Peak purity in peptide HPLC: the limits of spectral homogeneity

Every chromatographic purity figure rests on an assumption that the chromatogram itself cannot verify: that the main peak contains one compound. Area percent is arithmetic performed on integrated peaks, and it is indifferent to whether a peak holds one species or three that happen to leave the column together. The conventional check on that assumption is diode-array peak purity — an automated comparison of ultraviolet spectra taken across the peak, returning a pass or fail. It appears in method validation packages, in forced degradation reports, and occasionally as a line on a certificate of analysis.

For small molecules with distinctive chromophores, the check carries real weight. For peptides it carries much less, and the reason is structural rather than instrumental. Understanding where the inference breaks down changes what a stated purity figure can reasonably be read to mean.

What the algorithm computes

All commercial peak purity algorithms work from the same raw material: a diode-array detector records a full spectrum at every data point across the peak rather than absorbance at a single wavelength, producing a matrix of absorbance against wavelength and time. If the peak contains one absorbing species, every spectrum within it should be the same shape, differing only in magnitude as concentration rises and falls. If a second absorbing species is present and its concentration profile does not exactly track the first, the shape will change somewhere across the peak.

The implementations differ in how they express that comparison. Spectral contrast approaches normalise each spectrum to a unit vector and regard it as a direction in multidimensional space; the angle between a given spectrum and a reference spectrum — usually taken at the apex, on the assumption that the apex is least contaminated — becomes the purity angle. Against it the software computes a purity threshold: the angular deviation attributable to detector noise and solvent background alone. A peak is reported spectrally homogeneous when the purity angle falls below the purity threshold. Other vendors report a similarity factor on a 0–1000 scale, with values at or above roughly 990 conventionally read as consistent with a single component, and plot the comparison across the peak against a noise-derived threshold curve.

Both formulations share a property worth stating plainly. The threshold is not a fixed property of the compound; it is derived from the noise in that particular injection. A peak run at low absorbance produces a high threshold, which makes a pass easier to obtain and less meaningful — the same sample at higher loading can fail a test it previously passed, with nothing about the material having changed.

The evidence is asymmetric

Peak purity analysis is designed to reveal impurities, not to prove their absence, and the asymmetry is structural rather than a matter of software quality. A failure is informative: something in the peak has a spectrum that varies across it beyond what noise explains, and that requires an explanation. A pass is much weaker, because several distinct situations produce one.

An impurity that does not absorb ultraviolet light at the monitored wavelengths is invisible entirely — counterions, residual salts, and several classes of process residue fall here, part of why peptide content and chromatographic purity are separate numbers. An impurity present below the concentration at which its spectral contribution exceeds the noise threshold is equally invisible, and that floor is often well above the impurity levels that matter at the identification and qualification thresholds.

The most complete blind spot is exact co-elution. The algorithm detects a change in spectral shape across the peak; that change exists only because two species have different concentration profiles in time. Two compounds that co-elute with genuinely identical retention and identical peak shape produce a composite spectrum that is constant across the entire peak. The mixture is uniform, so the peak is spectrally homogeneous, and the software reports a pass on a peak containing two compounds. Chromatographic conditions engineered for a clean, symmetric main peak make this case more likely rather than less.

And finally, an impurity whose ultraviolet spectrum closely resembles the main component’s produces so small an angular deviation that it disappears into the threshold. Compounds that co-elute are usually chemically related, and chemically related compounds usually share a chromophore. This is the general case for peptides.

One chromophore, many species

Peptide chromatography is typically monitored between 210 and 230 nm, most often at 214 nm, because that region is where the amide bond of the peptide backbone absorbs. The choice is deliberate: it makes detection close to universal across sequences, with response scaling roughly with the number of peptide bonds rather than with the presence of any particular residue. That universality is exactly what defeats spectral discrimination.

Consider what the relevant impurities in a synthetic peptide preparation actually are. Deletion and truncation sequences arising from incomplete coupling differ from the target by one or more residues but are otherwise the same class of molecule, built from the same backbone. A deamidated variant differs by one amide converted to a carboxylate. An epimer differs only in stereochemistry at a single centre. An oxidised methionine differs by one oxygen on a side chain that does not absorb meaningfully in the monitored window. In every case the dominant chromophore — the backbone amide — is essentially unchanged, and the ultraviolet spectra of parent and impurity are near-superimposable. The purity angle stays small because there is almost nothing for it to measure.

The only structural features that give a peptide spectrum any distinguishing shape are the aromatic side chains: tryptophan and tyrosine, which absorb near 280 nm with molar extinction coefficients on the order of 5,500 and 1,490 M⁻¹cm⁻¹ respectively, and phenylalanine, which contributes weakly. Where the sequence contains them, a second detection wavelength becomes available and the ratio of absorbance at 280 nm to absorbance at 214 nm can itself act as a crude homogeneity indicator — a principle the A280 quantitation methods rest on. But many research peptides contain no tryptophan and no tyrosine at all, and for those the spectrum above 250 nm is essentially flat. Where an impurity differs from the parent at an aromatic residue, the ratio can move; where it differs anywhere else, it will not.

The practical consequence is that a peak purity pass on a peptide method is compatible with the presence of deletion and truncation sequences, deamidation products, and D-amino acid epimers at levels that would materially change the reported purity if they were resolved. The test has not failed; it was never sensitive to those species in the first place.

Where false results come from

The failure mode runs in both directions, and a reported failure is not automatically evidence of co-elution. Low signal-to-noise is the most common cause of a spurious fail, because the algorithm is comparing small numbers and noise dominates the comparison at the leading and trailing edges of any peak, where absorbance approaches baseline. Most software allows those regions to be excluded, which improves consistency and also quietly discards the part of the peak where an early- or late-eluting impurity would most likely show itself.

Gradient elution introduces a second artefact. The mobile phase composition changes across the peak, and if the background absorbance of the solvent changes with it — which it does, particularly with trifluoroacetic acid as the ion-pairing modifier and particularly on steep gradients — the measured spectra shift for reasons that have nothing to do with the analyte. Adequate background correction and a shallower gradient both reduce this, and the relationship between gradient slope and apparent resolution is one of several method variables that move a peptide purity figure without any change in the material. Detector saturation, poor integration limits, and baseline drift produce the same class of artefact. A peak purity result is therefore only interpretable alongside the system suitability data for the run that produced it.

What counts as stronger evidence

ICH Q2(R2), adopted in December 2023 alongside Q14, is explicit that specificity for peptides, proteins, and oligonucleotides is a harder problem than for small molecules, and directs that a second, orthogonal procedure be used where a single separation cannot establish it — ideally a genuinely different separation mechanism rather than a variation on the same one. The guideline also recognises technology-inherent specificity, where the detection principle itself supplies selectivity without a separate experiment, as with mass spectrometry.

In practice the useful orthogonal axes for a peptide are well defined. Changing mobile phase pH alters the ionisation state of acidic and basic side chains and reorders the elution of charge variants. Ion-exchange chromatography separates on charge rather than hydrophobicity and resolves deamidation products that reversed-phase methods often co-elute. Capillary electrophoresis separates on charge-to-size ratio in free solution, with no stationary phase at all. Hydrophilic interaction chromatography inverts the retention mechanism.

Mass spectrometric detection is the strongest single addition, since it distinguishes species that ultraviolet spectra cannot — but it carries its own blind spot that matters for exactly this class of impurity. Isomers share an exact mass. An epimer, an isoaspartate rearrangement product, and a cis/trans proline conformer are all indistinguishable from the parent by mass alone, which is why multiple heart-cutting two-dimensional liquid chromatography has been developed specifically for peptide peak purity work. That technique has its own difficulty: peptide retention responds sharply to small changes in mobile phase composition and temperature, so first-dimension peaks wander between runs, and the shallow gradients needed to resolve peptide impurities make the wandering worse.

Forced degradation remains the most accessible check available to a laboratory without a second instrument. Deliberately generating the degradants, then demonstrating that the method resolves them from the parent, establishes something a peak purity pass on an unstressed sample never can.

Reading a purity claim accordingly

None of this makes peak purity analysis worthless. It is cheap, it runs on data the detector already collected, and when it fails it has found something real. What it cannot do is convert an area percentage into a statement about molecular composition — and for peptides that gap is wider than for almost any other analyte class, because the backbone chromophore that makes detection universal also makes discrimination impossible.

A purity figure is therefore best read as indexed to a specific method, not as a property of the material — a point the wider treatment of what a 99% figure means develops further. The questions that separate a well-characterised analytical record from a bare number are concrete ones: at what wavelength was the peak monitored, was a second orthogonal technique run, was the method challenged against deliberately generated degradants, and was mass spectrometric confirmation obtained. A laboratory that can answer them is making a materially different claim than one reporting a single number and a peak purity pass.