pH 2.25–2.50, and in the absence of electroosmotic flow, and their recommended expression for mobility was μ = 85.4(Z/M^0.68) × 10⁻⁸ m² V⁻¹ s^{−1 [14]}. The reason carries over to any pH you might work at: computed charge is most trustworthy where few groups sit near their pKa, and least trustworthy in the pH regions where a titration is half complete.

Ion exchange. Net charge sets the first-order retention order in cation exchange, but not the whole of it. Using three synthesized series of positively charged peptide polymers spanning 5–50 residues and differing in hydrophobicity — Ac-(Gly-Lys-Gly-Leu-Gly)_n-amide, Ac-(Leu-Gly-Leu-Lys-Ala)_n-amide and Ac-(Leu-Gly-Leu-Lys-Leu)_n-amide — Burke and colleagues showed that non-specific hydrophobic interaction with both silica- and polymer-based ion-exchange packings contributes to retention, that adding acetonitrile to the mobile phase suppresses it, and that removing it may be necessary simply to elute peptides from the matrix at all ^[15]. A peptide that elutes later than its charge predicts is not necessarily more charged than you thought.

Electrospray. The naive rule is that a peptide picks up one proton per basic site. Machine-learning analysis of charge state distributions for hundreds of thousands of peptides found that about half depart from that expectation, splitting into an undercharging and an overcharging regime with distinct sequence determinants; peptides in the overcharging regime showed minimal dependence on basic site count ^[16]. Use basic-site count to anticipate the charge envelope, but expect the observed envelope to be data, not confirmation.

The window around zero charge

Near the pI, intermolecular electrostatic repulsion is at its weakest, and the practical consequence is a change in what the peptide does with itself. For the model ionic-complementary peptide RADA16-I, a net charge of +0.75 was found to be the ideal condition for forming regular amyloid fibrils; lower net charges favoured amorphous precipitates instead, while larger net charges destabilized fibrillar aggregates and promoted reversible dissociation of monomers from fibril ends. Adding either salt or guanidinium hydrochloride at moderate concentration, 0.3–1 M, shifted the monomer-fibril equilibrium toward the fibrillar state; the salt effect was explained by charge screening of electrostatic repulsion alone, while the guanidinium effect was also attributed to changes in peptide conformation ^[17]. Two things follow. Charge screening does part of what a pH change does, so ionic strength and pH have to be reported together. And the interesting regime is not a single point but a narrow band of low net charge in which the aggregate morphology, not just the rate, changes.

The standard advice to keep working pH away from the pI is sound but often over-extended. In the RNase Sa series, the pH of minimum solubility tracked the pI across variants of pI 3.5, 6.4 and 10.2 — but the pH of maximum activity and the pH of maximum conformational stability did not ^[6]. Moving off the pI buys solubility; in that series it did not move the activity or stability optima with it. That work is on a protein, so read it as a warning against conflating the three optima rather than as a number to transfer to a peptide.

What to record

A useful pI entry in a notebook is not a number. It is a number, the pKa set that produced it, and the spread that set implies — a few tenths of a unit for a short unstructured peptide ^{[4, 5]}, wider for anything long, highly charged, structured or modified. If the value matters, measure it, and record the ampholyte range and lot, the ionic strength, the temperature, the marker set and the focusing time alongside it ^[11]. Then convert to the quantity you will actually reason with: the net charge at the pH you are running, with the acknowledgement that this too is computed from constants borrowed from other molecules, and that the borrowing is worst where a group is close to titrating.