A peptide with no tryptophan and no tyrosine gives you almost nothing at 280 nm. The arithmetic is plain: for folded proteins in water, the molar absorption coefficient at 280 nm is predicted by ε280 (M-1 cm-1) = (#Trp)(5,500) + (#Tyr)(1,490) + (#cystine)(125), and the authors of that relation record that the predicted values are quite reliable for proteins containing Trp residues and less reliable for proteins that do not [1]. Strike the two aromatics and the only term left is 125 per cystine.
Dropping to 214 nm recovers a signal, because the backbone absorbs, but it does not recover proportionality to mass. Measured in the presence of acetonitrile and formic acid, the peptide bond has a molar extinction coefficient of 923 M-1 cm-1 at 214 nm, tryptophan is approximately thirty times higher, and phenylalanine, tyrosine and histidine approximately six times higher [2]. An area percent at 214 nm is therefore a share of absorbance weighted by composition — the problem the library's guide to what an HPLC area percent measures is built around. An impurity differing from the main peak by one aromatic residue is misrepresented in that percentage in a direction set by which residue it lost, and a co-eluting species with no chromophore contributes nothing at all.
What a nebulisation detector measures instead
Charged aerosol detection (CAD) and evaporative light scattering detection (ELSD) replace the optical question with a physical one. Column effluent is nebulised, the mobile phase is evaporated, and whatever was non-volatile survives as a particle; ELSD measures the light that particle stream scatters, while CAD is based on aerosol charging. In the review that set the two side by side, both are described as mass dependent, with a response that does not depend on the spectral or physicochemical properties of the analyte — the property that makes calibration against a single universal standard conceivable, so that, in the review's own wording, all other compounds or impurities can be qualified against it. The same review credits CAD with better sensitivity than ELSD and a dynamic range of up to four orders of magnitude, and records that CAD carries ELSD's central limitation, that response is affected by mobile-phase composition, addressed by inverse gradient compensation [3]. That is the review's summary; the measured bounds on each half follow.
Quasi-universal is the accurate word
A flow-injection survey of 29 solutes spanning acids, bases and neutrals put a figure on the uniformity. CAD response was characterised there as quasi-universal, with strong signals for compounds of low volatility at typical room-temperature operation; for relatively involatile solutes the response was reasonably independent of solute chemistry, varying by 12–18% RSD across mobile phases from buffered 95% acetonitrile (HILIC) to 10% acetonitrile (reversed phase), with somewhat higher response for basic than for neutral solutes [4].
Closer to peptide reversed-phase conditions, the same uniformity is reported, though qualitatively rather than as an RSD. Flow injection of six proteins of different molecular weights and different structures gave near-uniform CAD responses, and similar responses were observed for all six when the mobile phase composition included trifluoroacetic acid and acetonitrile — the conditions regularly applied to the separation of proteins by reversed-phase chromatography [5].
The most direct test of the no-standard claim benchmarked CAD against quantitative 1H NMR. Across 50 compounds with a wide range of physical and chemical properties, quantification using a generic calibrant and mass calibration gave an average error of 11.4% relative to NMR. Correction factors estimating the relative surface area of the particles in the detector, taking account of analyte density and charge, and calibrating on surface area rather than mass, reduced that average error to 7.1%; the improvement was largest for highly dense compounds, where mass calibration averaged 34.7% error against 5.8% for surface-area calibration [6]. Those were small molecules rather than peptides, but they fix the order of magnitude of the residual: a residual quantification error of roughly ten percent against NMR, not of zero.
The response is a power law and has to be calibrated as one
Signal is not proportional to injected mass in either detector, so a straight line through peak area against concentration is the wrong model over any wide range. One study varied evaporation temperature, filter constant and power function value systematically in a UHPLC-CAD analysis of fatty acid standards and of manufacturing batches of polysorbate 80, all of whose components lack a chromophore. Evaporation temperature and filter constant strongly affected detection limits expressed as signal-to-noise ratios; evaporation temperature could be raised to improve signal-to-noise, but analyte volatility was an important limiting factor. The power function value proved a strong tool for optimising response linearity, yet its optimum differed depending on analyte volatility. Because double-logarithmic transformation at the default power function value of 1.0 yielded satisfying universal results over a range of two orders of magnitude for every homologous fatty acid from C14 to C18, with less measurement time, log-log transformation was the favoured linearisation strategy [7].
The alternative is to let the instrument reshape the curve. The power function value is a multiplication factor applied to the power-law exponent of the equation describing CAD response, altering the signal output to optimise the quasi-linear range of the response curve. Applied to HPLC-UV-CAD impurity analysis of gabapentin, quantitation at the optimal power function value improved accuracy for the majority of the analytes to 90–105% recovery, against 81–115% for quantitation by single-point calibration or linear regression; the final method used a power function value of 1.30 and was validated with respect to ICH Q2(R1) [8].
Either route works; what matters is that the choice is fixed in the method and reported with the number, because areas produced at different power function values are not comparable.
Where the peak elutes changes its response
Because the signal is manufactured by nebulisation and evaporation, the composition of the eluent at the instant of nebulisation is part of the measurement, and under gradient elution that composition differs for every peak. The correction is to remove the variable. A second pump delivers an exactly inverse gradient relative to the analytical system, and both flows are mixed in a tee piece before introduction to the detector, so the detector sees a constant composition throughout the run. The detector is characterised by a nearly universal response at a given, constant mobile-phase composition for sufficiently non-volatile analytes, and the compensation extended that universal response from isocratic to gradient elution. The compensated data showed precision ranging from 4% RSD at 10 mg/L to 1.6% RSD at 780 mg/L, an average limit of detection with a 2 µL injection of 0.5 mg/L corresponding to 1 ng injected on column, and measurement of an analyte at a relative concentration of 0.05% versus the main component [9]. The test set was six sulfonamide drugs; the cost is a second pump, a mixing tee and the dilution from combining two streams.
The same compensation applied to ELSD made the uncorrected error explicit. Without compensation, ELSD response factors for six sulfonamide drugs differed by a factor of two when the mobile phase composition was varied between 10 and 90% acetonitrile. Compensation effectively eliminated that change, giving a nearly constant, albeit somewhat reduced, response, under which the ELSD behaved as a concentration-sensitive detector; analysis of the sulfonamides at the 0.05% level by gradient UPLC-ELSD with compensation was demonstrated [10].
The composition dependence is not an artefact of model compounds. In an ion-pair method for 19 underivatised amino acids and taurine on a porous graphitic carbon column, the response of the corona CAD to the amino acids was highly dependent on the eluate composition, whereas response factors were similar for amino acids in eluate with similar compositions; plots of injection amount against peak area regressed with r2 > 0.998 over roughly 0.1–1.6 nmol, the exception being arginine, which co-eluted with a background peak [11]. In the flow-injection survey, CAD response generally increased as acetonitrile rose from 10% to the 80–95% typical of HILIC, with signal-to-noise ratios about four times higher in the HILIC range than in the reversed-phase range [4].
Composition is one variable among several. A systematic comparison of four aerosol detectors under UHPLC found that mobile-phase flow rate, analyte concentration, mobile-phase composition, nebuliser and evaporator temperature, evaporator gas flow rate and signal filtering after detection all exerted non-linear effects on the responses, and must therefore be considered when designing separation conditions, particularly when gradient elution is performed [12].
The eluent has to be volatile, and its counterions are analytes
Everything non-volatile that reaches the nebuliser is measured, including what you did not intend to measure. A CAD method on a polymeric zwitterionic column was qualified for the simultaneous analysis of inorganic ions as salts and counter-ions, validated for four anions and two cations and extended to nine anions and three cations, and presented as a viable alternative to traditional ion analysis at parts-per-billion level [13]. A phosphate or sulfate buffer in the mobile phase is therefore not a background: it is the largest analyte in the run. Volatile additives are the only workable choice — the amino acid method above used nonafluoropentanoic acid as a volatile ion-pair reagent [11], and the protein work used trifluoroacetic acid with acetonitrile [5].
Counterions arriving with the sample are the same problem in a different place, and can be turned into a measurement. A mixed-mode HPLC method with evaporative light scattering detection quantified 14 positively and negatively charged counter ions simultaneously within 30 minutes, validated for specificity, accuracy, precision, limit of quantification, sample stability and carry-over as proposed by ICH; trifluoroacetic acid in synthetic peptides was the case of particular interest. Applied to small-molecule reference substances and to synthetic peptides, it revealed differences from the expected concentrations of the prepared peptide solutions and an unexpectedly high content of sodium in the synthetic peptides [14]. Read the other way round, that result says a relative area percent from a nebulisation detector is a percent of everything non-volatile in the injection, salts included, unless the salt peaks are chromatographically resolved from the peptide-related peaks and excluded deliberately.
The ion-pairing reagent also enters the analyte's own response. For cationic basic solutes, anionic reagents of increasing size in the mobile phase — formic, trifluoroacetic and heptafluorobutyric acid — produced somewhat increased CAD response, suggesting that salt formation with these reagents is contributory; the increase was not stoichiometric, which the authors read as pointing to a complex mechanism [4]. Changing the trifluoroacetic acid concentration in a peptide method is not a neutral change as far as the detector is concerned.
Sensitivity, measured against UV and against MS
On-column detection limits sit in the low-nanogram range. Under HILIC conditions, for solutes of low retention factor, CAD limits of detection were of the order of 1–3 ng on column, which the authors noted compares favourably with other universal detectors [4]; the inverse-gradient work reported 1 ng on column for sulfonamides [9]. For a set of cyclic peptides, microcystins were detected at 5–10 ng on the column using CAD, providing quantification limits comparable to those obtained using traditional UV detection [15]; those analytes were also measurable at 238 nm, so the comparison was a demanding one.
In the four-detector UHPLC comparison, with 25 µL injections, limits of detection were 10 ng/mL for the nano-quantity analyte detector, 76 ng/mL for the Corona CAD and 178 ng/mL for UV detection at 200 nm. Reproducibility of detector response for 11 analytes over ten consecutive separations was approximately 5% for the charged-aerosol detectors and approximately 11% for the light-scattering detectors, and the aerosol detectors were able to detect all 11 analytes in the test set whilst UV detection was applicable only to those analytes containing chromophores [12]. That last clause, not the detection limits, is the reason to reach for one. Whether an aerosol detector out-detects UV for a particular peptide depends on that peptide's absorptivity; where there is no chromophore the comparison does not arise.
The claim a mass percent supports
The specific thing these detectors license is a relative mass percent for species you hold no authentic standard for. That was done explicitly for a stressed multi-constituent formulation composed of amino acids and dipeptides, profiled by off-line two-dimensional RPLC × HILIC with ion-trap mass spectrometry and CAD. Describing the CAD as a mass-sensitive universal detector for non-volatile compounds with relatively consistent detector response, the authors set up a universal calibration function with a set of standards and then employed it to quantify unknown impurities, classifying them into those that need to be reported (>0.05% relative to the precursor compound), identified (>0.1%) and quantified (>0.15%); the dilemma of unavailability of authentic standards at that stage of research was thereby circumvented. The impurities comprised di-, tri- and tetrapeptides, cyclic dipeptides (diketopiperazines), pyroglutamic acid derivatives and their condensation products, and cross-validation with HPLC-MS/MS methods using synthesised authentic standards confirmed the results [16]. The same logic produced the composition of a protein reference material: three major components resolved by reversed-phase gradient HPLC were reported at relative compositions of 60.0%, 34.2% and 5.8% on the strength of the near-uniform CAD response [5].
Which way the correction runs matters. For partially purified microcystins, apparent purity ordered as 238 nm > UV total ion current > ESI total ion current > CAD, indicating that increased impurities could be detected using the CAD, thus providing a more accurate indication of compound quality; compounds purified by preparative HPLC were shown to have relative purities between 97% and 99%, which dropped to 90–94% when evaluated by CAD. The authors' conclusion was that a multi-detector strategy is essential, not that one detector replaces another [15].
What the measurement does not support
It does not support identity: a mass response carries no structural information, which is why, in the impurity profiling above, the relevant impurities above the reporting threshold were identified by ion-trap mass spectrometry [16].
It does not license treating an MS peak area as the same quantity. Relative ionisation efficiencies compiled for 241 peptides from tryptic digests of well-characterised monoclonal antibodies show that electrospray response for peptides has to be modelled rather than assumed: peptides shorter than 20 amino acids had efficiencies highly correlated with molecular weight, but a random forest model still predicted a test set only to a mean relative error of 23.9%, while for larger peptides a multilayer perceptron reduced the mean relative error from 60.5% under current best practices to 32.0% [17]. Those figures are prediction errors on modelled efficiencies for tryptic peptides from antibodies rather than for synthetic sequences, but they indicate the scale of the sequence dependence an aerosol detector is being used to avoid.
It does not support a claim about volatile or semi-volatile species. The semi-volatile species in the four-detector test set exhibited a more variable response on the aerosol detectors [12], and the optimal power function value in the polysorbate work differed depending on analyte volatility, with the signal-to-noise available from a higher evaporation temperature limited by the same property [7]. A small, relatively volatile fragment can be under-represented, and a fully volatile one is not represented at all.
And it does not, by itself, produce a purity figure of the kind the library's peak-integration guide dissects. The response-factor spread these studies measured — 12–18% RSD across mobile-phase composition for involatile solutes [4], around 11% average error against NMR with a generic calibrant and mass calibration [6] — sits on top of the integration error rather than instead of it. The arrangement those studies themselves used is what survives scrutiny: the aerosol detector in line with UV and with mass spectrometry, the linearisation setting fixed and declared, a volatile eluent throughout, and inverse-gradient compensation wherever response factors have to be comparable between an early- and a late-eluting peak.