Online coupling of reverse-phase and hydrophilic interaction liquid chromatography for protein and glycoprotein characterization

Chemicals and materials

Polyimide-coated fused-silica capillaries were purchased from Polymicro Technologies (Phoenix, AZ). PEEK tubing was purchased from Upchurch Scientific (Oak Harbor, WA). Four- and six-port switching valves, stainless 2-μm screens, two-way tees, and zero-dead-volume unions were purchased from Valco Instruments (Houston, TX). Jupiter C₁₈ packing materials (3-μm particles, 300-Å pores) were acquired from Phenomenex (Torrance, CA); TSKgel Amide 80 packing materials (5-μm particles, 300-Å pores) were obtained from Tosoh Bioscience (Montgomeryville, PA). Dithiothreitol (DTT), iodoacetamide (IAA), human serum, and all protein standards were purchased from Sigma-Aldrich (St. Louis, MO). Modified trypsin was acquired from Promega (Madison, WI); formic acid and trifluoroacetic acid were obtained from Fluka (St. Louis, MO). The high-performance liquid chromatography (HPLC)-grade water and acetonitrile used as solvents and for sample preparation were purchased from Tedia (Fairfield, OH).

Sample preparation

In the protein standards study, 100 μg of the protein standards were digested. The proteins were dissolved in 100 mM ammonium bicarbonate and reduced by 50 mM DTT at 60 °C for 60 min. The proteins were cooled to room temperature prior to alkylation in 100 mM IAA for 60 min at room temperature in the dark. Trypsin digestion was performed using modified trypsin (protein-to-enzyme ratio, 33:1) and incubation at 37 °C overnight for ca. 17 h. Digestion was terminated by heating to 90 °C for 10 min; samples were stored at −20 °C until required for use. In the peptide-glycan study, the tested glycoproteins were first deglycosylated with 1 unit of peptide-N ⁴-(acetyl-β-glucosaminyl)asparagine amidase (PNGaseF) at 37 °C overnight prior to trypsin digestion following the procedure described above. The resulting standard protein digest (0.5 μg/μL) was diluted 25 times; 5 μL (100 ng) was injected for the experiments. The human serum, which contained proteins at concentrations in the range 4.0–9.0 g/dL, was used without further purification. A solution (600 μL) of 100 mM ammonium bicarbonate containing 8 M urea was added to human serum (400 μL). The serum was then deglycosylated with PNGaseF (5 equiv.) and trypsinized as described above; 25 μg of the resulting peptide-glycan mixture was used for the experiment. Mouse chondrogenic cells (ATDC5) were grown in Dulbecco’s modified Eagle’s medium (DMEM)/F12 medium supplemented with 5% fetal bovine serum (FBS), 1% penicillin/streptomycin, and fungizone (1 μg/mL) at 37 °C. The cells were harvested by direct scraping, spun down, and lysed with 0.5 M triethylammonium bicarbonate (TEAB) at pH 8.5, and then subjected to sonication on ice using 25%-amplitude 0.5-s/3.5-s pauses for 30 min. The lysed cells were centrifuged at 13,400 rpm at 4 °C for 1 h to remove insoluble pellets; the collected supernatant containing the soluble proteins was quantified through a Bradford assay, following the manufacturer’s protocol, and then digested with trypsin, as described above, to yield peptides at a concentration of 0.5 μg/μL. A portion (10 μL) of the digest was used in the complex biological sample analysis.

RP-HILIC chromatographic equipment and conditions

The simplified diagram in Fig. 1a illustrates the system assembly. The micro-flow RP-HILIC system consisted of a UPLC system featuring two 65D isocratic syringe pumps and a system controller (Isco; Lincoln, NE), a gradient mixer, three four-port/two-position mechanical switch valves (valves 1, 2, 5), two six-port/two-position switch valves (valves 3, 4), and a sample injector valve (Valco Instruments; Houston, TX). Two trap columns and two analytical columns were used. The RP and HILIC trap [150-μm i.d. × 20-mm length] and analytical (150-μm i.d. × 800-mm length) columns were packed in-house using C18 and Amide80 materials, as described previously [29]. All connecting capillaries used were 75-μm i.d. × 200-mm long, unless otherwise stated; the flow-increment tube was 1.0-mm (0.04-in.) i.d. × 200-mm long; the flow restriction capillary was 20-μm i.d × 1-m long. The i.d. and length can be fine-tuned if a different gradient speed is desired. The separated analytes were detected and analyzed using MS. The switching valve positions and the solvent flow path at each step of the experiment are provided in Fig. 1b. Detailed descriptions are given below:

• Step I

  (RP, HILIC column conditioning): all valves were set to position A during the column conditioning. Pumps A and B were operated at constant pressure (7,000 and 6,000 PSI, respectively). The aqueous solvent (97.5% water, 2% acetonitrile, 0.5% formic acid; v/v/v) was directed to the gradient mixer through the sample injection valve to valve 2, where it was diverted to the RP trap and analytical column after flow-splitting; meanwhile, the organic solvent (2% water, 97.5% acetonitrile, 0.5% formic acid; v/v/v) was delivered through a two-way valve to the HILIC trap and analytical column after flow-splitting. The conditioning time for the RP and HILIC columns was 20 min.

• Step II

  (sample trapping in RP, HILIC trap): valves 3–5 were set to position B; the sample was injected via the injector and the hydrophobic analytes were extracted in the RP trap. Any unretained, hydrophilic analytes in the flow-through were then diverted through valves 2 and 3 to the solvent conversion tee, where they were mixed with the organic solvent. At this point, the flow rate, as indicated by the residual pump volume readings on the Isco UPLC, of the organic solvent was considerably higher (ca. 120 μL/min) than that of the incoming aqueous solvent (ca. 3 μL/min). The hydrophilic analytes in the then-organic medium were subsequently extracted by the HILIC trap column. The sample trapping time was 20 min.

• Step III

  (RP separation): valves 3 and 4 were set back to position A, the two-way valve was closed, and pump B was set to 8,000 psi (constant pressure). Valve 1 was then set to position B and pump B delivered organic solvent that slowly changed the aqueous composition inside the mixing chamber and generated the solvent gradient used for separation. The hydrophobic analytes trapped in the RP trap column at this point were eluted into the RP analytical column for gradient separation. The %B gradient progression over time follows the shape of “1 minus exponential decay” and is a function of the restriction flow rate [29]. Most peptides eluted from the RP column within 4 h, whereas the HILIC run was shorter at ca. 2.5 h. Longer experiments can be conducted for complex samples to take advantage of the greater plate number of the long column by modifying the gradient mixing rate. The post-column flow rate was measured using a volumetric capillary; it was ca. 1 μL/min throughout most of the gradient.

• Step IV

  (HILIC separation): The gradient mixer was prefilled with organic solvent prior to the HILIC run. Pump A was then set to deliver aqueous solvent at 8,000 PSI to generate the HILIC solvent gradient. The solvent gradient egressing from the mixer was diverted at valve 2 directly to the HILIC channel. The analytes trapped in the HILIC trap column were eluted into the HILIC analytical column for separation.

Mass spectrometry

The eluents from the RP and HILIC analytical columns were coupled online continuously to a MDS Sciex QSTAR XL quadrupole time-of-flight hybrid mass spectrometer (Applied Biosystems; Framingham, MA) through a nano-ionspray interface. Typical mass spectrometer parameters were set as follows: DP, 20 V; FP, 125 V; CUR, 20 arbitrary units (a.u.); GSI, 0 a.u.; GS2, 0 a.u.; nanospray voltage, 3000 V. Each MS/MS experiment was set to 8 h in length; the mass spectrometer was left to run in excess of the LC gradient length in information-dependent acquisition (IDA) mode to ensure all eluents were analyzed. In IDA settings, every 1-s survey scan in the range m/z 300–2000 was followed by up to three 1-s product ion scans in the range m/z 100–2,000. Selection criteria for product ion scans were the three most-intense peaks in each precursor ion scan having more than 50 counts and charge states between +2 and +4. Enhanced ion settings were used to improve sensitivity; 60-s dynamic exclusion and rolling collision energy were used to acquire non-redundant, high-quality product ion spectra. The calibration curve for the collision energy for the +2-to-+4 ions was determined through direct infusion of BSA tryptic peptides. The acquired MS/MS data were analyzed directly using the Paragon algorithm in the ProteinPilot 2.0.1 software (Applied Biosystems, Framingham, MA). The MS/MS spectra were matched against theoretical spectra generated from sequences in the Uniprot SwissProt (release 14.8) mouse (16,116 entries) and human (20,336 entries) protein databases (http://www.uniprot.org). Trypsin was set as the enzyme used; IAA was chosen as the cysteine alkylation agent; a thorough search with biological modifications settings, which contain a number of built-in modifications, was performed. The precursor mass accuracy and product ion mass accuracy were pre-set by the software. The identified peptides from the Paragon algorithm were grouped into minimal non-redundant protein sets by the ProGroup algorithm of the software. For protein identifications to be considered, a minimal unused ProtScore of 2.0 was needed. The unused ProtScore is the sum of individual peptide scores, based on identification confidence, that are not already assigned to higher-ranked protein sets; it served to minimize redundancy of protein identification based on common peptide sequences. The peptide score was based on statistical confidence of sequence matching according to the software; 99% confidence translates into a score of 2.0 and 95% confidence scores 1.3. In the complex sample analysis experiment, protein sets containing two or more peptides passing the 95% confidence threshold were automatically accepted, whereas other protein sets were manually confirmed. Each spectrum in the protein standard and peptide-glycan standard experiments was inspected manually. Possible N-glycan compositions were determined with the aid of free software [Glycomod [31] and GlycoWorkbench [32]]. The hydrophobicities of the detected peptides were calculated using Sequence Specific Retention Calculator (v. 3.0) [33].

Online system for micro-flow RP-HILIC

Previous attempts to combine RP and HILIC have invariably employed the so-called column-switching technique, with miniaturization of the first column to solve the problem of solvent incompatibility. To develop an RP-HILIC system in which both separations would be performed in a micro-flow scheme suitable for MS-based proteomics, we had to solve the problem of solvent incompatibility between the RP and HILIC columns through effective online mixing. Our system converts the aqueous solvent used in RP trapping to the organic solvent used for HILIC trapping through the application of various flow restrictions in the RP and HILIC channels. The aqueous UPLC channel, operating at constant pressure, delivered solvent at a lower flow rate, albeit with higher pressure, because of the higher restriction from the flow path. In comparison, the organic solvent channel UPLC pump operated at a lower pressure, but with a considerably higher flow rate, through the use of a much wider flow-splitting capillary (flow-increment tube in Fig. 1). Thus, the aqueous flow-through from the RP column, upon entering the organic channel, was readily mixed with a high volume of organic solvent, rendering it amenable for HILIC extraction (see Fig. 1 and the Experimental section for details).

Retention of hydrophobic and hydrophilic peptides in RP-HILIC

The coverage of peptides using hyphenated RP-HILIC was increased over that of either RP or HILIC alone, as readily demonstrated from the analyses of three protein standards: ribonuclease B, bovine serum albumin, and horseradish peroxidase. We analyzed the mixture of peptides and glycopeptides resulting from trypsin digestion of the proteins through tandem mass spectrometry (MS/MS) after separation using either RP or HILIC alone or the online RP-HILIC system. Figure 2 presents the mass spectral identification of a relatively hydrophilic peptide and a glycopeptide that eluted solely from the HILIC system; i.e., they were not detected during RP separation. Similarly, we detected a number of relatively hydrophobic peptides solely from the RP system that were unretained on the HILIC column (Table 1). Based on mass spectral evidence, the individual RP and HILIC separations resulted in the identification of a total of 58 unique native (non-glycosylated) peptides and 11 glycopeptides. The online RP-HILIC system recovered all of the native peptides and all but one glycopeptide from a single injection event; 53 native peptides were revealed in the RP part of the system and the remaining five, presumably unretained by the RP trap column, were captured by the coupled HILIC system. The HILIC system separated only six of the 10 recovered glycopeptides. Table 1 summarizes the identified peptides and glycopeptides, from the standalone RP, standalone HILIC, and RP-HILIC separations. As expected, the peptides and glycopeptides retained by the RP column generally contained longer, more-hydrophobic peptide sequences. In contrast, those retained by the HILIC system were either short and contained many polar residues (e.g., serine, threonine, glutamic acid) or they were rendered more hydrophilic as a result of extensive glycosylation. Figure 3a and b display example full-scan and product ion spectra of a detected glycopeptide from horseradish peroxidase, analyzed using RP-HILIC MS/MS. The ions of the peptide fragments as well as the characteristic N-glycan products are readily apparent, allowing characterization of the peptide sequence, the position of the glycosylation site, and the glycan structure.

Figure 4 displays mass spectrometric base peak chromatographs (BPCs) of the protein standards, recorded after performing standalone RP, standalone HILIC, and hyphenated RP-HILIC separations. From a comparison of the BPCs for the standalone RP separation and the RP separation of the RP-HILIC process, it appears that the hyphenated system did not deteriorate the performance of the RP peptide separation; we observe only minimal shifts in the retention times (within 1–2 s). Furthermore, relative to the BPC of the standalone HILIC, the HILIC chromatograph from the RP-HILIC system is considerably simpler because the RP trap column removed the signals of the relatively hydrophobic peptides. For example, chromatographic peaks corresponding to the peptide ion at m/z 742.00 are present in the BPCs of the standalone RP separation, the standalone HILIC separation, and the RP separation of the RP-HILIC system, but is absent in the BPC from the coupled HILIC system. In addition, the increased sensitivity in the coupled HILIC system, arising from the reduced sample complexity, is evident when comparing the intensities of the two peptide ions in the full-scan mass spectrum of the RP-HILIC system (Fig. 5) with those in the corresponding spectra after standalone HILIC; the latter are populated by co-eluting peptides that mutually suppress their signal strengths through competition during electrospray ionization. We suspect that such improved ionization efficiency will benefit complex sample analyses when some of the eluting peptides are present in only marginally detectable concentrations. In addition, because only one sample injection event is required for RP-HILIC, the coupled system is essentially capable of harnessing the combined separation power of the standalone RP and HILIC systems while consuming only half the amount of sample.

Retention of released glycans in deglycosylated glycoproteins

Next, we demonstrate the applicability of RP-HILIC to studies of a wide range of compounds in non-targeted analyses through the simultaneous separation of peptides and PNGaseF-released glycan chains from deglycosylated glycoproteins. We analyzed turkey ovalbumin, a previously uncharacterized glycoprotein featuring highly complex glycoforms, as well as the glycoprotein standard ribonuclease B after PNGaseF-mediated deglycosylation and trypsin-mediated digestion. For turkey ovalbumin, the mixture of peptides and highly hydrophilic glycan chains was successfully extracted and separated by the RP-HILIC system, enabling us to report 23 novel N-glycan compositions (Table 2). From ribonuclease B, all five known major N-linked glycans (GlcNAc₂Man₅, GlcNAc₂Man₆, GlcNAc₂Man₇, GlcNAc₂Man₈, GlcNAc₂Man₉) were separated and detected through unequivocal spectral evidence (Fig. 6); the peptide coverage was similar to that obtained using standalone RP (data not shown).

Because the released glycans are considerably more hydrophilic than typical glycopeptides, we obtained a larger discrepancy in their analyses when using the RP and HILIC systems alone for the efficient separation of the different components in the glycoprotein samples. The great number of similar glycoforms present in complex glycoproteins, such as turkey ovalbumin, makes them difficult to characterize without efficient separation, suggesting that RP-HILIC has great potential for allowing sophisticated analyses of highly heterogeneous mixtures.

Increased component coverage in complex biological samples

To test the performance of RP-HILIC in the analysis of highly complex proteomics samples, we analyzed the total protein extract of mouse chondrogenic cells without further purification. From a single injection of a 10-μg sample, we identified a total of 6,812 distinct peptides using the RP component of the RP-HILIC system, translating into the identification of 493 unique proteins (95% confidence). The coupled HILIC system contributed a further 985 distinct peptides and allowed the overall identification of an additional 11 proteins (see Electronic Supplementary Material Figure S1). We attribute the increased coverage mainly to the separation and detection of hydrophilic peptides in the coupled HILIC. The calculated average hydrophobicity of the detected peptides in the RP component was 30.5 (ranging from 2.48 to 63.4); in contrast, the coupled HILIC allowed the detection of considerably-more-hydrophilic peptides, with a calculated average hydrophobicity of 12.3 (from 3.17 to 20.24; see Electronic Supplementary Material Figure S2). Our results obtained using the RP and RP-HILIC systems reveal that at least 10% of the detected proteins saw increases in sequence coverage when we included the coupled HILIC component; this improvement is likely to benefit peptide-based proteomics applications, such as bottom-up quantification and characterization of post-translational modifications.

Finally, when we evaluated the performance of the RP-HILIC system in the separation of deglycosylated and digested human serum, we identified more than 80 potential glycans without prior immunodepletion of abundant serum proteins. We foresee that these glycan structures might be resolved through the combined use of RP-HILIC and mass-analyzed laser desorption ionization (MALDI)–time-of-flight (TOF)/TOF MS/MS. After RP analysis of the relatively hydrophobic peptides, it might be worthwhile to directly deposit the HILIC eluent onto a MALDI plate for offline analysis. Because MALDI-TOF/TOF systems display higher glycan ionization efficiencies than electrospray ionization MS systems, they might potentially reveal cross-ring cleavage information of N-glycans when using high-energy collision-induced dissociation [34–36].

Below is the link to the electronic supplementary material.