A purity figure is a statement about a chromatogram. Whether it is also a statement about the material depends on one thing: whether anything else in the sample reached the detector at the same time as the peptide. A single gradient cannot answer that question about itself. An impurity that co-elutes does not appear as a small peak — it is absorbed into the main peak's area and raises the reported number rather than lowering it. The evidence has to come from outside the run, from a second separation that orders the same molecules by a different rule. That is what orthogonal means here, and it is a narrower, more conditional claim than the word usually carries on a document.
The impurities that make this hard
A synthetic peptide's related impurities occupy two extremes at once. One group — amino acid substitutions, chain cleavages — spans a broad range of properties. Another, such as d-/l-isomers, sits so close to the target that similarity rather than diversity is the problem; the authors present both extremes together as what makes the analysis challenging, and a separation tuned to catch the first group is not necessarily tuned for the second [2].
Mass spectrometry does not close the gap. Compounds with the same mass-to-charge ratio are not readily differentiated by mass spectrometry and must therefore be separated chromatographically [1]. Epimers, positional isomers and sequence-scrambled variants are such compounds, so the burden falls back on the separation.
Nor is a second reversed-phase method automatically a second opinion. Characterizing peptide separations with peptide probes, Field and colleagues concluded that the dominating forces appear to be hydrophobic as well as electrostatic and polar in nature, and that it appears to be the accessibility differences between phases — to silanols, to the hydrophobic acetonitrile–ligand layer, or to a thin adsorbed water layer on the silica surface — which play a crucial role [4]. Two C18 methods running an acetonitrile gradient draw on that same set of interactions, so — this is a consequence of the mechanism described, not a comparison the authors ran — there is little reason to expect one to resolve an impurity the other buries.
Orthogonality is a property of a pair, not a badge on a method
No single method is orthogonal. Orthogonality describes how differently two systems order the same analytes, so it can only be evaluated for a specific pairing against a specific sample set. It is also quantifiable, and the quantification is not settled. Gilar and colleagues compared correlation coefficients, mutual information, box-counting dimensionality and surface fractional coverage across six peptide elution data sets acquired on different stationary phases, and found correlation coefficients less suited to the purpose than the alternatives; box-counting dimensionality and fractional coverage turned out to be mathematically related [11]. It follows — as a consequence of that comparison rather than a claim the authors make — that a reported orthogonality value is only interpretable alongside the metric that produced it.
The magnitudes are instructive. Yeung and colleagues evaluated 16 two-dimensional pairings using roughly 30,000 tryptic peptides per pairing, and reported that orthogonality generally increases in the order reversed-phase < strong cation exchange < hydrophilic interaction < strong anion exchange — with one exception: a high-pH reversed-phase first dimension paired with a low-pH reversed-phase second dimension gave the second-best value at 68%, six percentage points behind the best pairing, a strong anion exchange column at 74% [10]. Changing only the pH within reversed-phase chemistry bought most of what a wholly different retention mode bought. And the best pairing scored 74%, not 100% — though these systems were assessed for retention-space coverage in proteomic fractionation, not for the fraction of a peptide's impurities they would resolve, so the number should not be read as the latter.
One negative case is worth stating plainly, and it is not a literature finding: two methods differing only in gradient slope, run time, column length or flow rate are not orthogonal in any useful sense. They retain by the same mechanism, so they order the same molecules the same way. A longer gradient may resolve more, but it does not test a different hypothesis.
What changing the mobile phase pH buys
pH is the cheapest genuine change available, because it acts on the analyte rather than the column. Field and colleagues assessed 51 reversed-phase mobile phase compositions spanning pH 1.8 to 7.8, using selectivity differences between nine synthetic peptides on a C18 phase as the readout. Principal component analysis sorted them into groups of distinct selectivity, and the authors concluded that screening numerous mobile phases of differing pH, ion-pair reagent and ionic strength is important to maximize the probability of separating all peptides of interest in a complex mixture; high ionic strength was crucial for symmetrical peaks. They stated the scope carefully. The C18 observations were compared on three additional phases — alkyl amide, fluorophenyl and biphenyl — over a limited subset of mobile phases; with the exception of the ion-pair reagent, similar trends were obtained for the C18, fluorophenyl and biphenyl phases, which the authors take to intimate applicability to the vast majority of reversed-phase columns neutral or weakly polar in character. Their conclusions were not relevant for columns of a more disparate nature, such as those carrying a high degree of positive charge [6]. That exception is worth carrying into the next section: the ion-pair reagent is the one parameter whose trends the authors did not report as holding across those phases.
For peptides with secondary structure, pH does more than reset ionization. Chen and colleagues compared amphipathic α-helical peptides at pH 2.0 and pH 7.0 and reported that at pH 2.0, raising the trifluoroacetic acid concentration or adding sodium perchlorate to a phosphoric acid mobile phase both improved peak shape and increased retention through ion-pairing with the positively charged peptides, whereas at pH 7.0 added salt had little effect beyond improved peak shape. Temperature had a complex influence on selectivity, which the authors attributed to varying effects on peptide conformation [8]. A pH change can thus alter charge state and folding at once — which is why it generates real selectivity differences, and why its effects are hard to predict.
What changing the ion-pairing agent buys
Trifluoroacetic acid is the default additive — the concentration at which it is used is a matter of convention rather than of any measurement reported here — and there is evidence that the default suppresses the very differences an orthogonal check looks for. Assessing 30 column and mobile phase combinations, Petersson and colleagues found that for the peptides investigated, trifluoroacetic acid reduced selectivity differences between columns of diverse properties, which they suggest is probably due to its potential to form ion-pairs with peptides. The same dataset supplies the counterweight: trifluoroacetic acid still received the highest ranking for 40% of the column and mobile phase combinations, on better selectivity and/or peak shape [1].
Alternatives have been evaluated. In a screening protocol using five disparate columns and six mobile phases spanning pH 2.3 to 5.1, ammonium hexafluorophosphate at pH 2.3 was reported to produce excellent complementary separation and good peak shape, and methanesulfonic acid was demonstrated to be a good alternative to trifluoroacetic acid, which failed to generate optimum separation for the peptides investigated. Neither novel additive adversely affected C18 column stability in that work, though the authors recommend both be used with caution until long-term routine use has been established [7].
The underlying mechanism appears more modest than the convention implies. Dai and colleagues measured ion pair formation constants by capillary electrophoresis and found the order hexafluorophosphate > perchlorate > trifluoroacetate > chloride, but reported that the extent of ion pairing is not large: at a typical additive concentration of 20 mM, roughly 15%, 6% and 3% of the analytes were present as ion pairs with hexafluorophosphate, perchlorate and trifluoroacetate respectively [9]. Those measurements were made on prototypical basic drugs, not peptides, and the quantities should not be transferred. For those drugs, at least, the picture is of an additive that acts through an interaction involving only a minority of the analyte at any moment; whether the same proportions hold for peptides was not examined in that work.
What changing the stationary phase buys, and one trap
Column choice offers the widest selectivity range and the largest opportunity to fool yourself. Field and colleagues applied a peptide-specific characterization protocol to 38 stationary phases varying in ligand chemistry, base silica, end capping and pore size, using principal component analysis to identify columns of dissimilar selectivity as method-development candidates. The trap is their central negative result: the small molecule column databases, specifically Snyder's Hydrophobic Subtraction Model and the extended Tanaka protocol, failed to correlate with the peptide protocol [5]. A column pair selected as "very different" on small-molecule grounds is not, on that evidence, established as different for peptides. The same work found positively charged stationary-phase groups advantageous, yielding very different selectivity and improved peak shape.
Leaving reversed-phase altogether is the strongest available move. Yoshida and colleagues analyzed the impurities of four cyclic peptides — each nine residues, with neutral, acidic, basic or zwitterionic residues at positions 3 and 6 — across three polymer-based hydrophilic-interaction columns and seven mobile phases. Against reversed-phase profiles on an octadecylsilyl silica column, the two techniques proved mutually orthogonal: most impurities with small reversed-phase retention times had large hydrophilic-interaction retention times, and vice versa [3]. That inversion is what makes the pairing informative. How far it extends beyond this class of nine-residue cyclic peptides was not tested.
Two separate runs are not the same as a second dimension
Here is where the reasoning usually goes wrong. Suppose a peptide gives 98.2 area% by method A and 96.4 area% by an orthogonal method B. These numbers are hypothetical; the arithmetic below is worked, not measured.
Method A resolved 100 − 98.2 = 1.8% impurity. Method B resolved 100 − 96.4 = 3.6%, exactly twice as much. The main peak figure fell by 1.8 percentage points, a relative decrease of 1.8 / 98.2 = 1.83%.
The wrong conclusion is that the purity is 96.4%. Both figures are upper bounds, because co-elution can only inflate a main peak. What two separate runs establish is a range. If every impurity A resolved is also among those B resolved, the union is 3.6% and purity is at most 96.4%. If the two sets are entirely disjoint, the union is 1.8 + 3.6 = 5.4% and purity is at most 100 − 5.4 = 94.6%. The upper bound therefore lies somewhere from 94.6% to 96.4%, and no amount of one-dimensional data narrows it, because nothing in two independent chromatograms tells you which peaks in one correspond to which in the other. The bracket is also in area-percent units, and inherits every assumption about equal detector response that area percent carries.
What a second dimension actually demonstrates
Two-dimensional liquid chromatography answers a more specific question, because it physically transfers the first-dimension main peak into the second column. The logic is stated directly by Stoll and colleagues: the first dimension can be used to detect impurities over a broad range in properties, while the second can focus specifically on those species that might co-elute with the target peptide in the first [2]. The result is a claim about that peak, rather than a comparison of two peak counts.
The cost is method development. Obtaining sufficient peptide isomer retention and separation in the second dimension requires shallow gradients, well under 1% acetonitrile per minute [1]. Stoll and colleagues describe a two-step procedure — two scouting gradients, then a third separation used to build and refine a retention model — that places the target in the middle of the second-dimension chromatogram and increases the likelihood of resolving peptides with very similar properties; they demonstrated it on four model peptides and a degraded model peptide sample [2]. The companion study recommends, for that second dimension, a C8 or C18 column possessing no ionic functionality with an acetic acid / ammonium acetate mobile phase buffered at pH 5, reporting good selectivity at 25 °C for peptide isomers of molecular weight below 10 kDa [1].
What such an experiment demonstrates should be stated precisely: under those second-dimension conditions, no impurity above the detection limit was resolved from the target within the fraction transferred. It does not show that the peak is pure. A species that co-elutes in both dimensions remains invisible; the probability of that is lower for a well-chosen pair than for a single run, but it is not zero, and no orthogonality metric quantifies it.
Dimensions a chromatogram of the intact peptide will not supply
Two classes of information sit outside any purity number derived from separating the intact molecule. The first is stereochemistry at the residue level. Strege and colleagues describe a chiral liquid chromatography–tandem mass spectrometry method in which the peptide is hydrolyzed in deuterated acid — the label allowing correction for racemization introduced by hydrolysis itself — after which the free amino acids are separated on a chiral column and quantified, demonstrating recovery of d-amino acid substitutions at each residue of an octapeptide across the 0.1 to 1.0% range [13]. The measurement destroys the peptide to make itself possible, which is why it is orthogonal to every chromatographic purity figure.
The second is mass that never elutes or never absorbs. Melanson and colleagues combined three orthogonal approaches — quantitative NMR on the intact peptide, amino acid analysis by isotope dilution tandem mass spectrometry after hydrolysis, and a mass balance calculation — to assign purity to a candidate angiotensin II certified reference material, noting that both of the first two are susceptible to interference from related peptide impurities that must be measured and accounted for. The mass balance relied on a validated fluorine-19 qNMR method to quantify the trifluoroacetic acid counter-ion, present at nearly 25% by mass. The three methods agreed, and the final assignment was 691 ± 9 mg/g [12] — less than 70% of that material's mass was the peptide, in a sample characterized to reference-standard rigor. A chromatographic area percent has nothing to say about that quantity.
What is convention, and what to ask
Much of this is convention with thin primary literature behind it. That a certificate reports one method, and rarely names a second, is a documentary habit rather than a finding. So is quoting purity to a single decimal place, which implies a precision the method has not been shown to have. Injecting a mobile-phase blank to establish that a small peak comes from the sample rather than the reagents is routine bench practice, not a published result. And the rule that the lower of two orthogonal figures is the more conservative one follows from the direction in which co-elution biases area percent; it is a logical consequence, not a measurement.
What is genuinely unresolved: there is no established threshold value of any orthogonality metric above which a pair of methods can be called sufficient, and the available metrics are not interchangeable — Gilar and colleagues found correlation coefficients less suited to the purpose than the alternatives, while two of those alternatives, box-counting dimensionality and fractional coverage, turned out to be mathematically related to each other [11]. Peptide-specific column characterization exists [4, 5] but does not predict, for a particular sequence and its particular impurities, which pairing will expose a co-eluting species. And no published framework converts an orthogonality score into a probability that an impurity escaped both dimensions.
So the questions worth putting to a purity figure are narrow. Which two systems produced it, and do they differ in retention mechanism or only in gradient? Was the second method's peak count merely compared with the first, or was the main peak itself transferred into a second column? A purity number without its two methods attached is a measurement with its most important variable left out.