The peptides that defeat reversed-phase chromatography defeat it for a reason you can read straight off the sequence. A retention model fitted to tryptic peptides found that Phe, Trp, Ile and Leu strongly promote reversed-phase retention while His, Lys and Arg strongly reduce it, and that on bare silica, ethylene-bridged hybrid silica and amide HILIC columns those same residues acted in the opposite direction, His, Lys and Arg promoting retention and Phe, Trp, Ile and Leu demoting it [4]. A sequence built mostly from charged and small polar residues is therefore a weak reversed-phase analyte and a strong HILIC analyte. That inversion, not any general superiority, is what the technique is for.
Alpert named the mode in 1990 on exactly this basis: a hydrophilic column eluted with a mostly organic mobile phase retains solutes in order of increasing hydrophilicity, typically giving the reverse of the reversed-phase elution order, and he singled out phosphorylated amino acids and peptides among the troublesome solutes it suited [1].
The water-enriched layer, and what partitioning does not explain
The standard mechanistic picture is a layer of the aqueous component held at the polar stationary-phase surface, into which polar analytes partition out of the organic-rich bulk eluent. That layer has been sized rather than assumed. Benzene, a small hydrophobic probe that does not partition into it, was injected onto a silica column under HILIC conditions and found to be partially excluded from the pores up to a water content of about 30%, above which hydrophobic retention on siloxane bonds took over. The exclusion implied that roughly 4–13% of the pore volume of the silica phase was occupied by a water-rich layer at acetonitrile–water compositions between 95% and 70% (v/v) acetonitrile [2]. The authors were careful about the extreme of that measurement: in 100% acetonitrile the benzene retention volume came out smaller than either pycnometry or the column's basic physical parameters predicted, which they suggested might be attributable to the larger size of the benzene molecule, while noting that some influence of strongly adsorbed water surviving extensive purging with dry acetonitrile could not be entirely discounted [2].
Partitioning is not the whole mechanism, and treating it as the whole mechanism will mislead method development. A study that measured the adsorbed-water phase ratio and the distribution coefficients separately, and could therefore split retention into a partitioning term and a surface-adsorption term, found across five neutral test compounds that retention may be controlled by hydrophilic partitioning, by surface adsorption, or by both, depending on the characteristics of the compound [3]. Those were small neutral molecules, so the specific split does not carry over to peptides; the caution does. For peptides, the authors of a large retention-modelling study on zwitterionic phases concluded that HILIC interactions are driven both by distribution of the peptide into the water layer adsorbed on the stationary phase and by interactions specific to the functional groups of the packing material [6].
Why the gradient runs backwards
A HILIC gradient starts high in organic and descends. In the phosphoproteomics work that used HILIC as a first-dimension peptide fractionation, tryptic peptides were separated on a TSKgel Amide-80 column using a shallow inverse organic gradient, under which peptide retention was based on overall hydrophilicity and the separation produced was orthogonal to reversed phase [12].
Two consequences follow. The eluting composition is organic-rich rather than water-rich: in a systematic HILIC-versus-RPLC comparison run on a set of 56 basic pharmaceutical compounds, analytes eluted at an average of 82% acetonitrile in HILIC against 29% acetonitrile in RPLC at pH 6 [15]. And the aqueous component is not dispensed with: the retaining layer was sized under acetonitrile-water eluents rather than under dry acetonitrile [2].
Stationary phase chemistry is not a single axis
Ten pharmaceutically relevant model peptides run across seven HILIC columns — bare silica, amide, poly-hydroxyethyl aspartamide, diol and zwitterionic chemistries — clustered in a way that punishes the habit of treating a phase label as a selectivity. Hierarchical clustering and principal component analysis placed the zwitterionic phase in a cluster apart from the other columns, placed the two poly-aspartamide columns together, and placed the two bare silica phases in two different clusters, that is, at different selectivities. On the composite desirability score used, one of the bare silica columns performed best for that peptide set, followed by ZIC-HILIC [5].
Hydrophilicity ranks differently from selectivity. Retention models built on proteomics-derived datasets of roughly 30,000 tryptic peptides each, at pH 4.5, found the zwitterionic sulfoalkylbetaine (ZIC-HILIC) and phosphorylcholine (ZIC-cHILIC) phases similar in overall hydrophilicity to the previously studied amide phase but lower than bare silicas [6]. The two zwitterionic phases differ in the spatial orientation of their charged groups, and that difference surfaces in the residue coefficients: ZIC-cHILIC, whose positively charged quaternary amines are positioned distally, showed an increased contribution from Asp and Glu. Retention on both was driven by all the charged residues at that pH — Asp, Glu, Arg, Lys and His — but with a specificity that follows the orientation of the zwitterionic pair [6].
Know the quality of a retention model before you lean on it. Sequence-Specific Retention Calculator models optimised for the two zwitterionic columns reached R2 values of 0.967–0.976 between experimental and predicted retention [6]; the earlier model spanning bare silica, bridged-ethyl hybrid silica and amide reached R2 of about 0.92–0.97 [4]. Both were fitted on tryptic peptides. Neither licenses prediction for a synthetic sequence containing non-natural residues.
Orthogonality against reversed phase is measurable, and it is the point of the exercise. Correlation between HILIC and RPLC retention values for the same peptide sets was R2 0.197 for ZIC-HILIC and 0.137 for ZIC-cHILIC [6]. On synthetic material, an impurity-profiling study of four cyclic peptides of nine amino acids each, built from natural and non-natural residues with neutral, acidic, basic or zwitterionic residues placed at positions 3 and 6, screened three polymer-based HILIC columns (acidic, basic and zwitterionic) against seven mobile phases and found the two techniques mutually orthogonal: most impurities with small reversed-phase retention times had large HILIC retention times, and the reverse [18]. Ammonium acetate was identified as an effective additive, and polymeric selectors carrying vinylpyridine and phosphoryl choline residues were effective irrespective of the peptide features [18].
Pushing the ionic term the other way
Electrostatic repulsion–hydrophilic interaction chromatography (ERLIC) exploits the ion-exchange contribution deliberately. An ion-exchange column eluted with a predominantly organic mobile phase can retain solutes through hydrophilic interaction even when they carry the same charge as the stationary phase; with mixtures whose components differ greatly in charge, the repulsion can be used to selectively antagonise retention of the solutes that would otherwise be best retained. That permits isocratic resolution of mixtures that would normally require gradients, affords separations of highly charged peptides that are not readily resolved by other means, and allows phosphopeptides to be isolated selectively from a tryptic digest [13].
Equilibration, and the reproducibility it governs
HILIC's reputation for slow equilibration comes from isocratic work, and it does not transfer wholesale to gradients. Judged by retention-time stability, which proved more informative than the mobile-phase disturbance method or column efficiency measurements, full equilibration for isocratic analysis could take up to an hour and was found to depend on the nature of the stationary phase, on the pre-equilibrium storage solvent, and on the flow rate [7]. The operationally important result sits in the same study: in gradient elution a repeatable partial equilibration gave retention-time stability equivalent to that in isocratic analysis after an equilibration of only about 5 minutes, using as few as two preliminary conditioning runs, and this was demonstrated on the column that had taken the longest to reach full equilibration [7].
The follow-up work generalised the effect across a wide variety of HILIC stationary phases: repeatable partial equilibration in gradient elution was demonstrated after purging with as little as 12 column volumes, with retention-time relative standard deviations of on average about 0.15% after one or two conditioning (blank) runs [8]. Full equilibration was reported to be the favoured strategy for stationary phases that do not trap extensive water layers and for wider-pore, lower-surface-area materials, and temperatures somewhat above ambient shorten the time [8].
The condition attached to partial equilibration is strict, and it is where methods actually break. Because selectivity changes with the equilibration time, a series of analyses must use identical gradient conditions throughout — something the same study describes as facile on a modern instrument, but fatal if a sequence is edited mid-run [7]. Retention drift blamed on a column can instead be a shortened re-equilibration on one injection in a batch. Build the conditioning runs into the sequence and read them as conditioning, not as data [7, 8].
Sample diluent: an easy way to ruin a peak
Mismatch between injection solvent and mobile phase outweighed injection volume as an injection-side cause of poor peak shape in the one study that varied both systematically. With the mobile phase held at 95% acetonitrile (v/v) and weak and stronger bases injected on a hybrid bare silica phase, widening the mismatch across an injection-solvent range of 95% down to 0% acetonitrile gave progressively worse peak shape; on the 2.1 mm i.d. columns used, raising the injection volume from 1 to 20 µL in mobile phase also worsened peak shape, but by a considerably smaller margin than solvent mismatch over that range [9]. If a HILIC peak splits or fronts, change the diluent before you change the column.
The physical mechanism has been resolved quantitatively, on a small molecule rather than a peptide. For cetirizine on a 4.6 mm × 100 mm, 2.5 µm BEH HILIC column with an acetonitrile–water eluent and a sample diluent containing 7% and 9% water, the distortion was traced to four terms: the diluent-to-eluent excess of water propagating along the column at a velocity similar to that of the analyte, the significant drop in the analyte's Henry constant as water concentration in the eluent increases, the injected sample volume, and the pre-column dilution factor set by the characteristics of the instrument [10]. That last term is why a method can transfer badly between an HPLC and a UHPLC without anything else changing [10].
Neat acetonitrile is the usual default diluent, but not a universal one. On small-internal-diameter (0.1–0.3 mm) zwitterionic and charged columns at nano and micro flow, pure acetonitrile was demonstrated to be the best sample solvent for obtaining narrow peaks for retained compounds, yet distorted the peaks of compounds that eluted very close to the solvent peak. A highly aqueous sample solvent, which interferes with the partitioning of polar analytes into the stationary phase, was detrimental to the peak shape of retained neutral compounds but not problematic for unretained ones [11]. For charged compounds the same study found that water in the sample solvent favoured electrostatic attraction with the stationary phase, so retention time increased with the water content of the diluent while peak shapes stayed acceptable even when a large amount of water was present [11]. Because diluent choice is bounded by what the sample will dissolve in, note the alternative the authors reached for: a mixture of acetonitrile with methanol or isopropanol, for highly polar compounds with limited solubility in aqueous sample solvents [11].
Two other peak-shape faults deserve to be separated from diluent effects, because a diluent change will not fix them. Structural features including galloyl (trihydroxy benzene), catechol (benzene diol) and phosphate in nucleotides gave serious peak tailing, attributed to interactions with metals in the stationary phase or in the chromatographic hardware; the effect was moderated by including complexing agents in the mobile phase, by changing the stationary phase chemistry, or by altering mobile phase pH [9]. That phosphate observation was made on nucleotides, not on phosphopeptides, so treat it as a hypothesis to test when a phosphopeptide tails, not as a diagnosis.
Phosphorylated and glycosylated peptides
The hydrophilicity of the phosphate group is itself a retention handle. Working on an amide column with a shallow inverse organic gradient, phosphopeptides were selectively enriched and fractionated on the basis of their increased HILIC retention, achieved without derivatisation or chemical modifiers; IMAC enrichment applied afterwards to the HILIC fractions showed better than 99% selectivity, and a 300 µg equivalent of HeLa cell lysate digest yielded over 1000 unique phosphorylation sites, more than 700 of them new additions to the HeLa phosphoproteome as it then stood [12]. In the same work, protein identification numbers from HeLa tryptic digests were comparable to those obtained using strong cation exchange [12].
For glycopeptides sharing a peptide backbone the glycan sets the retention order, and the stationary phase interacts with the glycan's charge. Across HALO penta-HILIC, Glycan BEH Amide and ZIC-HILIC columns run on complex N-glycopeptides of hemopexin and immunoglobulin G, retention time increased with the polarity of the glycans attached to the same peptide backbone on every column tested, except on ZIC-HILIC when sialic acid was added to the glycan moiety, which caused electrostatic repulsion with the negatively charged sulfobetaine functional group and decreased retention [14]. The penta-HILIC column gave the best separation of that set and ZIC-HILIC the worst, and the columns showed potential for isomeric separation of fucosylated and sialylated glycoforms [14]. The ranking belongs to those glycopeptides on those columns; what generalises is the warning that a sialylated glycopeptide on a zwitterionic phase can move the wrong way.
What the electrospray gain is actually worth
The organic-rich eluent genuinely helps electrospray, but most of the headline numbers were measured on small molecules. In the 56-compound comparison, the median gain in sensitivity for HILIC over RPLC was equal to 4 whatever the pH, and the best signal-to-noise was systematically attained under HILIC for about 90% of the tested compounds and analytical conditions; the average gains of 7 at pH 3 and 10 at pH 6 were described by the authors as unreliable, because a few compounds behaved extremely [15]. They attributed the gain to better desolvation at 82% versus 29% acetonitrile at elution, and also to the high proportion of solvent influencing ionisation by modifying pH and pKa, since the weakest bases in the set (pKa 2–5) showed an unexpectedly strong gain of 20- to 100-fold [15].
Two things stop that number transferring to peptide work. The first is the ion source. Across three triple quadrupole instruments with brand-specific source geometries and 53 pharmaceutical compounds, the proportion of compounds with improved signal-to-noise in HILIC swung from 92% to 7% on one instrument on the strength of flow rate and mobile phase pH alone, and the gain in sensitivity observed in HILIC was found less significant on those recent platforms than on old-generation instruments, because improved sources evaporate highly aqueous reversed-phase eluents efficiently even at high flow rates [16].
The second is the analyte class, and here there is a direct peptide measurement. Comparing three HILIC columns against a C18 column for targeted quantification of proteotypic peptides by selected reaction monitoring, maximum HILIC sensitivity was achieved on an amide column without salt buffer, and signal did increase relative to reversed phase — but the intensity improvement was described as quite low compared with that obtained for small molecules, for two stated reasons: a higher matrix effect in HILIC, and a change of peptide charge states in organic solvent from doubly to singly charged [17]. Doubly charged ions dissociate more readily than singly charged ones and are the ones you want for SRM, so supercharging reagents were added to the mobile phase to shift the population back; under those optimised conditions the peptide signal improved by a factor of between two and ten for 88% of the 81 peptides investigated [17].
The practical reading is narrow. Reach for HILIC on a peptide because reversed phase will not retain or resolve it, and treat any ionisation gain as a secondary, instrument-dependent bonus that charge-state redistribution may partly consume. Volatile additives keep the eluent compatible with the mass spectrometer, and ammonium acetate was the effective additive in the cyclic peptide screen, which was run with UV rather than MS detection [18] — but the highest peptide SRM sensitivity in the amide-column comparison came without salt buffer [17]. Those two constraints pull against each other and have to be settled on the analyte in front of you, not on a general rule.