Now that several years have passed since the introduction of ultrahigh-pressure liquid chromatography (UHPLC), I think we all can agree that it is a technique that is here to stay. I'm not convinced, as some of the advertising might lead you to believe, that UHPLC will be replacing all LC methods in the near future. However, UHPLC has moved beyond the research and development environment into more routine use in many companies. Some companies use UHPLC for the development of analytical methods and then covert them to conventional LC methods for routine use; others use UHPLC in all areas for new applications; still others are staying with the traditional LC methods and equipment.   

Through contacts in my teaching and consulting activities, attendance at scientific meetings, and e-mail with readers like you, I have collected a number of tips and techniques that are used in practical applications to get the most out of UHPLC equipment. It is this subject that I would like to explore a bit in this month's "LC Troubleshooting." I will add more to this in the future, so if you have found particular tricks that you would like to share with other readers, drop me an email (

) and include "UHPLC" in the subject line. For the present discussion, I will make a binary division. LC (or as many call it, high performance LC or HPLC) will refer to equipment that operates up to the conventional pressure limits of 6000 psi (400 bar); UHPLC will refer to all equipment that operates at higher pressures. 

As is summarized in Table I, the passageways through the typical UHPLC system are smaller than those for conventional LC systems. To minimize dispersion problems when 5-μm diameter particles are used as a column packing, connecting tubing should be no larger than 0.007 in. i.d. (0.175 mm) in the areas where the sample travels: between the autosampler and column and between the column and detector. If 3-μm particles are used, 0.005-in. (0.125-mm) tubing is required. Even this latter tubing can be a challenge. I remember when Hewlett-Packard first introduced their 1190 system, which was one of the early low-dispersion LC systems. Originally it was plumbed with 0.005-in. tubing, but after numerous problems with tubing becoming blocked in routine operation, they had their service engineers replace the 0.005-in. tubing with 0.007-in. i.d. tubing, which is less prone to blockage. The further challenges of UHPLC with sub-2-μm particles require even less extracolumn peak broadening, which means that even smaller tubing diameters are used; typically, 0.0025 in. i.d. (0.00625 mm), which is even more prone to blockage.  

Table I: LC versus UHPLC particulate challenges 

One of the complaints I commonly hear about 3-μm particle diameter columns is that they become blocked much more easily than 5-μm columns. The primary reason for this is that 5-μm columns use 2-μm porosity frits to hold the particles in the column. This frit size is not sufficiently small to retain all the particles in many 3-μm particle columns, so they typically use 0.5-μm porosity frits on the outlet and either 0.5- or 2-μm frits on the column inlet. With sub-2-μm particles, the 0.5-μm frit is not satisfactory, so 0.2-μm frits are used commonly. Those of you who have had a microbiology class along the way will remember that 0.22-μm porosity filters are used to remove bacteria from solution. Ever see an analytical chemistry laboratory that is bacteria-free? 

So how do these items affect the practical use of UHPLC in the laboratory? With 3- or 5-μm particles, sample filtration often can be avoided by simply placing the samples in a benchtop centrifuge, turning the speed up to the maximum for 5 min to settle any particles, and transferring the supernatant to the injection vial. As a safety measure, I strongly recommend using an 0.5-μm porosity in-line filter between the autosampler and guard column or analytical column to catch the occasional sample particle that was missed in the centrifugation step. With sub-2-μm columns and 0.2-μm frits, the centrifugation step might not be sufficient. If this is the case, every sample must be filtered through a 0.2-μm porosity frit — a very slow process. Whether centrifugation or filtration is used, a 0.2-μm porosity in-line filter is strongly recommended as a safety precaution.  

Another source of particulate matter that can block the column is the mobile phase. Conventional LC applications with 3- or 5-μm particle columns are fairly tolerant to minor levels of mobile-phase contamination. For example, if the buffer or other aqueous mobile phase is filtered through a 0.5-μm porosity solvent filter, it can be used for at least a week in most laboratories, and many laboratories extend this to two weeks. For UHPLC applications, many workers observe sufficient microbial growth in aqueous mobile phases that fresh buffers must be made daily and filtration through a 0.2-μm porosity filter is required. In either case, the reservoir should be replaced with a clean one each time mobile phase is replenished, so that a minor microbial contamination does not amplify itself in the next batch of mobile phase.  

Still another source of particulate matter is the wear of piston seals and injection valve rotors. I recently saw a UHPLC presentation that discussed the use of a continuous seal wash to help keep the piston cool so that the heat generated by the seal gripping the piston did not cause the seal to melt. The data also showed a reduction of piston seal lifetimes of fourfold at the highest pressures tested. Similarly, the increased pressure of the UHPLC system requires that the rotor seal and stator of the injection valve be pushed together more tightly, increasing friction and reducing the seal lifetimes. So it is imperative to pay close attention to system software that counts piston and injection cycles so that seals can be changed on a timely basis to minimize particles from these sources. 

What does all this tell us? As one of my friends terms it, "chromatographic hygiene" must be at a much more stringent level for UHPLC than for LC applications. But many workers take care to keep particulate matter out of the system and can use UHPLC reliably in routine analysis with great results.  

When most of us think about the increased pressure that is observed with a UHPLC system, it seems logical that it results from smaller-particle columns and higher linear velocities. We get faster separations at higher pressure and don't think much else about it as long as the equipment can handle the increased pressure. However, it isn't quite so simple. It has been known for more than 40 years (1,2) that pressure can change the molar volume of solutes; this can result in changes in selectivity, or relative peak spacing in the chromatogram. I know that this came as a surprise to me when we were testing one of the first UHPLC systems. We observed that the peak spacing for a closely eluted triplet changed dramatically as we changed the pressure from 5400 psi (370 bar) to 12,800 psi (880 bar).  

With the right compounds, changes in peak spacing can be very dramatic. In the example of Figure 1 (3), a change in column pressure was achieved by adding a restriction capillary between the column and detector; the same flow rate and other conditions were used. You can see that as the pressure was increased from 43 bar (625 psi) to 739 bar (10,700 psi), there are two peak reversals. Peaks 3 and 4 (diphenhydramine and acetophenone) reversed retention order, as did peaks 5 and 6 (protriptyline and nitrobenzene). These changes are attributed primarily to a differential change in the molar volume of the solutes (3).  

Figure 1: Pressure selectivity effects: Separation observed at (a) 43 bar and at (b) 739 bar by the addition of a restriction capillary after the column. Column: C18; mobile phase: acetonitrileâphosphate. Peaks: 1 = thiourea, 2 = propranolol, 3 = diphenhydramine, 4 = acetophenone, 5 = protriptyline, 6 = nitrobenzene. Adapted from reference 3. 

What does this mean in practical terms? With the relatively small changes in pressure encountered in conventional LC systems (for example, a factor of two when the flow rate is doubled), most of us don't ever notice this effect. However, one of the common deployments of UHPLC technology in the pharmaceutical industry takes advantage of the speed of UHPLC to develop new methods in the research laboratory more quickly than could be done with conventional LC equipment. But there also is a realization that many methods are performed in a production environment where the equipment, skills, and environmental conditions might not be as favorable. So to improve the chances of success, the UHPLC methods are converted to conventional LC methods on 3- or 5-μm particle columns that can be operated on existing, lower-pressure equipment. This can save in training, equipment costs, and method downtime. However, when pressure selectivity is taken into account, scaling a method from UHPLC to LC might not be as simple as initially thought. This is of special concern for methods that must be capable of detecting many small peaks, such as stability-indicating or impurity methods. This does not mean that the UHPLC development and LC application strategy cannot work, but it does mean that special care should be taken to check for possible changes in peak spacing when the system pressure is changed. For a nice discussion of other effects of pressure in LC applications, see the introduction to reference 3. 

When liquid passes through a bed of particles, the friction between the two phases creates heat (sometimes called viscous heating). As with pressure effects, with the 3- or 5-μm particles and 1–2 mL/min flow rates used with conventional LC methods, we seldom notice this effect. However, when the particle sizes are reduced to <2 μm and flow rates (or more properly, linear velocities) are increased, a significant amount of heat can be generated. As has been discussed here before (for example, in reference 4), a change in column temperature can have a dramatic effect on the separation. In fact, peak reversal can occur from changes in column temperature. The use of temperature can be a powerful tool to help optimize a separation, but it also can be the source of problems in a separation if it is not carefully controlled. The effect of temperature on peak shape is shown in the cartoons of Figure 2. In Figure 2a, a fully temperature-equilibrated column is shown. The incoming mobile phase, column bed, and column oven are all at 70 °C. The uniform temperature distribution across the column means that all the molecules in a single band travel at the same speed, no matter where they are relative to the column diameter, and a relatively narrow peak results. If solvent preheating is ignored, the incoming mobile phase can be at a significantly lower temperature than the rest of the system, as shown in Figure 2b. Here, the solvent at the center of the column is at a lower temperature than that at the edges of the column, where the column oven has its strongest influence. As a result, molecules in the center of the column travel more slowly than their warmer siblings closer to the column wall. This causes peak broadening. The opposite can happen if heat is created within the column, as is illustrated in Figure 2b. Here, the center of the column is warmer than the edges, where heat is dissipated into the column wall and oven, so the molecules in the center travel more quickly than the ones at the edge. Once again, peak broadening results.  

Figure 2: Effect of uneven column heating: (a) column and mobile phase fully equilibrated, (b) cold incoming mobile phase, (c) frictional heating within column. 

The combination of these temperature effects can be interesting. In one case, we noted that a sub-2-μm column generated the narrowest peaks (highest plate number) at 65 °C. Upon further investigation, we found that the oven was not performing very well, so that most of the column heating was provided not by the column oven, but by the preheated solvent (the oven was, in fact, acting as a radiator to cool the column). So under these conditions, the solvent coming into the column was warmer than that leaving (and at the column walls). On the other hand, the small particle size and high linear velocity created heat within the column, so the temperature at the column outlet was higher than at the inlet. This also caused peak broadening. But at 65 °C, the cooling effect of a poorly performing oven and the heating effect of frictional heating balanced each other and the column temperature was uniform, resulting in the best performance. Although this was an interesting observation, it certainly is not conducive to easily transferred methods. It is very important to preheat the mobile phase for methods operating above ≈40 °C, to have a column oven that works well, and to pay attention to possible influences of frictional heating. 

UHPLC can be a great tool to speed method development and routine analysis. However, it is not without potential problems. In this discussion, we saw that cleanliness is extremely important to reliable operation. Solvents should be filtered and changed more often than with conventional LC systems. More attention may need to be given to preventive maintenance of pump and injection rotor seals. Care should be taken to minimize particulate matter in samples either through centrifugation or filtering. Additional protection of the column should be undertaken by installing a 0.2-μm in-line filter just downstream from the autosampler in every system.  

We also reviewed two phenomena that can be a surprise when using UHPLC. Pressure selectivity is seldom observed with conventional LC methods, whereas it might be more prevalent in UHPLC, especially when UHPLC methods are scaled to lower pressures for routine operation. Similarly, frictional heating is not a common problem under conventional LC conditions with 3- or 5-μm particles, but can potentially change selectivity and column efficiency with UHPLC. 

As my mother always told me when I was learning to drive as a teenager, "Be careful. Pay attention to what you are doing." This still is good advice to her chromatographer son decades later. 

"LC Troubleshooting" Editor John Dolan has been writing "LC Troubleshooting" for 

 for over 25 years. One of the industry's most respected professionals, John is currently the Vice President of and a principal instructor for LC Resources, Walnut Creek, California. He is also a member of 

's editorial advisory board. Direct correspondence about this column to "LC Troubleshooting," 

, Woodbridge Corporate Plaza, 485 Route 1 South, Building F, First Floor, Iselin, NJ 08830, e-mail 

For an ongoing discussion of LC troubleshooting with John Dolan and other chromatographers, visit the Chromatography Forum discussion group at 

(3) M.M. Fallas, U.D. Neue, M.R. Hadley, and D.V. McCalley, 

Articles

LC Troubleshooting

John Dolan presents some practical pointers to get the most out of your UHPLC system.

UHPLC Tips and Techniques

The FDA is responsible for regulation of dietary supplements as foodstuffs through the 1994 Dietary Supplement Health and Education Act (DSHEA) amendment to the Food Drug and Cosmetic Act (FDCA). Product standardized practices were set, but compliance was voluntary and no legal requirement existed for manufacturers. Due to public and industry concern, in 2003, the FDA proposed requiring dietary supplement manufacturing to adhere to current good manufacturing practices (cGMPs). The final rule was issued in June 2007 and is in full effect June 2010 (1). Basic GMPs require implementing comprehensive procedures to ensure product quality and safety. Because many dietary supplements are largely derived from botanical sources, they must be tested for pesticide contaminants in order to meet cGMP regulations. As a result of this requirement, labs are working to develop and validate methods, an endeavor that is complicated by the wide range of pesticides and matrices to be tested. 

Although other researchers have used established methods like those found in the FDA Pesticide Analytical Manual (PAM) (2) as an alternative, we developed a QuEChERS (quick, easy, cheap, effective, rugged, and safe)-based method for analyzing pesticides in dietary supplements. QuEChERS is an approach developed by Anastassiades and colleagues (3) as a simple, rapid, effective, yet inexpensive way to extract pesticide residues from fruits and vegetables, followed by a novel dispersive solid-phase extraction (dSPE) cleanup of the extract. Because of these benefits, the approach has become popular and has been expanded to include numerous other matrices. We chose QuEChERS as an alternative to PAM-based methods because of its speed, simplicity, and low solvent use, as well as its ability to produce good extraction efficiencies for relatively polar pesticides (3).  

QuEChERS extraction and dSPE tubes, QuEChERS internal and quality control standards, and cartridge SPE tubes (cSPE) were from Restek Corporation (Bellefonte, Pennsylvania). The 46-pesticide mix, representing different chemical classes that have been reported previously in dietary supplements (4–6), was a custom standard produced by Restek Analytical Reference Materials group (Bellefonte, Pennsylvania).  

 Typically, QuEChERS methods use 10–15 g of material and are ideal for commodities with high water content (>80%). Therefore, to prepare for a QuEChERS extraction with a dry commodity, it is critical to use a reduced amount of material and wet it with water prior to extraction. In this work, 1 g of dietary supplement powder was combined with 9 mL of water. After shaking to mix well, the wetted supplement was fortified with 200 μL of a 2 ng/μL pesticides spiking solution resulting in a 400-ng/g spike level, relative to the original commodity. Also, 100 μL of QuEChERS Internal Standard Mix for GC–MS analysis was added. The sample was allowed to soak for 2 h. 

 The EN 15662 QuEChERS method was used for sample extraction (7). A 10-mL volume of acetonitrile was added to the wetted sample. After a 1-min shake, buffering extraction salts (Q-sep Q110, Restek) containing 4 g magnesium sulfate, 1 g sodium chloride, 1 g trisodium citrate dihydrate, and 0.5 g disodium hydrogen citrate sesquihydrate were added. Following another 1-min shake, the sample was centrifuged for 5 min at 3000 U/min with a centrifuge (Q-sep 3000, Restek). Lastly, 5 μL of a quality control standard containing anthracene was added to a 1-mL aliquot of extract to monitor for potential losses of planar compounds to a graphitized carbon black during cleanup. 

 Based upon preliminary studies, we knew that while the extraction part of QuEChERS would be successful, the dSPE cleanup step probably did not have the capacity to handle the matrix complexity and amount of most dietary supplements. Thus, we compared dSPE to a cSPE cleanup and established a procedure that uses a QuEChERS extraction, cSPE cleanup, and GC–TOF-MS for determinations of 46 pesticides in dandelion root powder. 

Two approaches were explored for extract cleanup: dSPE and cSPE. For dSPE, 1 mL of extract was added to a Q210 dSPE tube containing 150 mg  of magnesium sulfate and 25 mg primary secondary amine (PSA), shaken for 2 min, and then centrifuged for 5 min. The resulting final extract was then analyzed by GC–TOF-MS. 

For cSPE cleanup (8), 1 mL of extract was processed with a 6-mL Combo SPE cartridge (Resprep, Restek), which is designed for pesticide residue cleanup and contains 500 mg of a form of graphitized carbon black and 500 mg PSA. To prepare the SPE cartridge, approximately 300 mg of magnesium sulfate was first added; then the cartridge was rinsed with 20 mL of 3:1 acetonitrile–toluene, which was discarded. For cleanup, 1 mL of extract was loaded onto the prepared cartridge and then eluted with 50 mL 3:1 acetonitrile–toluene. The eluent was then evaporated and solvent exchanged using dry nitrogen gas and a 35–40 °C water bath. Evaporation was allowed to proceed until approximately 0.5–1 mL eluent was left, at which point about 3 mL of toluene was added. The mixture was evaporated to just under 0.5 mL, and then the evaporation vessel was rinsed with toluene to bring the sample to a final volume of 0.5 mL. The resulting final extract was then analyzed by GC–TOF-MS. 

 Matrix-matched standards were prepared at 80 pg/μL, as 80 pg/μL is the expected final concentration in extract of the 400-ng/g matrix spikes (assuming 100% recoveries). Matrix-matched standards were prepared by adding standard solution to the final extract (postcleanup) from a control sample. Actual recoveries were calculated by comparing peak areas for fortified samples that were extracted and cleaned up, to areas of a matrix-matched standard, using the internal standard quantification method with PCB 52 from the internal standard mix added before extraction.  

 A LECO Pegasus III GC–TOFMS system was used and all data were processed with software from LECO (ChromaTOF, LECO Corporation, Saint Joseph, Michigan). GC was performed using a 30 m × 0.25 mm, 0.25-μm 

 column (Rxi-5Sil MS, Restek) with a constant flow of helium at 1.5 mL/min and a fast autosampler splitless injection of 1 μL, purge valve time of 1.5 min, into a 5-mm single gooseneck liner with wool (Restek). The inlet temperature was 250 °C and the GC oven program began with a temperature of 90 °C (1.5 min) followed by an increase at 8 °C/min to 340 °C. Total run time was 32.75 min. Electron ionization at 70 eV was used with a source temperature of 225 °C. Data acquisition was from 45 to 550 u at an acquisition rate of 5 spectra/s.  

One aspect of this investigation was to compare the applicability of two sample cleanup methods, dSPE and cSPE, for QuEChERS extracts of pesticides in dietary supplements. Several formulations of dSPE were tested having different combinations and varying amounts of PSA, C18 and GCB sorbents. While dSPE has the advantage of improved speed and less solvent usage, it does not have the sorbent capacity to adequately clean up these samples (Figure 1). Figure 1 shows a comparison of the effectiveness of dSPE and cSPE to remove common matrix compounds like fatty acids, sugars and other co-extractives. The blue plot on the left side shows the fatty acids of dandelion root processed with dSPE, while the green plot is that of dandelion root processed by cSPE. Because cSPE uses more sorbent, it is a better choice for dietary supplements (and other complex samples, for example, spices, essential oils) as it can remove a higher amount of matrix components, such as fatty acids, sugars, and pigments. QuEChERS methods developed for dietary supplements of botanical origin can benefit from the extra sorbent capacity of cSPE, which reduces GC inlet–column contamination and chromatographic interference from complex botanical matrices.  

Figure 1: Comparison of sugars and fatty acids removal by extracted ion chromatograms of m/z 60. Left: dandelion root processed with dSPE. Right: dandelion root processed with the cSPE method described in the Experimental section. 			

Even with effective extraction and cleanup techniques, dietary supplements can be challenging to analyze due to their complexity. Coelutions are common and pesticide residues can be overwhelmed by abundant matrix compounds not only qualitatively, but also by interfering with quantitation masses. Figure 2 plots the total ion chromatogram (TIC) and extracted ion 

 312 corresponding to the quantitation mass for carfentrazone ethyl. It is clear that target pesticide signals can be obscured in the TIC. The software mentioned earlier was used to identify target pesticides by comparison with empirical reference spectra, after automatic peak find and spectral deconvolution algorithms, as well as performing calibration and quantification. TOF-MS makes this powerful data processing possible by providing fast acquisition rates and unbiased mass spectra, and by having picogram-level sensitivity in full mass range mode, which also allows the potential for finding nontarget pesticides. An alternate GC–MS approach for targeted pesticides in dietary supplements would be to use selected ion monitoring (SIM) with a typical quadrupole mass spectrometer.  

Figure 2: Total ion chromatogram (green) and extracted ion chromatogram of m/z 312 corresponding to the quantitation ion for carfentrazone ethyl (blue). The inset shows the enlarged carfentrazone ethyl peak with a quantitation signal-to-noise value of 105.			

 Overall, the combination of QuEChERS extraction, cSPE cleanup, and GC–TOF-MS used in this method produced good recoveries for most compounds tested (Table I). Although early eluted compounds trended toward slightly lower recoveries, most analytes, including more polar compounds, showed excellent recoveries. The potential for good recoveries of polar pesticides is a major advantage of QuEChERS methods. This difference is due to the use of a polar extraction solvent, acetonitrile rather than petroleum ether (hexanes), which is used in PAM 303, for example. The lower recoveries of early eluted compounds can be due to evaporative loss during concentration steps, due to their higher volatility. Additionally, in the case of planar compounds (for example, the chlorinated aromatics), reduced recoveries can be due to interaction with the graphitized carbon black sorbent used to remove pigments and other matrix compounds, although the planar quality control standard, anthracene, did not show drastic losses during cSPE. Overall, the chromatography and recovery results seen for a broad range of pesticides in dandelion root demonstrate the utility of the QuEChERS approach for dietary supplement testing.  

Table I: Retention time data, recoveries, class, and type of the pesticides used in this work for dandelion root extracts.			

Demonstrated here is a QuEChERS extraction, cartridge SPE cleanup, and GC–TOF-MS approach that helps accomplish the pesticide testing now required for dietary supplements. The basic methodology presented here for dandelion root can be modified for other analytes and matrices and illustrates the advantages of the QuEChERS approach for labs developing cGMP methods. Analytical benefits include reduced interferences and good recoveries, even of polar compounds. Other benefits include an overall savings of both materials and sample preparation time compared to the PAM-based methods, and better expected reproducibility due to the straightforward procedure with fewer manual preparations. 

The authors would like to acknowledge LECO Corporation for the generous use of their instrument. 

Julie Kowalski, Michelle Misselwitz, Jason Thomas, and Jack Cochran

 are with Restek Corporation, Bellefonte, Pennsylvania.

(1) US Food and Drug Administration, Current Good Manufacturing Practice in Manufacturing, Packaging, Labeling, or Holding Operations for Dietary Supplements, Docket No. 1996N-0417 (formerly No. 96N-0417), CFSAN 200441 (2007) 34752.  

US Food and Drug Administration, Pesticide Analytical Manual, Volume I: Multiresidue Methods

(3) M. Anastassiades, S.J. Lehotay, D. Stajnbaher, and F.J. Schenck, 

(4) J.W. Wong, M.S. Wirtz, M.K. Hennessy, F.J. Schenck, A.J. Krynitsky, and S.G. Capar, 

(5) J.W. Wong, M.K. Hennessy, D. G. Hayward, A. J. Krynitsky, I. Cassias, and F.J. Schenck, 

(7) Foods of Plant Origin — Determination of Pesticide Residues Using GC–MS and/or LC–MS/MS Following Acetonitrile Extraction/Partitioning and Clean-up by Dispersive SPE (QuEChERS-method). (EN 15662 Version 2008.)

(8) M. Okihashi, Y. Kitagawa, K. Akutsu, H. Obana, and Y. Tanaka,

Article

The authors evaluate a method for detecting pesticide residues in dandelion root powder.

Evaluation of QuECheRS, Cartridge SPE Cleanup, and Gas Chromatography Time-of-Flight Mass Spectrometry for the Analysis of Pesticides in Dietary Supplements

The previous installment in this series looked at how moderate peak tailing affects resolution and degree of peak overlap. When peaks tail, even moderately, extra time between them is required to attain the same resolution as for nontailing peaks due to the overall increase in peak widths caused by tailing. Even when the same resolution is attained, however, the degree of area overlap between tailing peaks is greater than for nontailing peaks, and more so for the amount of excursion of the first peak into the second. Thus, additional resolution is required for tailing peaks to achieve the same degree of peak area overlap as a pair of nontailing symmetrical peaks.   

The amount of area overlap of one peak into another can be measured as the percentage of a peak's area that lies underneath the adjacent peak, where the peaks are separated with a specific degree of resolution, as in the previous two installments in this series (1,2). In that case, where the peaks were of equal size, resolution of 2.0 was needed to bring the degree of overlap of a first tailing peak into the second down to about the same level seen with symmetrical peaks at resolution of 1.5, at around 0.5%. This was despite the moderate increase in peak separation required to keep the measured resolution the same for tailing peaks as for nontailing ones.  

The degree of peak overlap can be  calculated easily enough with the theoretical peak model being used here, but it isn't a realistic measurement of the effects that less-than-ideal peak separation has on actual area measurement of peaks that are co-eluted in a chromatogram. With nonselective detectors, it isn't possible to measure the two peaks independently, as was done here to obtain the degree of overlap. Even with a mass spectrometer or other selective detector, there often isn't sufficient selectivity to obtain 100% discrimination between overlapping peaks, so questions of the effects of peak overlap on quantitative measurements are not answered by overlap calculations. 

In some cases, mathematical peak deconvolution can be used to tease apart two partially merged peaks, but practically speaking, for routine applications, we have to fall back on the integration parameters available in chromatography data processing systems. To test how well merged tailing peak areas are estimated, I exported a series of synthesized chromatograms with moderately tailing peaks using asymmetries (

) of 1.4 in which the peak resolution was set at 1.0, 1.5, and 2.0. Three common peak integration methods were selected for the estimation of peak areas: a vertical drop from the valley point between the peaks down to the baseline; a tangential skim drawn from the valley out to the down-sloping side of the second peak; and an exponential skim underneath the second peak. Somewhat unrealistically, no noise was added to the generated peaks, but in this case, I was more interested in the measurements themselves and not the degree of uncertainty across multiple measurements. These results show how one particular data processing system performed. Others would vary, but I expect they might produce very similar results. 

Table I enumerates the results of this tailing peak faceoff, and Figure 1 shows what the separations looked like at resolution = 1.0. As was the case with symmetrical peaks (1), increasing peak separation and resolution improved the accuracy of the area measurements significantly. And the influence of the area determination method influenced the measurements in much the same manner. The best results were obtained with a simple vertical drop from the valley between the peaks to the baseline below; the measured values had less than a 4% error in all cases. The exponential skimming technique performed well except for the second, smaller peak when the resolution was less than 1.5. In this case, the error went from about 6% up to nearly 35%. The tangential method delivered a comparably large error of over 8% for the second of two relatively well-separated peaks, and it produced very large errors for the second peak in the cases of the less-resolved peak pairs.  

Table I: Area measurement of partially resolved tailing peaks. Peak 1 is 10Ã larger than peak 2; both have asymmetries of 1.4. The appearance of the peaks at resolution = 1.0 is illustrated in Figure 1, to which the letters (a), (b), and (c) refer. 

Overall, at resolution 1.0, the valley baseline error for the second, smaller peak was actually somewhat less for the unequal-sized tailing peaks than for the symmetrical peaks, due perhaps to the slightly larger separation of the tailing peaks, while errors for the tangential and exponential methods were consistently larger with unequal-sized tailing peaks than for symmetrical ones. This seems to say that the increased separation between tailing peaks that was required to maintain a degree of area overlap originally seen with nontailing peaks is insufficient to reduce peak area integration errors to similar levels in both cases.  

Figure 1: Influence of integration method on the measured areas of unequal sized asymmetrical peaks. Peak pair with Rs = 1.0 (a) Vertical drop from valley to baseline; (b) tangential skim underneath the second peak; (c) exponential skim off the first, larger peak. 

Another observation gave me second thoughts about peak integration in these situations. In the default state, chromatography data handling software has a number of settings that control various automatic thresholds for the onset of exponential skimming versus the vertical drop or other integration methods. These settings were left at their default values for this short study, which caused the automated selection of the area measurement method to change from a vertical drop in the case of resolutions equal to 1.5 and 2.0 to the exponential skim method at resolution of 1.0. The result was a sudden "automatic" decrease in the second peak's area when the resolution dropped off to 1.0, as determined without any intervention, a change that would have resulted in a -34.4% error if the system were allowed to run unobserved. As it was, I forced the use of the vertical drop method to observe its effects. Although I did not explore the effects of changing the relative sizes of the peaks to values other than 1:1 and 10:1, changes in peak size also can trigger unintended shifts in the integration method selected automatically by any data handling system. Therefore, in low-resolution situations, careful examination of the results over the entire range of intended magnitudes of all affected peaks, both analyte and matrix components, must be performed while taking positive control of the integration settings.  

The aforementioned issues with peak area determination and resolution arise partially from the fact that peak resolution does not depend upon the relative sizes of the peaks. Thus, even though the area measurements of two peaks of equal size, tailing or not, can be quite different compared to measurements of two peaks of significantly different sizes, the resolution of the two pairs of peaks is identical: it depends only upon the peak widths measured at base, or on the widths at half-height — as are sometimes easier to measure — converted to base widths.  

A number of measurements related to resolution have been described in the chromatography literature, including the peak capacity, which is the number of additional peaks of a specified resolution that could fit between two well-separated peaks. For example, if two peaks have resolution 

 = 6, then three additional peaks of the same width would fit between them if the added peaks were spaced exactly evenly in-between, for a peak capacity of 3. A similar term, the effective peak number, refers to the peak capacity in the special case where the resolution between added evenly spaced peaks equals 1.0. The separation number, or Trennzahl, uses the peak widths at half-height to calculate the number of peaks that would fit between our well-separated pair with 

= 1.177 and usually is applied to consecutive members of a homologous series of compounds such as 

-hydrocarbons in retention index calculations. Details of these measurements can be found in reference 3 as well as in many other chromatography textbooks. 

Another resolution-related measurement is of interest to the current discussion. As stated earlier, none of these calculations takes into account the effect of differing peak sizes on peak area overlap and integration errors. The discrimination factor, however, does incorporate peak height and it can be a more meaningful metric for evaluation of these concerns (4). The discrimination factor is related to the separation power originally described by Kaiser (5, see also reference 3 for more information) and is calculated graphically as shown in Figure 2 from the relative heights of the smaller peak and the valley above the baseline.  

In Figure 2a, the discrimination factor is 

 = 0.72 in the case of equal-sized symmetrical peaks and in  Figure 2b with unequal-sized symmetrical peaks, 

 = 0.30. In both cases the resolution equals 1.0, but the smaller discrimination factor of the peak pair with a 10:1 height ratio in Figure 2b reflects more accurately the reduced degree of separation available when the peak sizes are significantly different.  

Figure 2: Measurement of the discrimination factor: (a) peaks of equal height; (b) peaks of unequal height. 

In the case of tailing peaks of equal heights at resolution 1.0 (1), the discrimination factor also equals 0.72. The discrimination factor does not distinguish between tailing and symmetrical peaks with the same size and resolution, but remember that the tailing peak pair are separated in time by a few more seconds to keep the same resolution as the nontailing peak pair. In the case of the unequal size tailing peak pair with a 10:1 size ratio (Figure 1a), the discrimination factor drops down to 0.04. Compared to a value of 0.30 for the unequal-sized symmetrical peak pair, the effect of peak tailing on the discrimination factor is dramatic, perhaps even overstated. Nevertheless, the discrimination factor does provide a useful indication of how much a larger peak will affect the quality of separation with an adjacent peak both in terms of the peak size and the presence of peak tailing. 

The effects of peak size and peak tailing on the degree of area overlap and measured peak areas are significant. Changes in peak separation, resolution, and symmetry strongly affect the accuracy of peak area measurements. Conventional resolution measurements do not take into account the effects of disparities in peak size, nor do they completely accommodate the effects of peak tailing. The peak discrimination factor provides some indication of how well a given separation discriminates adjacent peaks from each other and does take into account the effects of peak size and asymmetry. 

Careful evaluation of the data processing and area measurements of disparate, partially merged, or tailing peaks can identify potential pitfalls that can be imposed by changes in concentration, peak retention time drift, and asymmetry incurred as chromatography components age and are replaced. 

 "GC Connections" editor John V. Hinshaw is senior Research Scientist at Serveron Corp., Hillsboro, Oregon, and a member of 

's editorial advisory board. Direct correspondence about this column to 

, Advanstar Communications, 485 Rt. 1 S, Bldg F, 1st Floor, Iselin, NJ 08830, or contact the author via e-mail: 

Basic Relationships in Gas Chromatography

 (Advanstar, Cleveland, 1994), pp. 75–83. 

 (Akademische Verlagsgesellschart, Leipzig, 1960), p. 33. 

GC Connections

John Hinshaw concludes his series on the effects of peak tailing.

When Peaks Collide, Part III

Polyphenols represent one of the most studied classes of natural compounds. Interest in these ubiquitous secondary metabolites, in recent times in particular, stems from their physiological contribution to the human diet and the numerous health benefits associated with their consumption. In addition, polyphenols are responsible for a range of organoleptic properties, including color, astringency, bitterness and mouth-feel, as well as ageing properties (1).  

Clearly, the analysis of these compounds is of significant importance in various fields, particularly in the food and beverage and nutraceutical industries. However, the accurate analytical determination of phenolic compounds is hampered by the sheer complexity of the numerous chemical classes and species classified as polyphenols. These broadly include phenolic acids (hydroxybenzoic and cinnamic acids) and flavonoids, the latter class being subdivided into flavonols, flavanols (and their polymeric derivatives, the proanthocyanidins) and anthocyanins (anthocyanin-glycosides) (Figure 1).  

Figure 1: General structures of the principal classes of phenolics. R is commonly H, OH or OCH3, while R1 may be OH, O-glycosyl, or O-acyl-glycosyl.			

The investigation of polyphenols is therefore closely related to the development of suitable analytical techniques for their identification and quantification. High performance liquid chromatography (HPLC), and reversed-phase LC in particular, is most often used for the routine analysis of phenolics, while in recent years the hyphenation of HPLC to mass spectrometry (MS) has become an indispensable tool for the detailed investigation of these compounds (2). 

The resolving power of chromatographic methods is commonly measured in terms of peak capacity: the number of peaks that can theoretically be resolved with unit resolution using a given separation method. The peak capacity (

) for gradient separations can be calculated according to the following equation (3):   

 the average (4σ) peak width. Common gradient reversed-phase LC methods typically provide peak capacities in the order of 100–150 (for ~1 h analysis times). Considering that the peak capacity should significantly exceed the number of randomly distributed analytes in a given sample to provide a high likelihood of separation (4), and the fact that typical phenolic extracts contain in excess of 100 species, the limitations of HPLC are evident.  

Importantly, in recent years a number of significant developments in HPLC have contributed to deliver improved performance. These include the use of ultrahigh-pressure liquid chromatography (UHPLC), new developments in stationary phase morphologies (monolithic columns and superficially porous particles), high temperature liquid chromatography (HTLC) and multidimensional LC separations. In this contribution we discuss the beneficial application of some of these new technologies to improve the analysis of polyphenols in a variety of samples.  

Cocoa beans, wine, blueberries and green tea were purchased from a local supermarket. Epicatechin, catechin and uracil standards were from Sigma-Aldrich (Atlasville, South Africa), and anthocyanin standards from Polyphenols Laboratories (Sandnes, Norway). Dimeric procyanidins were isolated by hydrophilic interaction chromatography (HILIC) fractionation from a cocoa extract. Mobile phases were prepared from HPLC-grade solvents (Sigma) and Milli-Q deionized water (Millipore, Milford, Massachusetts). 

Extracts of green tea and cocoa were prepared as reported previously (5,6). For extraction of blueberry anthocyanins, 6.681 g of blueberries was frozen and ground under liquid nitrogen and extracted with methanol–water–formic acid (60:37:3), followed by evaporative preconcentration. Filtered wine samples were directly injected. 

HPLC–UV analyses were performed on an Alliance 2690 HPLC equipped with a photodiode-array (PDA) detector (Waters, Milford, Massachusetts). UHPLC experiments were performed on an Acquity UPLC system with PDA detector (Waters). HPLC–ESI-MS analyses were performed on a UPLC system coupled to a Waters Ultima Q-TOF mass spectrometer operated in positive ionization (PI) mode (6). Blueberry anthocyanins were tentatively identified by MS and MS

  spectra and comparison of retention times with literature reports (7). The following columns were used: XBridge C18 (250 mm × 4.6 mm, 5 μm) and Acquity BEH C18 (50 and 100 mm × 2.1 mm, 1.7 μm, both Waters), Zorbax SB C18 (50 mm × 4.6 mm, 1.8 μm, Agilent, Waldbronn, Germany) and Develosil Diol-100 (250 mm × 1 mm, 5 μm, Nomura Chemical, Japan). 

Cocoa procyanidins and green tea flavonols: 

HPLC analyses at 25 °C were performed using 0.1% formic acid (A) and acetonitrile (B) at a flow rate of 1 mL/min: 0–0.63 min (2% B), 0.63–27 min (2–18% B), 27–45 min (18–25% B), 45–51.75 min (25–100% B), 51.75–56.25 min (100% B). UHPLC analyses were performed on a 100 mm Acquity column at 50 °C using the following gradient at 0.3 mL/min: 0–0.17 min (2% B), 0.17–7.4 min (2–18% B), 7.4–12.48 min (18–25% B), 12.48–14.53 min (25–100% B), 14.53–15.6 min (100% B). UV detection was performed at 280 and 370 nm for procyanidins and flavonols, respectively. 

Mobile phases consisted of 7.5% formic acid (A) and acetonitrile (B). HPLC analyses were performed on the XBridge column at 25 °C and 1 mL/min using the following gradient: 0–1 min (1% B), 1–12 min (1–13.5% B), 12–24 min (13.5–23.5% B), 24–28 min (23.5–28.5% B), 28–35 min (28.5% B). UHPLC analyses were performed on two coupled 100-mm Acquity columns at 50 °C and 0.06 mL/min as follows: 0–3 min (1% B), 3–34 min (1–13.5% B), 34–67 min (13.5–23.5% B), 67–78 min (23.5–28.5% B), 78–98 min (28.5% B). UV detection was performed at 500 nm.  

Analyses were performed on the XBridge column using the same mobile phases as for wine anthocyanins. A flow rate of 1 mL/min was used throughout with the following gradients: 0–50 min (1–25% B), 50–55 min (25–40% B) at 25 °C; and 0–50 min (1–20% B), 50–55 min (25–40% B) at 70 °C. A postcolumn split (1:4) was used before eluent introduction into the MS. 

Conditions were as reported elsewhere (5,6). 

For procyanidins, experiments were performed on the XBridge column at 25 °C and on a 50-mm Acquity column at 25 °C and 50 °C, and for anthocyanins on the XBridge column at 25 °C and 70 °C. Mobile phases were as specified for each of these samples above, with the composition varied to ensure similar retention factors at all temperatures. Uracil was used as unretained marker, and all experiments were performed in duplicate. van Deemter curves and kinetic plots were constructed according to reference 8. 

Since the inception of HPLC, there has been a continuous drive to improve the performance of the technique. The most widely practiced approach is to reduce the particle size (

) of column packing. The benefit of this approach is illustrated by the van Deemter curve, which plots chromatographic efficiency, in terms of plate height (

 is the number of theoretical plates) as a function of linear velocity of the mobile phase, 

. The optimal linear velocity, uopt, is that mobile phase velocity (flow rate) which delivers the minimum plate height (

) and therefore maximum efficiency. The effect of 

 on the van Deemter curve is demonstrated in Figure 2a for a dimeric procyanidin analyzed on 5- and 1.7-μm columns. It is clear that uopt increases and 

, implying that faster and more efficient analyses are possible on the smaller phase for equal column lengths.  

Figure 2: (a) Comparison of van Deemter curves for procyanidin dimer on 5- and 1.7-Î¼m columns at 25 Â°C and 50 Â°C. (b) Corresponding kinetic plots of efficiency versus  analysis time. 			

In fact, this picture is not entirely accurate, since decreasing 

 leads to a very important decrease in column permeability. Pressure constraints ultimately limit the column length and therefore efficiency attainable, so that very high efficiencies are easier to achieve for larger particle sizes (a result of the higher permeability of the larger 

 phases). Note though that the price to pay for such high efficiencies is a very significant increase in analysis time. 

This dilemma has been partially overcome by the availability of a new generation of high-pressure instrumentation. In Figure 2b the isocratic efficiency as a function of total analysis time for the procyanidin dimer is compared for the same 5- and 1.7-μm columns, the latter operated at pressures up to 1000 bar. Note the significant decrease in analysis time achieved on the UHPLC column for an efficiency of 25,000 plates (equal to that provided by a conventional 250-mm, 5-μm column). Moreover, UHPLC clearly offers faster analysis for efficiencies up to 

It is relatively straightforward to practically exploit the benefits of UHPLC because HPLC methods may nowadays easily be adapted to UHPLC column formats and instrumentation, essentially providing the same separation mechanism. To date this approach has received comparatively little attention for polyphenol analysis (9,10), with the focus primarily on very fast analysis on short columns, essentially providing the same efficiency as conventional HPLC columns. 

Increasing the temperature of HPLC separations provides an additional means of improving performance.11 This is primarily a result of faster analyte diffusion in the mobile phase at higher temperatures. The effect of this on plate height behavior is demonstrated once again for the dimeric procyanidins on the 1.7-μm column in Figure 2: While the maximum column efficiency (= 

) remains roughly unchanged, the optimal linear velocity increases and efficiency-loss at high flow rates is reduced. At higher analysis temperature the flow rate should be increased to operate close to uopt, in this manner providing the same efficiency faster. Note that even though the mobile phase viscosity is inversely proportional to temperature, the requirement of operating at higher flow rates means that the overall pressure remains roughly constant. 

In fact, elevated temperature offers a complementary means of further increasing the speed of UHPLC separations for efficiencies up to 

 ≈ 80,000. The benefits of this approach are illustrated in Figure 3, where conventional HPLC analysis of cocoa procyanidins and green tea flavonols are compared to UHPLC analysis at 50 °C. This approach provides similar separation in a third of the analysis time, demonstrating obvious advantages for routine analysis of polyphenols.  

Figure 3: Comparison of HPLC analysis of cocoa procyanidins and green tea flavonols at 25 Â°C on a 5-Î¼m XBridge column (a) and (b) respectively, and at 50 Â°C on a 1.7-Î¼m column (c) and (d) respectively. Detection was performed at 280 nm and 370 nm for procyanidins and flavonols, respectively. ÎPmax = 193 and 505 bar for HPLC and UHPLC analyses.			

For anthocyanins (anthocyanidin-glycosides, Figure 1), analysis temperature plays a particularly important role. These compounds are known to exist in solution in a number of different chemical forms (12). On-column interconversion between these species causes significant band broadening (13), leading to the relatively poor efficiency commonly observed for HPLC analyses of anthocyanins. Higher analysis temperatures speed up the interconversion reactions — leading to less band-broadening and improved efficiency. This is demonstrated for the van Deemter curves of four common anthocyanins in Figure 4. At conventional flow rates (that is, 1 mL/min on 4.6-mm i.d. columns), an improvement in efficiency of 25–130% is achieved at 70 °C compared with 25 °C. In this manner elevated temperature may be used to improve the efficiency of conventional gradient reversed-phase LC analysis of anthocyanins, as demonstrated for blueberry anthocyanins in Figure 5. Also note that temperature may affect the selectivity of the separation, as is evident from the relative elution order of some anthocyanins in Figure 5.  

Figure 4: van Deemter plots for four anthocyanins on a 5-Î¼m XBridge column at (a) 25 Â°C and (b) 70 Â°C and (c) corresponding kinetic plots of efficiency versus retention time for malvidin-3-O-glucoside 			

The previous examples demonstrate how analysis speed for polyphenol analysis may be increased (for the same efficiency) or efficiency increased for the same analysis time. Another approach especially suited for complex phenolic fractions, is to exploit the benefits of UHPLC and elevated temperature for high-efficiency analysis. This is demonstrated for the analysis of red wine anthocyanins on a 200-mm, 1.7-μm column operated at 50 °C in Figure 6 (14). This set-up provides isocratic efficiencies of up to 40,000 plates, and gradient analysis delivers much improved resolution for an acceptable analysis time of 90 min. Note that this analysis is performed at pressures below 300 bar, implying that conventional instrumentation may be used (provided system delay and void volumes are optimized for such analyses).  

Figure 5: Extracted ion chromatograms for the gradient HPLCâMS analysis of blueberry anthocyanins on a 5-Î¼m XBridge column: (a) at 25 Â°C, and (b) at 70 Â°C, both at 1 mL/min. Cy = cyanidin, Dp = delphinidin, Pt = petunidin, Pe = peonidin and Mv = malvidin.			

However, caution should be exercised in the application of elevated temperature for phenolic analysis. As is well documented in literature, anthocyanins in particular are readily degraded at relatively mild temperatures. Fortunately this aspect may be evaluated quantitively. Using the approach of Thompson and colleagues (15) it at can be demonstrated that analysis times of less than 2 h at 70 °C do not lead to on-column degradation of anthocyanins (13).  

Figure 6: Comparison of gradient reversed-phase LC analysis of red wine anthocyanins on a (a) 5-Î¼m XBridge column at 25 Â°C and 1 mL/min and a (b) 1.7-Î¼m Acquity column at 50 Â°C and 0.06 mL/min (14). UV detection at 500 nm. ÎPmax = 190 and 295 bar for HPLC and UHPLC analyses.			

Finally, in the context of unidimensional HPLC analysis, it should be mentioned that a number of additional developments in column technology in recent years hold some promise for HPLC analysis of polyphenols. These include the use of monolithic (16) and superficially porous particles (17). However, limited application of these column formats to polyphenol determination has been reported to date (18).  

Comprehensive two-dimensional liquid chromatography (LC×LC)

The challenges associated with HPLC analysis of phenolics may be demonstrated using an example: procyanidins (Figure 1) consist of a mixture of monomeric, oligomeric and polymeric compounds composed of flavanol building blocks. As the degree of polymerization (DP) increases, the number of possible isomers increases exponentially (a result of the possible combinations of monomers as well as different inter-flavanol bonds). For example, 48 isomers of dimeric procyanidins containing catechin and epicatechin building blocks are possible (19). Considering that optimized high-efficiency HPLC methods provide peak capacities of maximally 400, no unidimensional HPLC method is capable of providing complete separation of higher molecular weight (MW) procyanidins in a single analysis.  

Comprehensive two-dimensional HPLC (LC×LC) is a promising method for the separation of such complex samples. LC×LC delivers peak capacities an order of magnitude or more than unidimensional LC. In LC×LC, two separation methods are combined in such a manner that the separation in each dimension is retained. As a consequence, the overall peak capacity in a two-dimensional space is (ideally) the product of peak capacities in each dimension. Admittedly, this ideal is only achieved if certain criteria are met, most importantly that both separations are based on different (orthogonal) mechanisms.  

Comprehensive 2-D LC may be performed in off-line, on-line or stop-flow modes (19). In on-line LC×LC, fractions from the first dimension are transferred in real time to the second dimension column, usually using multi-port two-position switching valves. This approach provides relatively fast (approximately equal to 1-D HPLC methods) separations, although resolution in the second dimension is sacrificed to meet sampling rate requirements. Off-line LC×LC, by contrast, results in much longer analysis times, but since no constraints are placed on the second-dimension separation much higher peak capacities may be obtained. (Stop-flow LC×LC is less often applied, and provides peak capacities and analysis times in between the off-line and on-line approaches). 

In fact, the application of on-line LC×LC using a variety of separation modes has been demonstrated for diverse phenolics in wine, beer and herb samples (18,21,22). In our group, we have investigated the off-line coupling of HILIC and reversed-phase LC to improve procyanidin determination (5). Off-line coupling was chosen to obtain the maximal practical peak capacity through optimal second-dimension analyses. In HILIC, procyanidins are separated according to polarity, which effectively provides molecular weight separation based on the number of hydroxyls (individual isomers are co-eluted) (23). Reversed-phase LC, in contrast, provides isomeric separation according to hydrophobicity — although complete separation of isomers in not possible in a single analysis. HILIC and reversed-phase LC therefore provide highly orthogonal separations. The off-line comprehensive HILIC×reversed-phase LC analysis of cocoa procyanidins is illustrated in Figure 7, where it is evident that the combination of MW (HILIC, 

-axis) information delivers a complete picture of the oligomeric procyanidin content of this extract. The practical peak capacity for this separation (according to reference 24) is ≈3500.  

Figure 7: HILICÃreversed-phase LC analysis of cocoa procyanidins. Numbers correspond to degree of polymerization of procyanidins, A indicates A-type procyanidins. Refer to reference 5 for further details.			

The same approach may also be used beneficially for the analysis of other classes of phenolic compounds. This is illustrated in Figure 8 for flavonol glycosides in green tea (practical peak capacity 2500). A very structured elution pattern reminiscent of comprehensive GC is evident from this figure: retention in HILIC increases with the number of attached sugars, while the different flavonol moieties as well as individual mono-, di-, and tri-glycosides are resolved in the reversed-phase LC dimension. It may be concluded that HILIC×reversed-phase LC, and LC×LC in general, holds significant promise for the in-depth investigation of complex phenolic fractions such as encountered in numerous natural products.  

Figure 8: HILICÃreversed-phase LC analysis of green tea flavonols. Data taken from reference 6. Myr = myricetin, Q = querectin, K = kaempferol, gal = galactoside, glc = glucoside, rha = rhamnoside, xyl = xyloside, rut = rutinoside and hex = unkown hexose.			

The analysis of polyphenols represents a severe analytical challenge, and important advances in natural phenolic research hinge on the development of suitable separation strategies. Fortunately, recent developments in HPLC show promise for improving polyphenol analysis. 

A summarized comparison of conventional HPLC, UHPLC and elevated temperature LC and LC×LC analyses of different phenolic compounds is presented in Table I. These examples demonstrate that the combination of small 

 columns and temperature primarily delivers improved peak capacity production rates. This advantage can be practically exploited in one of two ways: to achieve similar peak capacities to conventional HPLC 3–4× faster (for example, cocoa procyanidins, green tea flavonols); and to improve the peak capacity for a given analysis time (for example, blueberry anthocyanins). These techniques show particular promise in improving current routine HPLC methods for polyphenol determination. In addition, application of UHPLC columns operated at elevated temperatures allows improved total peak capacity for complex phenolic fractions (for example, red wine anthocyanins).  

Table I: Comparison of peak capacities and peak capacity production rates for conventional HPLC, UHPLC, elevated temperature LC and LCÃLC analysis of phenolic compounds in various samples.			

For very demanding separations, off-line comprehensive 2-D LC offers greatly improved separation power compared with HPLC (>20× higher peak capacities) at similar peak capacity production rates. The choice of the most suitable separation methodology will therefore depend on the requirements for a particular phenolic sample in terms of speed, instrumental simplicity or resolving power.  

It should be noted that state-of-the-art mass spectrometric (MS) techniques such as high resolution MS and tandem MS methods, not addressed in the current contribution, represent a complimentary strategy for improving polyphenol analysis. In fact, MS offers an alternative 'separation' dimension (in terms of 

), increased selectivity for target analysis and a powerful structural elucidation tool. Combined with advanced chromatographic separation, LC–MS will prove an even more powerful tool for detailed phenolic determination.     

Clearly, the judicious application of recent advances in HPLC demonstrably offers significant benefits for phenolic analyses, and shows promise for the in-depth investigation of the phenolic compounds found in numerous natural products — an endeavor that, to date, has been hampered by the lack of suitable separation methods.  

This article highlights a number of recent developments in HPLC for polyphenol analysis.

Improving HPLC Separation of Polyphenols

Chromatographers are always interested in higher efficiency. This is motivated by the resolution equation, which shows that increases in column efficiency always result in improved resolution. Chromatographic efficiency is affected by a large number of experimental variables and its optimization can be achieved in many different ways, depending upon how many variables one is willing to adjust. These include pressure, temperature, particle size, column length, and eluent velocity. In the early days of high performance liquid chromatography (HPLC), the selection of column formats (particle size, type, and column diameters) was rather limited and thus, optimization often was done by adjusting operational variables such as eluent velocity, column temperature, and operating pressure. Nowadays the selection of column formats is substantially wider and one can find a number of particle sizes between 1.7 μm and 5 μm, and numerous column lengths are achievable by coupling columns in series. This makes optimization of these nominally "discrete" variables possible (that is, particle size and column length).   

 Optimization of performance (meaning efficiency versus time) in HPLC has been studied for decades. Van Deemter was clearly among the first who understood how plate count on a given column could be optimized by varying the eluent velocity (1). When column length is also allowed to vary, one can use the Poppe plot (2) or kinetic plot (3,4) techniques that consider pressure and time constraints as part of the optimization process. Ultimately, one also can vary particle size, and this optimization has been studied by several LC pioneers including Halasz (5), Knox (6), Horvath (7), and Guiochon (8). We recently summarized these distinct optimization schemes and emphasized the different chromatographic results that these approaches entail (9). The purpose of this column is to propose a simple, stepwise optimization procedure that is solidly founded and to demonstrate its practical utility by applying it to the development of an ultrafast isocratic separation for pharmaceutical applications. 

As the march toward more-efficient separations continues, so does the demand for faster separations without sacrificing efficiency. This has been stimulated by the present and persistent need to deal with more and more samples in the same or less time. Clearly "fast" means different things in different applications. For example, "fast" can be a 10-min separation for organic impurities analysis in pharmaceuticals, or it can be a 1-min separation for cleaning analysis in drug manufacturing. In this column, we focus on the implementation of fast LC in pharmaceutical drug product dissolution testing. In such a test, one is primarily interested in quantifying the active pharmaceutical ingredient (API) released from dosage forms, thus, only moderate plate counts are required; here, analysis speed is more important than higher efficiency. Dissolution samples often are analyzed by two techniques, direct UV or HPLC–UV, in almost equal proportions as shown by an online survey of dissolution studies (10). Table I compares these two techniques. A thorough review of the use of HPLC in dissolution studies can be found in a chapter written by Wang and Gray (11). While UV–vis spectroscopy is cheaper and simpler, HPLC offers many advantages, such as better specificity and a greater linear dynamic range. HPLC is also more versatile especially for early drug development when different formulations and strengths are screened. The main disadvantage of traditional HPLC is its relatively slower speed (about 3–5 times slower than UV analysis). Clearly, increasing the speed of HPLC makes it more attractive in this context.   

 This work focuses on using theoretical guidance to develop better ultrafast isocratic separations (approximately 30 s) for dissolution testing, with the goal of making the speed of HPLC competitive with UV analysis. In addition, ultrafast HPLC is critical to the implementation of automated dissolution testing systems with online HPLC analysis (11). There are many approaches to speed up HPLC analysis. One popular way is to use monolithic columns at very high eluent velocities; sub-1-min separations have been achieved (12,13). However, our recent review of different technologies for enhancing speed suggests that the most effective way of achieving very high speed is the 

 use of higher temperatures, higher pressures, and smaller particles (14). In addition, substantially lower solvent consumption can be achieved with narrow-bore columns packed with small particles compared to monolithic columns. Therefore, in this column, we focus on optimizing an ultrafast separation under high temperature and high pressure conditions.  

Table I: Criteria of UV and traditional HPLC analysis for dissolution test* 

The goal of most separation optimizations is to achieve either the highest number of plates in a given analysis time or to achieve a given number of plates in the shortest time. For simplicity, here we will discuss only optimization with the goal of achieving the highest plate count in a given analysis time (that is, a fixed column dead time, 

), and refer interested readers to the extensive supporting literature on optimization for more detail (5–9,14). First, we briefly review three different optimization schemes in order of increasing complexity and then propose a simple, practical, stepwise optimization procedure. 

 In this case, one has already chosen a particle size and column length. The only variable left to optimize is velocity. Here, the van Deemter equation (1) is used to calculate the eluent velocity that gives the minimal plate height, provided that this velocity (and flow rate) can be reached within the pressure limit and flow limit of the instrument. Equations 1 and 2 give the van Deemter equation and the optimal linear velocity, respectively, where 

 is particle size, and A, B, and C are van Deemter equation coefficients. It should be noted that throughout this column, it is assumed that the plate height is related to velocity through the van Deemter equation. This is always a rather good approximation as long as one does not work far from the minimum in the plate height curve.  

However, the problem of operating at the van Deemter optimum velocity specified by equation 2 is that one frequently cannot achieve the separation in the desired time (using 

 as a proxy for the analysis time) because column length is already chosen. To achieve the target 

, one needs to operate at the velocity given by equation 3, where 

 are the total and interstitial porosities of the column, respectively (typically about 0.55 and 0.38), and λ is the ratio of the porosities.  

The resulting plate count obtained at this velocity can be calculated easily by inserting equation 3 into equation 1. This plate count is always lower than the plate count calculated based upon the van Deemter optimum velocity; however, it is also the more realistic estimate. 

 In this case, one has chosen a particle size but needs to optimize both column length and velocity. This is precisely the optimization that is dealt with by the Poppe and kinetic plot techniques. For a given analysis time, there is exactly one optimal combination of column length and velocity that will produce the maximum plate count. Equations 4 and 5 give the optimal column length and velocity; 

 is the maximum operating pressure, η is mobile phase viscosity, and 

 is the column dead time. It is important to note that the optimal velocity as described here is almost 

 the same as the van Deemter optimal velocity.  

 In this case, one simultaneously optimizes the particle size, column length, and eluent velocity. The solution to this optimization is that one must work at the van Deemter optimum but then must also use the particle size, column length, and velocity that are dictated by equations 6–8. This scenario often is referred to as the Knox–Saleem limit (6), which is the absolute best separation one can do assuming the optimal particle size and column length are available and the optimal velocity can be delivered by the pumping system.  

 It is easy to show that at a given analysis time the three-parameter optimization gives the highest maximum plates followed by the two-parameter and then the one-parameter optimization. Table II gives examples of the optimal conditions and the resulting predicted plate counts for the three scenarios when a separation with 4-s dead time is needed, which corresponds to a total run time of about 30 s (for example, retention time is 24 s with a 

 of 5). For one-parameter optimization where a 30-mm-long column packed with 1.8-μm particles is preselected, the predicted plate count is 49% lower than that predicted using the three-parameter optimization, and the predicted optimum velocity results in an operating pressure much lower than the instrument maximum. With the two-parameter optimization, a longer column is needed (53 mm) to improve the plate count, but this value is still 29% lower than that predicted for the three-parameter case. When three-parameter optimization is done, a 29-mm-long column packed with 1.0-μm particles is needed but the calculation predicts that almost 15,000 plates can be obtained with a column dead time of only 4 s.  

Table II: Comparison of three optimization schemes for a t0 = 4 s separation* 

The comparison in Table II is shown in the Poppe plot given in Figure 1. The 

-axis is plate count as a measure of efficiency, and the 

-axis is "plate time" as a measure of speed. We refer interested readers to the review articles by Desmet (3) and Neue (4). The blue line represents solutions to the three-parameter optimization conducted at 40 °C and 1000 bar. The values on this line also are known as the "Knox–Saleem limit." The red curve shows the results of two-parameter optimization for 1.8-μm particles packed in columns operated at 40 °C and 1000 bar; this curve also is known as the Poppe curve. Obviously, the three-parameter optimum (blue diamond) and the two-parameter optimum (red square) fall on the "Knox–Saleem limit" line and the Poppe curve, respectively. The corresponding plate counts (

-axis values) for these three points on the 

 = 4 s line show that the three-parameter optimization yields the most plates, followed by two-parameter optimization, and then by one-parameter optimization (green circle).  

Figure 1: Graphical comparison of the three optimization schemes of Table II. 

However, it is also clear from Table II that the optimum column length in the two-parameter optimization and both the optimum column length and particle size in the three-parameter optimization are not commercially available. In reality, one must make compromises and choose the closest available columns (15,16). One needs to make such decisions wisely so that the deviation from the true optimum does not cause a large loss in plate count. Based upon this thinking, we propose a simple stepwise optimization procedure (see Table III) to maximize the plate count in a given analysis time (

) at a predetermined operating pressure and temperature. We believe this approach will allow one to arrive quickly at the optimum conditions that can be achieved with existing column technologies. In this column, we show an application of this approach to develop an ultrafast isocratic separation for a dissolution test of ciprofloxacin extended release tablets.  

Table III: Proposed stepwise optimization procedure for isocratic separations 

 Ciprofloxacin extended-release 500-mg tablets were chosen because a range of sample concentrations can be obtained at different time points. Dissolution was conducted in a USP apparatus 2 (Distek model 2100, Distek, Inc., North Brunswick, New Jersey) at 50 rpm, 37.0 ± 0.5 °C, using 900 mL of 0.1 N hydrochloric acid as the dissolution medium. Samples (10 mL) were taken at 15, 30, 60, 90, and 120 min and filtered through 2-μm glass fiber filters before LC analysis. For UV analysis, a 10-mm flow cell was used and samples were diluted 100-fold before absorbance measurement at 276 nm. A calibration standard was made from a ciprofloxacin hydrochloride reference standard from USP. 

Chromatographic experiments were conducted on an Agilent 1290 Infinity LC system controlled by Chemstation software (Agilent Technologies, Santa Clara, Calfornia). The 1290 was equipped with a high-pressure mixing binary pump, autosampler, column thermostat, and a photodiode-array detector. To achieve maximal performance in ultrafast separation, it is critical to minimize extracolumn broadening. A 220 mm × 80 μm capillary stainless steel tubing was used between the injection valve and a 1.6-μL microheater in the column thermostat. A 300 mm × 75 μm PEEKSil capillary was used to connect the column to the detector. A prototype 800-nL detector flow cell with a 10-mm pathlength was used with a 160-Hz data acquisition rate. These modifications decreased the extracolumn variance by up to 40% compared to the standard configuration. Separations were conducted at 80 °C in an acidic mobile phase containing 0.1% trifluoroacetic acid. The columns used were a ZORBAX Rapid Resolution HD SB-C18 column (75 × 2.1 mm, 1.8 μm, Agilent Technologies) and a Poroshell 120 SB-C18 column (75 × 2.1 mm, 2.7 μm, Agilent Technologies). Both have StableBond bonded phase chemistry known to be stable at low pH and higher temperature operating conditions. 

Quality by design (QbD) is a hot topic in the pharmaceutical industry (17). A key element of applying QbD in analytical method development is to first develop the method performance criteria and then optimize the method to meet those requirements (18). In Table IV, we summarize the method performance criteria and method operational intent of the ultrafast separation for dissolution testing. We then develop the optimal method by using the stepwise procedure outlined in Table III.  

Table IV: Method performance criteria and operational intent for ultrafast dissolution testing 

 Equations 6–8 clearly suggest that ultrafast separations require small particles, higher pressures, and higher temperatures (9). Therefore, we operated the columns at 80 °C and at 90% of the maximum pressure allowed by either the instrument or the column hardware. In addition to fully porous particles, we also evaluated 2.7-μm superficially porous particles that had been shown to generate performance close to that of sub-2-μm fully porous particles but at lower pressure (19–22). 

Flow studies were first conducted on the RRHD SB-C18 1.8-μm column and the Poroshell 120 SB-C18 2.7-μm column using butyrophenone (

 = 5) to obtain the van Deemter coefficients. Flow resistance also was calculated from the pressure drop measured during the flow studies. Column porosities were determined by inverse size-exclusion chromatography (23). All these parameters are required to use equations 3–8 to calculate the optimal particle size, column length, and velocity. Tables V and VI give the calculated optimal conditions following the proposed three-step procedure for the RRHD column at 1100 bar and the Poroshell 120 column at 550 bar, respectively.  

Table V: Stepwise optimization for t0 = 4 s separation on ZORBAX RRHD SB-C18 at 1100 bar and 80 Â°C* 

The calculation results in step 1 suggest that the optimum particle sizes for 

 = 4 s separation are 0.9 μm for the RRHD column and 1.2 μm for the Poroshell 120 column. Clearly, one has to make compromises and choose the closest available particle sizes, which are 1.8 μm for RRHD and 2.7 μm for Poroshell 120. Thus, the step 2 optimum deviates from the step 1 optimum; 36% and 41% lower plate counts are predicted for RRHD and Poroshell 120 compared to the Knox–Saleem limit. Finally, the optimum column lengths in step 2 are not available, so another compromise must be made. Interestingly, a 75-mm-long column is needed in step 3 for both RRHD and Poroshell 120. An important conclusion drawn from this is that 

smaller particles do not always mean shorter columns for fast analysis

. The optimum column format strongly depends upon many variables, including pressure, temperature, analysis time, and the kinetic characteristics of particles (that is, the van Deemter coefficients). In summary, the calculation predicts that ultrafast separations satisfying the criteria in Table IV can be achieved on both RRHD and Poroshell 120 particles although RRHD offers higher plate count.  

Table VI: Stepwise optimization for t0 = 4 s separation on Poroshell 120 SB-C18 at 550 bar and 80 Â°C* 

The guest authors describe an optimization procedure to achieve the highest plate count in a given analysis time.

LCGC North America-11-01-2010

UHPLC Tips and Techniques

Evaluation of QuECheRS, Cartridge SPE Cleanup, and Gas Chromatography Time-of-Flight Mass Spectrometry for the Analysis of Pesticides in Dietary Supplements

When Peaks Collide, Part III

Improving HPLC Separation of Polyphenols

A Simple Approach to Performance Optimization in HPLC and Its Application in Ultrafast Separation Development

Evolution of Chromatography: One Scientist's 51-Year Journey

Vol 28 No 11 LCGC North America November 2010 Regular Issue PDF