Peak Purity: What a Diode Array Detector Can and Cannot Prove

A chromatographic peak looks like one thing. Spectral peak purity is the method that tests whether it is, using data the detector was already collecting.

What a diode-array detector records

A single-wavelength detector measures absorbance at one wavelength as the peak passes. A diode-array detector disperses the light across an array and measures many wavelengths at once, so it records a full ultraviolet spectrum many times per second across the width of the peak.

That turns a peak from a curve into a stack of spectra. The extra dimension is what the analysis uses.

The principle

A pure peak is one compound at every point across it. One compound has one absorbance spectrum, whose shape does not change with concentration — only its height does.

So if the spectra collected at the front, apex and tail of a peak have the same shape after normalising for concentration, the peak is consistent with a single compound. If the shape changes across the peak, something else is present and its proportion varies across the elution window.

How the software reports it

Implementations differ, and the numbers are not comparable between vendors. The common approaches are:

• A purity angle against a threshold angle. Spectra are treated as vectors and the angle between them measured. The software also computes a threshold representing the angle expected from noise alone. Purity angle below threshold means the variation is indistinguishable from noise.
• A similarity or match factor. Usually scaled so that 1000 or 100 is a perfect match, with a stated pass value.
• A purity plot. The deviation is plotted across the peak, so a reader can see where in the peak the problem occurs, which is more informative than a single number.

Why it fails on the case you most want it to catch

The test detects a difference in spectra. Two peptides that differ by a deletion of a non-aromatic residue, or by deamidation, or by D-to-L inversion, have essentially identical ultraviolet spectra, because their chromophores are the same.

A peak containing both will pass a spectral purity test cleanly. This is the central limitation: the impurities most likely to co-elute with a peptide are often the ones most likely to share its spectrum.

The other conditions it needs

• Enough signal. At low absorbance the spectra are noisy and the threshold angle rises until everything passes. A pass on a small peak may mean nothing.
• Not overloaded. At high absorbance the detector response goes non-linear, spectra distort at the apex, and a pure peak can fail.
• A stable baseline. Solvent absorbance changing across a gradient contributes its own spectral variation, which is why background correction matters.
• A co-eluting species that actually absorbs. Something with no chromophore is invisible to the test entirely.

What a pass and a fail each mean

A fail is strong evidence: the peak contains more than one species, and that conclusion is reliable when the signal and baseline conditions are met.

A pass is weak evidence. It means no spectrally distinguishable second component was detected, under those conditions, at that signal level. It is not a demonstration that the peak is one compound, and a certificate that presents it as one is overstating what the method does.

What actually settles co-elution

An orthogonal separation — a different column chemistry, a different pH, or hydrophilic interaction chromatography instead of reversed phase — sorts on a different property and separates pairs the first method merged.

Mass spectrometric detection across the peak is stronger still, because it distinguishes species by mass rather than by spectrum and therefore catches exactly the deletion and truncation impurities that defeat the spectral test. The general problem is set out in co-elution and peptide purity.

Reading it on a document

Where a peak purity result appears, the useful form names the software’s metric, gives the value and the threshold, and ideally shows the purity plot. A bare statement that the peak was pure, with no metric and no threshold, records that the software was run rather than what it found. Related reading: reading an HPLC chromatogram.