General research or specific experimental protocols involving trace and ultratrace chemical analysis and detection is an extremely challenging and daunting task for scientists. As well as the inherent chemical background that is present when a sample is collected, numerous chemical analyses and methods rely on a pure solvent for experimental procedures. For example, many chemical extraction approaches use a solvent to extract target chemicals of interest from a substrate (for example, soil) or collection sorbent. After the extraction step, an evaporation–condensation–concentrating step is performed to reduce the extraction volume to improve detection of trace components. Unfortunately, this volume reduction step will also concentrate many of the impurities present within a solvent and can further confound research efforts. Many of the solvent artifacts introduced into a sample are from chemical impurities that are not completely removed by standard manufacturing processes or techniques for the solvent, as well as the glassware and storage vials used. Regardless of the source, minimizing and understanding these chemical artifacts is critical for trace-level detection and is crucial for unambiguous analytical results. 

Acetone is one of the most widely used solvents in the laboratory environment. A common chemical production method for acetone is through the cumene process (1). In the cumene process benzene is alkylated with propene resulting in cumene (isopropylbenzene), which is subsequently oxidized to give phenol and acetone. A wide variety of impurities may exist in the final crude acetone fraction (for example, aliphatic aldehydes, olefins and carbonyl impurities, and more specifically diacetone alcohol, mesityl oxide, and hydroxyacetone), which is subsequently purified to obtain the quoted purity. After the acetone has been obtained in the laboratory, further use or storage may form additional chemical artifacts. Hunchak and Suffet (2) and Lafleur and Pangaro (3) examined acetone–hexane artifacts produced during a Soxhlet extraction process. The chemical artifacts produced through the various procedures and reactions include diacetone alcohol and mesityl oxide as well as phorone, mesitylene, and isophorone (2,3). 

The work presented here examines the purity of various acetone sources using comprehensive gas chromatography and mass spectrometry (GC×GC–MS) for broad spectrum chemical analysis. The purpose of this work is to characterize the impurities present in a solvent as a possible approach to track or match acetone sources for chemical forensic investigations. In particular, in the preparation of the toxin ricin, acetone is a common solvent used for the extraction of the castor oil. Previous work developed a simple solid-phase microextraction (SPME) approach to detect the use of acetone as an extraction solvent in the final crude ricin preparation (4). The motivation for this study is to provide additional sample processing information regarding the type or grade of acetone used in the production of ricin or other extracted materials. As broad spectrum chemical analysis becomes increasingly important GC×GC–MS has emerged as a powerful analytical tool. Comprehensive GC×GC is a multidimensional separation technique that separates a sample sequentially on two different columns, which are chosen to separate mixtures based on distinct — ideally orthogonal — chemical properties. Multidimensional separations have the advantage of greatly improved separations compared to conventional single-dimension approaches (5–9). The work presented here is the first step, examining the impurities present in various acetone sources. However, the overall importance of understanding the possible impurities present in a solvent is vital and can be applied to any trace analytical method. 

 Table I lists the 21 acetone sources obtained and examined for this study. The mention of trade names or commercial products in this work does not constitute an endorsement, criticism, or recommendation for their use. 

Table II: Experimental conditions for chemical analysis.			

 Chemical analyses were performed on a Leco Pegasus 4D GC×GC–MS system (LECO, St Joseph, Michigan) equipped with a Gerstel cooled injection system (CIS4) and multipurpose sampler (MPS2) (Gerstel, Baltimore, Maryland). The separation used ultrahigh-purity helium as the carrier gas, set at 1 mL/min, constant flow. Because the type and number of impurities were largely unknown for each acetone source, two different comprehensive GC×GC column sets were used (see Table II, A×C and B×C) for all 21 acetone sources. The first dimensional column was either a polar or nonpolar type and the secondary column was of moderate polarity and remained constant throughout the study. Unless stated otherwise the experimental conditions are listed in Table II. LECO ChromaTOF software (version 3.32) was used for data collection and analysis. Higher level data analysis was performed using an in-house software tool (Data Analysis Tool Extension, DAnTE, 

In this study 21 different acetone sources (brands) were analyzed to examine the trace impurities inherently present within the neat acetone source. Table I lists the 21 acetone sources, which range from high-purity laboratory grade to hardware-store grade to commercial fingernail-polish remover. Within this acetone sample set the supply containers included dark amber glass, clear glass, plastic, and metal. All acetone sources were quoted above 99% purity. In comparison to the extraction solvent artifacts referenced by Hunchak (2) and Lafleur (3), this study attempts to examine the impurities inherently present within an acetone source. To minimize acetone-related artifacts during the analysis, a cooled injection approach is used where 5 µL of acetone is slowly injected into a cooled injection system (CIS) with the majority of injected acetone vented during this process (see Table II for experimental details). Following the acetone injection process into the CIS, the solvent vent is closed and the CIS is rapidly heated to inject the remaining collected residual impurities. 

Figure 1: Representative 2D TIC obtained from (a) a 1-ÂµL split injection (split ratio = 50:1) and (b) a 5-ÂµL CIS injection, bottom chromatogram, of acetone A11 (hardware store brand) using the column set of AÃC. Within the chromatograms depicted the blue color represents the baseline signal and the red represents higher signal values. The signal intensity for the two chromatograms is set to 1 Ã 106 . The black outlined region represents the saturated and unsaturated hydrocarbon compound elution region. The dashed vertical line in the chromatogram (a) denotes the start of data acquisition (that is, 700 s) for the 5-ÂµL CIS injection separation.			

To offer a baseline perspective of the purity of a low-grade acetone source, Figure 1 shows a comparison between a 1-µL split injection of moderate split ratio (50:1) and a 5-µL CIS injection of acetone A11 (hardware-store brand) obtained using column set A×C. In these and the following 2D chromatograms, the blue color represents the baseline signal and the red represents higher signal values of compounds. As can be seen in Figure 1, at this 1-µL split injection level (top chromatogram) a number of trace compounds (peaks) are visually apparent, but with acetone the dominant compound in this separation. The major impurity compounds are listed in Figure 1 and include 2-pentanone, 4-hydroxy-4-methyl- (diacetone alcohol), and toluene. Pentane is observed as carryover from the syringe wash solvent used between injections. The horizontal solute bands at the end of the chromatogram are due to column bleed components. In general even this low-grade acetone source appears to be relatively pure at this injection level. However, when a 5-µL CIS injection of the same A11 is examined the number of impurities observed dramatically increases as shown in the lower chromatogram in Figure 1, where the vertical dashed line denotes the start of the data acquisition for the larger injection volume. These impurity components encompass a wide range of polarity, volatility, and analytical stability. Depending upon the source and type of acetone, numerous trace impurities are found in all 21 acetone sources and all the sources can easily be differentiated by visual inspection of the chromatograms. 			

Figure 2 summarizes the separations obtained for five of the acetone sources (A3, A9, A20, A23, and A27) for both column sets (that is, A×C and B×C) acquired using the 5-µL CIS injection approach. A number of observations are illustrated in Figure 2. In general, among all the acetone sources the number and amount of impurities can be extremely large and variable. The general chemical classes observed include saturated and unsaturated hydrocarbons, organic acids, ketones, and branched aromatic hydrocarbons according to their chromatographic elution order and mass spectral information. Interestingly, regardless of the column set used (that is, A×C or B×C) various solute zones appear as large streaks typically early in the 2D chromatograms, which was also observed in Figure 1. This solute zone streaking is probably due to the chemical mismatch between the solute and stationary phase (for example, nonpolar solute and a polar stationary phase and vice versa) of the first dimensional column so that there is a lack of adequate solute focusing during the injection process. For example, in acetones A3 and A9, both acetic acid and propanoic acid are found by using column set A×C with good solute zone shape; however, acetic acid is absent within the data collection window used and propanoic acid appears as a large streak when using column set B×C (Figure 2). Obviously the presence of these trace acids in the solvent could have deleterious effects for some solutes (for example, reactions producing artifactual solutes) and could confound trace-level applications. Importantly, these organic acids could easily go undetected if only a nonpolar-type column were used. Conversely, many of the other acetone sources have a large number of what appear to be hydrocarbon-type impurities according to the mass spectral information. These hydrocarbon-type compounds appear as streaks early in the 2D chromatograms when using column set A×C (for example, acetone sources A9 and A27 in Figure 2, as well as A11 in Figure 1 as highlighted by black outlined regions), but they appear as narrow solute zones with column set B×C. This hydrocarbon source region is outlined for acetone source A9 in Figure 2 and contains numerous straight chain, branched, and cyclo hydrocarbon compounds as indicated by their electron impact mass spectra. Varying amounts of these hydrocarbons can be found in all the acetone sources examined and the solute zone shape is more clearly defined when using column set B×C. Again, these solutes may go undetected if using column set A×C due to the lack of solute focusing or streaking. This result highlights the importance of using a number of different primary columns of different chemical nature when examining for trace-level components in a broad spectrum chemical analysis. (Note: As an aside, when using a common second dimensional column and changing only the first-dimension column the down time of the GC×GC–MS instrument is minimal, as the mass spectrometer may not need to be vented during this column switch.) 

Figure 2: Representative 2D TIC comparing the impurities observed for five of the acetone sources (A3, A9, A20, A23, and A27) for both column sets (AÂµC and BÂµC) and a 5-ÂµL CIS injection. The black outlined areas depict the elution region of the saturated and unsaturated hydrocarbon compounds. The difference between the elution regions and the solute zone shape for the two different column sets AÃC and BÃC is illustrated. The signal intensity for the chromatograms is set to 1 Ã 106			

The hydrocarbon region outlined in A9 in Figure 2 is examined more closely in Figure 3 with a comparison between acetone sources A9 and A27. In comprehensive 2D separations compared to conventional 1D GC separations, chemically similar compounds will create patterns or groupings of solute zones (for example, a homologous series of saturated hydrocarbons) within the 2D separation plane. These chemical groupings or families in a 2D chromatogram can be of help in identifying unknowns, when no standard components are available, in the absence of commercial MS library spectra or when the resulting experimental MS spectra are very similar to others. In Figure 3, three 

 values are selected to highlight groupings of saturated hydrocarbons (

 = 69), and alkyl substituted benzene rings (

 = 91), which also captures possible mesitylene, cumene, and their isomeric impurities. In general, as expected, the A9 acetone source appears to be of higher chemical purity than A27 as depicted by the smaller number of solute zones. Interestingly, acetone source A9 also appears to have a greater fraction of saturated and unsaturated hydrocarbon compounds that are eluted later in time. This implies that A9 has a larger fraction of higher boiling point hydrocarbons than does A27, which may indicate a certain type or degree of purification process. Finally, a series–family of alkyl substituted benzyl compounds (

= 91) is observed in both acetone sources, but the amounts present in acetone A27 are much larger. Again the presence of these benzyl compounds may indicate the type or extent of the purification process used.

Figure 3: Magnified chromatographic region outlined in Figure 2 for acetone sources A9 and A27, column set BÃC. For clarity reconstructed SICs are selected to highlight groupings of saturated hydrocarbons (m/z = 57, signal intensity = 5 Ã 105 ), olefins (m/z = 69, signal intensity = 5 Ã 105 ) and alkyl-substituted benzene rings (m/z = 91, which captures possible meitylene, cumene and their isomeric impurities, signal intensity = 5 Ã 104 ). 			

All 21 acetone sources were found to contain four of the common acetone impurities expected; diacetone alcohol (2-pentanone,4-hydroxy-4-methyl-), mesityl oxide (3-penten-2-one,4-methyl-), phorone (2,5-heptadien-4-one,2,6-dimethyl-), and phenol, but at varying amounts. Figure 4 illustrates a heatmap plot summarizing the results obtained from these four compounds for the 21 acetone sources using the A×C column set. In Figure 4, the solute zone volumes (absolute detector counts) are based upon the average of three replicate injections of the acetone source, which were analyzed in a random order. The average solute volumes illustrated in Figure 4 are calculated based on a log10 scale, where the blue color represents the lowest signal intensity and the red color represents the largest signal intensity of the compound. As illustrated in Figure 4 there appears to be a large variation in the amount of these four compounds for each acetone source. This indicates that acetone CIS injection process is not contributing appreciably to the presence of these compounds and that the components are likely residual from the manufacturing or distribution process. From Table I, acetone sources A23, A24, and A25 are from the same distributor and the same lot (that is, different bottles); acetone sources A9 and A10 are from the same distributor, but different lots (this is also true for acetone sources A5 and A6; however, A5 displayed a large isopropyl alcohol impurity of approximately 1:1 relative to acetone, chromatogram not shown); acetone sources A11, A27, and A28 are commonly available hardware store brands. These "source" trends are also observed in Figure 4, where these acetone sources tend to cluster closer to each other. The exception maybe the closer clustering of A9 and A20 to each other compared to the A9 and A10. Again this clustering is only using these four expected impurities, where the actual number of compound impurities is extremely large. As shown in Figure 2 the number, type, and amount of impurities present between these two sources (A9 and A20) is very different.  

Figure 4: A heatmap plot summarizing the results obtained using the signals from diacetone alcohol (2-pentanone,4-hydroxy-4-methyl-), mesityl oxide (3-penten-2-one,4-methyl-), phorone (2,5-heptadien-4-one,2,6-dimethyl-) for the 21 acetone sources using the AÃC column. The default settings for heatmap construction were used. This allows hierarchical clustering through a complete linkage of the Euclidean distance. 

For broad spectrum screening or trace component detection against complex background mixtures, impurities present within a solvent can significantly confound research efforts. In this work comprehensive GC–MS is used to assess the impurities present in various commercial acetone brands. A large number and broad range of different chemical impurities were found, dependent upon the acetone source. These chemical impurities ranged in complexity from simple organic acids, alkanes, and olefins, to a variety of substituted benzyl compounds (to highlight just a few). In general, dependent upon the analytical questions trying to be answered, a chemist should not become complacent and should be cautious of the impurities that may be present (and how they may change with different brands) and interfere with a trace-component analysis. This is highlighted by the presence of the organic acids present in various acetone sources and the subsequent need for two different column sets to better capture and understand the range of impurities present. Moreover this work addresses only those compounds that are stable and volatile for the GC system and does not address those components more labile or nonvolatile that may also be present in a solvent to alter or affect results. 

Funding for this work was provided through contract AGRHSHQDC07X00451 to Pacific Northwest National Laboratory by the United States Department of Homeland Security, Science and Technology Directorate. 

is a senior research scientist in the Chemical and Biological Signature Sciences Group at Pacific Northwest National Laboratory. He has a PhD in Analytical Chemistry from Michigan State University and a BS in Chemistry from Michigan Technological University. Dr Wahl's principal research interests focus on interdisciplinary chemical and bioanalytical problems that require advanced high-resolution separation techniques. Multidimensional separations for ultratrace analysis methods for chemical warfare agents and related compounds, biological, explosives and other signature compounds of interest are a main focus.

received a BS in Biochemistry from Portland State University and a MS in Forensic Science from National University. She has experience in a variety of analytical techniques including separations, mass spectrometry, and spectrophotometry in addition to writing standard operating procedures and quality control manuals, and applying statistical data analysis. 

 is a staff scientist within the Chemical and Biological Signature Sciences Group at Pacific Northwest National Laboratory in Richland, Washington. Dr Wahl has a PhD in Analytical Chemistry from Michigan State University and a BA in Chemistry from The College of Wooster. Dr Wahl's research interests include the development of MALDI mass spectrometry for microorganism analysis and the development of analytical methods for chemical and biological forensics. 			

(1) V.J. Aristovich et al., Method for Purifying Acetone, US Patent 6340777. 

Article

Investigating Solvent Purity Using Comprehensive Gas Chromatography: A Study of Acetones

R.C. McDowall is the principle of McDowall Consulting and director of R.D. McDowall Limited, and "Questions of Quality" column editor for LCGC Europe, Spectroscopy's sister magazine.

In this two-part article series I'll be discussing the role of system suitability tests (SSTs) in the context of analytical instrument qualification (AIQ) and the generation of quality analytical results. I'll look at this first from the perspective of regulatory requirements based upon the 

) general chapter 1058 (1058) on AIQ (1). As much of this discussion is also good analytical science I will also look at this from the perspective of ISO 17025 (2), the interpretation of the ISO 9000 quality management system for testing and calibration laboratories. I'll use the ISO 17025 term of calibration to be equivalent to AIQ and discuss the reason why I have taken this approach.  

Specifically, I want to discuss the question why SSTs are not a substitute for AIQ for the initial qualification or requalification of a chromatograph (that is, a periodic check or operational qualification) under GMP or calibration under ISO 17025. The reason for this is that there is a common misconception in some laboratories that SSTs can be used to qualify a chromatograph. This is wrong. Using SSTs as the sole instrument qualification approach will leave any laboratory exposed to regulatory action or nonconformance as the chromatographs cannot be demonstrated as being fit for their intended purposes. Therefore, we will explore the reasons for this from both the regulatory and quality perspectives. 

Table I: Selected U.S. good manufacturing practice (GMP) regulations for instruments and equipment 

An SST is simply a check to see if the chromatographic system (instrument modules, column, and mobile phase) is capable of undertaking the analysis you require. It is performed immediately before the samples you have prepared in the laboratory are committed for analysis. At it's simplest, an SST is the single injection of a standard solution of reference compounds to see if the separation is "OK." The only criterion used for assessing if the system is acceptable is a visual check by the chromatographer to assess if the separation is within expected boundaries. However, there are no measurements made: Just a visual assessment that things are right.  

At the other end of the scale are the formal SSTs described in the pharmacopeias used in pharmaceutical QC laboratories. Here a minimum of five replicate injections of a standard are injected and the calculated peak areas and other chromatographic criteria (retention time window, peak shape, peak tailing, resolution between peaks, and so forth) are objectively compared with predefined specifications defined by the pharmacopeia (peak area repeatability) and the laboratory (all other chromatographic criteria). If the SST injections meet all the predefined criteria then the chromatographic system is deemed acceptable and the samples committed for analysis.  

Good manufacturing practice (GMP) regulations from either the United States or Europe state simply that equipment used in the manufacture of drug products should be fit for their intended use as can be seen in Table I, which presents the U.S. regulations (3). Also, shown in Table I are the requirements for ensuring that scientifically sound specifications for the qualification work are written and that qualification and calibration of instruments is documented correctly. Note also that if an instrument fails to meet its specification it must be taken out of service until the problem is rectified and retested to show that it meets its specification. 

Table lI: Selection of the main ISO 17025 sections covering instrument qualification 

However, for chromatographic equipment these GMP regulations are expanded by either the 

) chapter 2.2.46 (5), which have both specified requirements for SST for chromatographic analysis to demonstrate that a chromatograph is fit for the analysis it will undertake on the day of analysis. SSTs must be run each time before an analysis is undertaken and each SST is specific for an individual method with predefined acceptance criteria — for example, precision, peak shape, and resolution from other analytes. If an SST fails, then the samples cannot be assayed.  

The principle of a point-of-use check, such as an SST, is applicable to any analytical instrument or system and this is performed just before an analysis to demonstrate correct performance, for example, a check to see if the chromatographic system or an analytical balance works before committing samples. This point of use check is used in other types of instrumental analysis — for example, balance and pH meter checks — so it is not an unusual approach and is good analytical science. 

 <1058> on analytical instrument qualification (AIQ) in an earlier "Questions of Quality" column (6) that looked at the impact of this on the chromatography laboratory. This informational general chapter consists of six major sections:  

Analytical instrument qualification process. 

 <1058> that concern our debate of AIQ versus SST are the components of data quality and the analytical instrument qualification process. The quotations from 

 <1058> come from these two sections. I will focus my comments on the first section of 

<1058> that discusses the components of analytical data quality. In my earlier column (6) I skipped over this section but I would like to go into further detail, as the content is good analytical science and is applicable to all laboratories. 

Figure 1: The Data Quality Triangle modified from USP . 

OK, so what is ISO 170252 and what is accreditation? First question first, ISO 17025 is an international standard that both testing and calibration laboratories can work against; for the purpose of this column I will focus on testing  laboratories mainly. A major difference between ISO 17025 and GMP is that compliance with the requirements of the ISO standard is voluntary rather than imposed by government agencies. However, the advantage of working to ISO 17025 is that a laboratory is working against an internationally recognized standard. To answer the second question, accreditation is the process of ensuring that the laboratory has implemented and complies with the requirements of the ISO 17025 standard and continues to do so. This is achieved by an independent audit by technical officers from the local accreditation body coupled with external experts in the specific areas offered by a laboratory. Each country in the European Union has an accreditation body (A2LA for the United States and UKAS for the United Kingdom) that is responsible for the accreditation of laboratories in their respective country. However, ISO 17025 goes further than GMP in that the test methods used by a laboratory and the performance of each one are also covered within the scope of a laboratory's accreditation and are included in the assessments made by accreditation bodies. 

When considering ISO 17025 we have to remember that this standard is a specific interpretation of ISO 9000 for testing and calibration laboratories. In this column I will focus mainly on the requirements for testing laboratories that use chromatographic equipment to perform their work. The applicable sections of ISO 17025 are presented in Table II and are discussed and interpreted in more detail below. Please note that key portions of the standard, as were the U.S. GMP regulations, are highlighted for the purposes of this column and the full sections should be interpreted in practice. 

The first requirement for equipment is in section 5.5.2 that notes that all laboratories are required to implement and operate a program for the maintenance and calibration of laboratory equipment. These calibration checks are only required where the equipment setting can significantly affect the test or analytical result (for example, oven temperature, gas or liquid flow, detector wavelength accuracy, and photometric linearity for GC or LC modules). Proper equipment use combined with periodic servicing and calibration does not ensure that an instrument, in this case a chromatograph, is performing adequately; there also has to be point of use checks of overall system performance that can be built into test methods (for example, chromatographic SSTs). These checks should be documented and should be satisfactorily completed before the equipment is used or before results are accepted. 

Reconciling GMP Regulations and ISO 17025 Requirements

How can we have a common understanding of both the GMP and ISO 17025 requirements so that we can have a debate that is independent of applicable regulations or requirements? This all comes down to definitions and what we mean by specific words. So here are the key terms used in either regulations or standards and let's see how we can reconcile them to help us in the debate:  

 In GMP the term "equipment" is used to cover anything used in the manufacture of pharmaceutical products that includes the laboratory. Similarly, ISO 17025 uses the term equipment to cover the wide range of equipment, instruments and systems used in the testing laboratory. This is in contrast to 

 <1058>, which uses the term instrument. However, in practice the two terms are synonymous as we can refer to chromatographic instruments or chromatographic equipment. I prefer the term instrument and will use that in the remainder of this article. 

 Both GMP and ISO 17025 refer to calibration in contrast to qualification used by 

 <1058>. Are the two terms equivalent? Yes, in my view. Calibration is a simpler term to understand and has been defined by EU GMP as the set of operations that establish, under specified conditions, the relationship between values indicated by a measuring instrument or measuring system, or values represented by a material measure and the corresponding known values of a reference standard (7). I would argue that the work undertaken in qualification and calibration of an instrument is essentially the same: Does it work for your purpose and can you demonstrate that by using calibrated test instruments and traceable reference standards? 

 All regulations and guidance refer to these terms so there is full agreement and thus no need for interpretation.  

An example of how a laboratory can fall foul of a regulator is found in the U.S. Food and Drug Administration (FDA) warning letters publicly available on the Agency website (

). Shaanxi Hanjiang Pharmaceutical Group, a Chinese company, received a warning letter in January 2008 following an inspection of its facilities in September 2007 (8). Quoting from the warning letter we can see that: 

3. Calibration of measuring devices was not traceable to the national standards. 

Several balances and thermometers were not traceable to national standards. 

We acknowledge your firm's submission of a certificate of metrology <redacted> which expires on 16th April 2008. However, this certificate is deficient in that it does not include identification of the instrument for which it is meant for and does not include a statement of traceability.

So we can see the requirement for traceability outlined in ISO 17025 can be mirrored in GMP. The only problem is that there is no explicitly written requirement for this as you can see in Table I. However, as you can see in 211.160(b) in Table I there is a need for the laboratory controls to be "scientifically sound," as traceability of calibration is good science. Therefore, if you have not calibrated or qualified your instruments with appropriate and traceable reference standards then the data you produce from those instruments will be poor.  

In this article, I've looked at the regulations and quality standards for qualification and calibration of chromatographic instruments. In part II we'll look at how the data quality triangle can be applied to this from a practical perspective. 

 is principal at McDowall Consulting, Bromley, Kent, UK. He is also a member of 

 general chapter <1058> Analytical Instrument Qualification. 

(2) ISO 17025 General requirements for the competence of testing and calibration laboratories, International Standards Organisation, 2005. 

(3) Good Manufacturing Practice for Finished Pharmaceuticals (21 CFR 211), 2008. 

(5) EP Chromatography chapter 2.2.46 Chromatography. 

(7) EU Good Manufacturing Practice Glossary. 

(8) Shaanxi Hanjiang Pharmaceutical Group, FDA Warning Letter, January 2008 (

Can you qualify or calibrate your chromatograph using a system suitability test? No. However, for the doubters among you I'll spend the rest of this two-part article series explaining why.

Why System Suitability Tests Are Not a Substitute for Analytical Instrument Qualification or Calibration, Part I

Ron Majors, editor of "Column Watch" and "Sample Prep Perspectives," has been with 

 for over 26 years. Currently a senior scientist with Agilent Technologies, Wilmington, Delaware, Ron is known industry-wide as one of the premier chromatography experts in the field. He is also a member of 

In most forms of classical liquid chromatography and high performance liquid chromatography (HPLC), totally porous particles such as silica gel, alumina, and ion exchangers have been used in most modes of chromatography. These materials are useful because they present a high surface area, show good retention and sample capacity, and provide a base for the further bonding of organic moieties.  

In the early 1960s, HPLC as we know it today hardly existed. Although researchers had predicted theoretically that liquid chromatography could be more efficient and faster, little practical work had been done to realize these predictions. At that time, most chromatographers used open glass columns packed with porous particles larger than 100 µm. Most of these particles were obtained by sorting wide size ranges of irregularly shaped particles into narrower cuts using sieves. Chromatographers used gravity as the pressure source to move liquids through the glass columns, collected analytes at the column exit, and measured the analytes in the effluent using off-line techniques such as UV–vis spectroscopy. By today's standards chromatographic peaks were very broad. Attempts to speed up the chromatography by applying positive pressure only resulted in broader peaks because of slow mass transfer into and out of these large porous particles. Something had to be done to speed up chromatography. Clearly for high-speed separations, the solute migration distance or pore depth had to be decreased or the rate of solute diffusion had to be increased. 

The theoretical work of Purnell (1), Golay (2), and Knox (3) and experimental GC results in Horvath's PhD thesis (4,5) predicted that using a support with an impenetrable hard core coated with a thin outer layer of a porous solid should provide high-speed solute mass transfer in the stationary phase. Horvath and coworkers (6) first demonstrated the so-called porous layer bead (PLB, also referred to as a pellicular or superficially porous particle, SPP) support for the ion-exchange chromatography of nucleotides. This SPP consisted of a glass bead solid core with a thin layer of polystyrene–divinylbenzene polymeric resin modified with anion-exchange groups on its surface. Figure 1 depicts how this new concept compared to the large porous particles of yesteryear. Figure 1a shows a typical large porous particle used in the early days. When a solute came down the column it had to diffuse into the pores of the particles (millions of times as it moved through the column) and because of this slow diffusion into and out of the stagnant mobile phase, a great deal of time is required. Meanwhile, the mobile phase that the solute was originally in has long since passed the particle. This slow mass transfer is a source of band broadening in chromatography. Of course, the temperature of the column could be increased to decrease mobile phase viscosity but this resulted in only minor improvement and at the time was inconvenient.

 Figure 1: Comparison of solute mass transfer (migration) distances in (a) large porous particles with irregular shape (b) early SPPs (~1970), and (c) small porous particles with spherical shape (mid-1970s). 

Figure 1b depicts the concept of the SPP. A spherical solid glass bead of 40 µm in diameter was used as the base material. On the outside the glass bead was coated with a thin layer of ion-exchange resin (6), silica gel (7), or controlled surface porosity beads (8). The layer was usually 1–2 µm thick. Now the solute would come down the column and diffuse into the SPP but the diffusion distance was limited by the underlying solid core. By limiting the deep pores, the SPP diminished the pathlength for diffusion in the stagnant mobile phase. Diffusion could only occur in the thin layer on the surface so that rapid mass transfer took place. From a theoretical point of view, the thinner the layer the faster will be the mass transfer and the better the efficiency. Of course, as the layer becomes thinner, both the capacity factor (retention) and the sample capacity (loading) of the SPP decreases so that its separation potential is less. Thus a compromise must be made between the layer thickness and the capacity. 

Compared with the large, porous particles used earlier in LC, SPPs provided much faster separations (in the 20–30 min range, rather than several hours), more efficient separations (height equivalent to a theoretical plate, HETP, in the 1–2 mm range), and better limits of detection. Users could dry pack the dense, solid core pellicular packings into 50-cm and 1-m stainless steel columns easily using mechanical vibration. Many chemical bonded phases were developed to provide various separation mechanisms especially for normal-phase, reversed-phase, and ion-exchange chromatography. Many successful applications using SPP columns were developed in the early 1970s. 

The invention of commercial SPPs in the late 1960s to early 1970s along with the advent of the flow-through UV detector were the two occurrences that kicked off the development of HPLC. Of course, the breakthroughs in the sizing and packing of small silica porous particles in the early 1970s (9,10) that resulted in further efficiency increases led to the demise of large SPPs. The microparticulate packings (depicted in Figure 1c) improved mass transfer even more and improved efficiency by at least another order of magnitude. In addition, the high surface area of these small porous particles provided much greater sample capacity than SPPs and these packings could also be used for preparative work and with larger sample injections for better detection. Readers interested in the complete story can refer to an earlier publication I wrote of the subject of the commercial development of SPPs and microparticulate packings (11). 

Through the decades, newer versions of SPPs have occasionally reappeared to address alternate solutions to specific problems. Although the large particle SPPs were widely used in the 1970s they gave way to the microparticulate packings which took over usage in the 1980s and early 1990s and remain the favored HPLC packing today. After the 1970s, the large-particle SPPs were relegated to previously validated methods or as an inexpensive, easily replaced packing material for guard columns. 

Rather than a 40-µm-diameter particle, the first modern SPPs were 5.0 µm in diameter with a 0.25-µm layer and 300-Å pores (12). Instead of a glass bead, solid silica cores were used as a substrate. Then, a thin film of silica nanoparticles was coated onto this core. These SPP columns were recommended for the rapid separation of biomacromolecules such as proteins (13). Large biomolecules have very low diffusion coefficients, about 1/10th that observed for a small molecule. The thin shell of the SPP allowed the slowly diffusing proteins to only partially penetrate the packing (since the solid core prevented further diffusion). Compared with totally porous particles of the same size, mass transfer was more rapid and a high flow rate could be run without sacrificing much column efficiency (12). In Figures 2a and 2b, 5-µm porous particles and 5-µm superficially porous particles are depicted. For the totally porous particle (Figure 2a), assuming that the solute molecule diffused to the center of the particle and then back out again, the total solute migration distance in this particle would be 5 µm (2.5 µm in and 2.5 µm out). The large biomolecule will diffuse very slowly in this distance resulting in rather broad peaks especially as the flow rate was increased for a more rapid separation. 

Figure 2: Comparison of diffusion distances in modern HPLC packings: (a) 5-Âµm totally porous particles, (b) 5-Âµm SPPs, and (c) 2.7-Âµm SPPs. 

Contrast this to Figure 2b that depicts a bonded phase SPP of the same particle diameter. Assuming a protein with diffuse into and out of the thin layer the total migration distance would be 0.5 µm (0.25 µm in and 0.25 µm out). This distance would be 1/10th that of the totally porous particle. Thus the diffusion time is much less and the peak width should be narrower. Even at higher flow rates, the peak widths would be narrower than the porous particle at the same flow rate (12). This comparison can be made by looking at the chromatograms of Figures 3a and 3b. Here, for a couple of proteins, lysozyme and myoglobin, were separated at increasingly higher flow rates on a 75 mm × 2.1 mm column packed with totally porous wide-pore (300-Å) bonded silica particle (Figure 3a) and a superficially porous wide-pore (300-Å) bonded silica particle (Figure 3b). As is usual for proteins, the chromatograms were run under gradient conditions. Had the chromatograms been run under isocratic conditions, the differences in column efficiency would have been even greater. Under gradient conditions, the SPP column showed narrower peaks than the totally porous particle column as a result of more rapid mass transfer. One can see that the small peak on the leading edge of lysozyme remains partially resolved on the SPP column. Note that the chromatograms were aligned to make the comparisons easier. Also, as the flow rate increased, the retention times on both columns decreased proportionally. Also, the gradient times had to be adjusted as the flow rate was increased.

Figure 3: Comparison of totally porous particle and SPP columns at increasingly higher flow rates for biomacromolecules under gradient conditions. (Chromatograms are aligned for comparison.) Columns: (a) 75 mm Ã 2.1 mm, 5-Âµm Zorbax Stablebond C18, totally porous particle; (b) 75 mm Ã 2.1 mm, 5-Âµm Poroshell 300 Stablebond C18, superficially porous particle; mobile phase A: 95% water, 5% acetonitrile with 0.1% trifluoroacetic acid; mobile phase B: 5% water, 95% acetonitrile with 0.07% trifluoroacetic acid; temperature: 70 Â°C; detection: UV at 215 nm. Peaks: 1 = lysozyme, 2 = myoglobin. 

To illustrate the resolving power of a superficially porous wide-pore particle column, Figure 4 shows the rapid gradient separation of eight proteins in just over 0.75 min at a very high linear velocity for a 2.1-mm i.d. column.

Figure 4: High flow rates with 2.1-mm i.d. SPP column for high resolution and fast protein separations. Column: 75 mm Ã 2.1 mm, 5-Âµm Poroshell 300SB-C18; mobile phase A: 0.1% trifluoroacetic acid; mobile phase B: 0.07% trifluoroacetic acid in acetonitrile; gradient: 5â100% B in 1.0 min; flow rate: 3.0 mL/min; temperature: 70 Â°C; pressure: 250 bar; detection: UV at 215 nm. Peaks: 1 = angiotensin II, 2 = neurotensin, 3 = RNAse, 4 = insulin, 5 = lysozyme, 6 = myoglobin, 7 = carbonic anhydrase, 8 = ovalbumin. 

Previous results show that the superficially porous 300-Å particles are suitable for the rapid separation of macromolecules. For small molecules, the recent development in sub-2-µm porous particles (1.5–1.9 µm) has allowed for high-speed separations. Indeed, short 50-mm columns packed with sub-2-µm particles give rise to separations in the minute range and longer columns (up to 150 mm) can generate tens of thousands of theoretical plates. However these columns can generate very high back pressure due to the small particle size, sometimes necessitating UHPLC systems.  

Once again, only recently, the concept of SPPs was revisited. The new generation of SPPs, also called fused-core-, shell-, porous-shell, and Poroshell particles, was introduced. These new SPPs have a particle diameter of 2.6–2.7 µm but with a 0.5-µm porous layer and 90–120 Å pores are more suited to small molecule separations. These 2.6–2.7 µm SPPs provide a lower pressure drop (roughly 40–60% lower) than that observed for the sub-2-µm porous particles yet the short diffusion pathlength (slightly lower C term of the van Deemter equation) and other favorable particle characteristics give rapid solute mass transfer indicative of a sub-2-µm porous particle. Using a special synthesis process, very narrow particle size distributions of SPP can be achieved, much narrower than can be obtained with porous particles. Thus, these dense particles can be packed very efficiently into an HPLC column. This favorable particle configuration gives a lower A term (eddy diffusion) in the van Deemter equation. In addition, studies have reported a reduction in the B term (14). By a combination of the favorable lowering all of three terms in the van Deemter equation, efficiency of a column packed with these SPP particles rivals the sub-2-µm particles. Thus, high efficiency at lower pressure allows users to adapt their conventional LC systems without having to invest in UHPLC hardware. On the other hand, having UHPLC hardware allows one to use very long SPP columns to generate high numbers of theoretical plates. Figure 5 compares two high-efficiency separations performed with SPPs and totally porous sub-2-µm particles (15). Most comparisons done nowadays are shown on short columns in high-throughput applications, usually with some minute time scales. This example shows very long columns that together generated about 100,000 theoretical plates for both types of packings. Both sets of columns had 55 cm × 2.1 mm dimensions. In addition to the plate counts, the overall separation times were roughly the same. The major difference in the current chromatograms was the column pressure drop. For the totally porous sub-2-µm particle column, the pressure was just over 1000 bar (15,000 psi), while for the 2.7-µm SPP column, the pressure was only 547 bar (8200 psi).

Figure 5: High efficiency separations on 2.7-Âµm SPP and totally porous particle sub-2-Âµm columns. Top: 55 cm Ã 2.1 mm (3 Ã 15 cm, 10 cm), 2.7-Âµm Poroshell 120 EC-C18 (SPP); mobile phase: 56:44 acetonitrileâwater; flow rate: 0.202 mL/min; temperature: 40 Â°C; pressure: 547 bar. Bottom: 55 cm Ã 2.1 mm (3 Ã 15 cm, 10 cm), 1.8-Âµm ZORBAX RRHD Eclipse Plus C18 (totally porous particle); mobile phase: 59:41 acetonitrileâwater; flow rate: 0.23 mL/min; temperature: 40 Â°C; pressure: 1001 bar. (Reproduced with permission from reference 15.) 

Commercial Availability of Modern SPP Columns

So far, there are only a few companies that have introduced superficially porous particles for macromolecular and small molecule separations (see Table I). Besides a 2.6-µm particle size, Phenomenex (Torrance, California) has introduced a 1.7-µm SPP that would have a similar pressure drop to the sub-2-µm UHPLC materials but should offer even higher efficiency, perhaps stressing modern UHPLC instrumentation in terms of extracolumn contributions to band broadening. Because of the acceptance of SPPs for both HPLC and UHPLC, it is anticipated that additional companies will introduce products in the near future. In addition, more phases for the sub-3-µm SPPs are under development so that there will be as many chemistries available as for sub-2-µm and regular HPLC particles.

Table I: SPP columns for reversed-phase chromatography 

The concept of SPP supports has been around a long time in LC and now has reemerged in a format that rivals separations achieved on small porous particles with sub-2-µm diameters. The newer SPP supports provide rapid separations of large biomolecules as well as small molecules, the latter with lower pressure drops than the sub-2-µm competitive columns. Although not discussed here, SPPs for small molecules have thicker porous layers and thus their sample capacity is not greatly reduced as was observed in early large-particle SPPs. One can do the calculation but the available surface area is only reduced 25% compared to a totally porous particle. However, the particle density for the SPPs is higher and therefore the amount of stationary phase packed into an SPP column is comparable to that packed into a totally porous particle column. Thus, the total sample capacities of the two types of columns are similar. Because the 2.6–2.7 µm SPP columns for small molecules and the 5.0-µm SPP columns for macromolecules have regular 2.0-µm-porosity inlet and outlet frits, they are less prone to plug than the sub-2-µm particle columns that have smaller porosity frits. Clearly, there is a bright future for such packing materials as users become more familiar with the technology. 

"Column Watch" Editor Ronald E. Majors is Senior Scientist, Columns and Supplies Division, Agilent Technologies, Wilmington, Delaware, and is a member of 

's editorial advisory board. Direct correspondence about this column to "Column Watch," 

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

R.P.W. Scott, Ed. (Butterworths, London, 1960), p. 139. 

(4) Cs.Horvath, Ph.D. Thesis, Universitat Frankfurt am Main, Germany, 1963. 

(6) Cs. Horvath, B.A. Preiss, and S.R. Lipsky, 

(13) R. Ricker, C. Woodward, and R.E. Majors, 

(14) F. Gritti and G. Guiochon, submitted for publication. 

(15) J. Link, W. Long, W.E. Barber, and X. Wang, "A Critical Performance Comparison of Column Options Using Poppe and Kinetic Plots", Pittcon 2010, March 2, 2010, paper 1230-4. 

Column Watch

This installment will trace the development of SPP supports to the present-day versions that rival UHPLC supports.

The Increasing Role of Superficially Porous Particles in HPLC

What are glycoproteins and why have they become so very important in the biotech industry today? Why are so many biotech firms emphasizing glycoproteins for their product pipeline, especially antibodies?

Glycoproteins represent but one class of protein variants, alongside other classes, such as phosphoproteins, membrane proteins, or hydrophobic proteins, but they have become one of the most intensely studied and commercialized class, as real or potential biotherapeutics today (1–6,59,60). Perhaps one of the best discussions of the chemistry and biochemistry of glycoproteins is that by Walsh (7). Figure 1 illustrates, in schematic fashion, a typical antibody, which is just one type of glycoprotein, perhaps the most commercially successful and viable biopharmaceutical products now on the market. Though this particular glycoprotein has its glycans (carbohydrate chains, circles in Figure 1) on only one amino acid in the antibody heavy chain, at two sites (because there are two heavy chains in these antibodies), most glycoproteins have several sites of attachment and several different glycans. Even monoclonal antibodies have several different variants or isoforms (different isoelectric points or pIs), because of the expression of a set of glycoforms on the heavy chains. Glycoproteins with more than one site of glycosylation reflect heterogeneity from each site. Or, because of changes in certain amino acids (for example, glutamic acid to pyroglutamate, deletion of C-terminal lysine or asparagines deamidations, and others). And, of course, as the number of glycans (oligosaccharides, often with several antennae or side chains) incorporated is increased in other glycoproteins, sometimes dependent on the cell expression system, the number of possible variants or glycoforms (that is, isoforms or variants or posttranslational modifications) also increases (7–12). 

Glycoproteins appear to be uniquely important in cell functioning, stability in vivo, cellular distribution in vivo, membrane transport, and even cellular recognition or differentiation (62–64). Glycoproteins are often important integral membrane proteins, where they play a role in cell-cell interactions (62). They are also important for white blood cell recognition, especially in mammals. Classes of glycoproteins include: collagens, mucins, transferrin, ceruloplasmin, immunoglobins, histocompatability antigens, human chorionic gonadotropin (HCG), thyroid-stimulating hormone (TSH), alkaline phosphatase, certain plasma proteins of coldwater fish, various proteins involved in hormone and drug action, calnexin, calreticulin, notch, and specific glycoproteins on the surface membranes of platelets (62). 

Figure 1: A typical antibody structure, indicating arrangements of heavy chain (HC), light chain (LC), and location of glycosylations at one specific arginine amino acid on the HCs (PhD Thesis, Ling Chen Santora, Northeastern University, 2001). 			

A listing of some FDA-approved monoclonal antibody biopharmaceuticals can be readily found on their website (9,10). These include Rituxan, Remicade, Zenapak, Herceptin, Zevalin, Humira, and others. Quite aside, there is an equally large number of nonantibody, glycoprotein biopharmaceuticals on the markets today, including Enbrel, TNKase, tPA, Aranesp, and Xigris (9). Perhaps one of the most popular and useful texts that deals just with proteins in general, as well as glycoproteins, is that by Creighton (11). A very useful overview or synopsis of the major analytical approaches or techniques in glycobiology is that edited by Townsend and Hotchkiss (12). A similar overview was edited by Jackson and Gallagher (13). It is very likely that more publications in analytical biotechnology have appeared and will continue to appear dealing with glycoproteins than any other class of biopharmaceuticals (14–24).  

Glycans is a commonly used term that refers to carbohydrate or oligosaccharide chains, often having multiple chains or antennae (for example, biantennary, triantennary, and others) (54–56). 			

Some very recent texts are dealing specifically with glycoanalysis, and virtually all of the currently available analytical methods (54,55). Glycobiology refers to the biology associated with these glycans on glycoproteins, separate polysaccharides alone, quite apart from protein attachments. The analysis of glycans attached to glycoproteins, or for that matter of free-standing polysaccharides or monosaccharides, is an enormous field, which is today an expected assay to be performed on virtually all glycoprotein biopharmaceuticals destined for marketing (56). The United States Pharmacopeia (USP) has recently put forth for review and public comments a proposed, new, general information chapter that discusses glycan analysis methods and analytical strategies (56). It also discusses, at some length, approaches for the direct analysis of glycoproteins and the analysis of released, derivatized, or nonderivatized glycans.  

Glycoproteins have certain, perhaps novel, attributes that have made them the preferred biopharmaceutical products for numerous applications and indications (7–10). They are readily expressed in both mammalian and vegetable cell expression systems, often in high yields, against numerous targets or applications (9). They have long shelf-lives in the dry state, show good stability profiles, generally are not immunogenic in the monomeric state, and are cleared from the body readily after delivering an effective performance. They are also demonstrating good membrane transport, have higher than general water–biofluid solubility, are not rejected by the body when humanized, and can be manufactured with high reproducibility, lot-to-lot and batch-to-batch (7,8). They are also often produced in high yields, are readily purified using a number of process techniques, and as antibodies, can be designed to target very specific disease sites. Though, at times, they have been manufactured with human, viral impurities, expression thru vegetable systems appears to overcome these problems at times (25,26). It has been estimated that, today, over 50% of all biopharmaceuticals now going through the various pipelines in the world's biotech companies are glycoproteins in nature. This would also include fusion proteins, often using the Fc region of a humanized antibody to improve delivery and stability of the partner protein. The Fab (antigen binding) region of an antibody targeted against a specific site can also be fused with another protein that delivers the true therapeutic activity. And, one might reasonably expect that glycoproteins may well increase in usage and commercialization with time, as they have already over the past 10 years (27–29). 

What then are the problems remaining for even greater utilization of glycoproteins as biopharmaceuticals in the future? Surely, one of these problems relates to our current ability to fully or well characterize glycoprotein variants, which often comprise a dozen or more species in a final drug substance. This is not a trivial issue, especially when we now begin to consider marketing biogenerics, biosimilars, or biologic follow-ons, as more and more glycoprotein, proprietary products come off patent (61). With most humanized antibodies, there are usually two sites of glycosylation, N-linked, as above, on the two heavy chains. And, depending on the cell expression system, these may have a limited number of glycans attached, still leading to a dozen or more final variants (also known as glycoforms). However, other glycoproteins, depending on how they are expressed, may have dozens of variants present, all of which may well form the drug substance. 

These are not always easily resolved by even two-dimensional chromatographic or electrophoretic methods (for example, multidimensional liquid chromatography [MDLC], two-dimensional gel electrophoresis [2DGE], and differential gel electrophoresis [DIGE]). And, in the case of glycoproteins (not antibodies), each of these variants may have both N- or O-linked glycans, with several amino acids bearing such attachments. As the number of glycans attached to a single variant increases, our ability to fully characterize each variant decreases, perhaps exponentially. It is difficult enough to fully characterize a single variant, once isolated and pure or homogeneous, but if these are not baseline resolved before mass spectrometry (MS) detection in LC or capillary electrophoresis (CE) modes, their characterization again becomes problematic. And, using bottom-up methods, rather than both top-down and bottom-up simultaneously, putting the right peptides to the right parent protein again becomes even more problematic. When Humpty Dumpty fell off that famous wall, all the king's men could not put Humpty Dumpty back together again. Thus, complete characterization of an individual glycoprotein may become much, much easier when we can baseline resolve it from all other variants present. And, this may only be feasible when we can finally do true, top-down protein sequencing (TDS) of even high MW glycoproteins, including all sites of glycan attachment and the nature of each glycan present at any amino acid attachment site. That is not yet here, but it is indeed truly coming (

What do we mean by true, complete characterization of a glycoprotein, as opposed to the euphemistic expression, well-characterized biopharmaceutical product (WCBP)? Complete characterization is what we just suggested, that we know the complete amino acid sequence, we know every site of glycan attachment, and we know the specific glycan sequence at each such site. And, it also means that we know the relative or absolute levels of each variant in the final drug substance, after complete characterization. WCBP does not mean this, but it means that we have practically (almost) characterized the drug substance, and that we can make it the very same way, batch-to-batch with a well-controlled process. The process then defines the product, but only because we were not able to more fully or completely characterize each and every major variant in the drug substance. Without the ability to demonstrate true, complete characterization of a biogeneric glycoprotein mixture, it seems that the regulatory agencies may well require a clinical demonstration of safety and efficacy (29). With the ability to demonstrate true, complete characterization and quantitation of a biogeneric glycoprotein mixture, the agencies most likely will not demand clinical testing, perhaps only animal testing? Thus, as we move into the world of biogenerics and follow-on biologics (FOB), there will be more and more pressure for the biogeneric firms to be able to demonstrate true, complete glycoprotein mixture characterizations, if they wish to bring their biogenerics to market ASAP.  

Why Can We Not Fully Characterize Complex Mixtures of Glycoproteins Today, or Can We? What Do We Do Instead?

If we have generated a complex mixture of new, unknown glycoproteins as our drug substance, the regulatory agencies would like us to characterize, as best possible, each and every major variant present. And, they would like us to demonstrate batch-to-batch chemical equivalency for that same mixture of glycoproteins, in the same relative or absolute amounts produced, batch-to-batch. Thus, they want a demonstration of complete variant structures, as best possible, and then batch-to-batch chemical equivalency or lot release testing. This is often easier said than done, of course, especially if one is dealing with a complex mixture of glycoprotein variants and they are not resolvable by any analytical separations technique. It then becomes almost impossible to demonstrate complete protein characterization, unless one is able to do intact, top-down sequencing for each species present. Figure 2 demonstrates a typical, 2DGE plot for a mixture of yeast proteins, using isoelectric focusing in the first dimension followed by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) in the second. This technique is often employed within contract service organizations (CSO) to "identify" variants, using bottom-up methods with database searching routines, and to provide the identify of known, putative proteins (54–56). Such databases, of course, do not usually exist for proprietary or generic biopharmaceutical glycoproteins. 2DGE and other analytical separation methods are routinely used to "demonstrate" lot-to-lot chemical equivalency after the individual, major spots have been fully or well characterized as being a part of the drug substance.

Figure 2: Silver-stained, two-dimensional gel electrophoresis pattern of 50 Âµg yeast extract. 			

True characterization of each spot would require a combination of both top-down, intact MW determination followed by peptide mapping with identification of all glycopeptides present, the amino acid location of each glycan and its sequence (as well as anomericity, 

 orientations of hydroxyls, and all linkages). This is not yet true TDS, it is just determination of the intact MW of the glycoprotein combined with bottom-up peptide mapping and peptide mass fingerprinting (PMF) to deduce the complete sequence, location, and nature of each glycan present. This will only work, however, if every single variant is fully resolved from the next, so that only a single variant goes into the matrix-assisted laser desorption ionization–time-of-flight mass spectromtry (MALDI-TOFMS) or high performance liquid chromatography–electrospray ionization mass spectrometry (HPLC-ESI-MS) system for top-down and bottom-up sequencing. There are numerous MS approaches described in the literature (

) that have been described as being useful to fully or partially characterize glycoproteins. However, this is not yet true top-down protein sequencing, which will come later, perhaps this year or next? 

If two or more glycoproteins are in a single 2DGE spot, their intact MWs can be determined by MALDI-TOFMS or ESI-MS, but it then becomes somewhat problematic to do peptide mapping on a mixture of the original proteins and put the jigsaw puzzle together in the right arrangements for each protein. It is not impossible, just very tricky. And, if a spot contains more than two glycoproteins, the puzzle becomes all the more problematic and perhaps impossible to solve. This is quite aside from the obvious problem of ion suppression, whenever multiple species of similar structure enter the MS instrument. Hence, there remains the basic, fundamental need to resolve each and every single glycoprotein variant before efforts are initiated to do characterizations. If we were able to perform true TDS in the MS system, even for unresolved parent proteins, then we could do complete characterizations of unresolved proteins. That is almost here, but we have not yet seen it fully demonstrated for a complex mixture of glycoproteins in a drug substance (

Thus, many FDA filings as CMCs (Chemistry, Manufacturing, and Controls) for NDAs (New Drug Applications) contain alternative approaches to WCBPs, without actually demonstrating true and complete glycoprotein characterization. These approaches have always been limited by the existing analytical technologies. Thus, HPLC profiles of the drug substance, showing partially resolved variants, or 2DGE of partially resolved variants, or HPCE or MDLC of partially resolved variants have been routinely used to demonstrate batch-to-batch chemical equivalency (along with other analytical data) (37,38). Figure 3 illustrates the ability to use both cation-exchange chromatography and capillary isoelectric focusing (cIEF) to partially resolve all the variants present in a precommercial batch of a humanized monoclonal antibody destined for market (37). However, full resolution and chemical characterization of each individual variant present was not, at that time, possible.

Figure 3: The ability to use both cation-exchange chromatography and capillary isoelectric focusing (cIEF) to partially resolve all the major variants present in a precommercial batch of antibody variants (37). Treatment of the original mixture of variants with carboxypeptidase B (CPB) removes C-terminal lysines and causes the three major peaks to collapse to one peak by both cation-exchange chromatography and cIEF.			

However, determining the specific nature of each glycoprotein has usually not been possible, and thus glycoprofiling is routinely used to demonstrate the total glycans present on the entire mixture or individual variants. This technique, known as glycan analysis, releases N- and O-linked glycans, separates them from the residual protein mass, and then characterizes the glycans present to provide a glycoprofile (1DGE, 2DGE, high performance capillary electrophoresis [HPCE], hydrophilic interaction chromatography [HILIC], MALDI, 2-AB MALDI, and others). There are numerous analytical methods now available for performing glycoprofiling (54–57).  

This adds to a partial WCBP characterization profile and provides confidence that lot-to-lot chemical equivalency can or may be demonstrated. However, it does not provide true, complete chemical characterization of each and every glycoprotein present in that drug substance. Nor does it demonstrate the absolute amounts of each variant present, but only relative percent peak areas, which also helps demonstrate chemical equivalencies for lot release purposes. In general, virtually all prior chemical characterizations of complex mixtures of glycoproteins, including antibodies, have not been able to do true, complete characterization of each and every variant present. This is changing, slowly but surely, and eventually, it seems clear we will be able to do this, even for a complex mixture of glycoproteins that does not allow for complete baseline separation of each and every variant present. Partial resolution may suffice if we can do complete sequencing via top-down methods that also provides location and nature of each and every glycan on each and every glycoprotein. This is not impossible, just slow in coming but almost here. These techniques require the ability in the MS system to perform amino acid sequencing for each peptide in the peptide map of the glycoprotein, to demonstrate the location of every glycan on a specific amino acid in every glycopeptide, and then to sequence that glycan to determine its exact chemical structure. These are, by now, somewhat routine measurements, at least for glycopeptides, if not quite yet for glycoproteins. And, if used together with top-down, intact MW determinations of the parent glycoprotein, such techniques are fully capable of providing complete chemical characterizations. Again, baseline resolving each, intact glycoprotein, either in MDLC or 2DGE, before the top-down and bottom-up MS routines above, will virtually guarantee obtaining a complete chemical characterization of every glycoprotein in a drug substance. And, without any ion suppression occurring, either.  

When Can We Fully Characterize Complex Mixtures of Glycoproteins Today?

There are several ways by which complex mixtures of glycoproteins can, in principle, be fully characterized, but these will require some form of multidimensional separations (39). Separation approaches such as MDLC or 2DGE are capable of providing upward of 5000 peak capacity, and thus should be able to fully, baseline resolve most, if not all, recombinant glycoprotein drug substances (39–41). However, such high resolving analytical techniques are perhaps too complex to be used in the biotech industry for drug substance, chemical characterizations, or chemical equivalency (lot-to-lot release testing), without database searching. Both MDLC and 2DGE in proteomics usually rely on using established, databases together with PMF and peptide mapping to identify known proteins. And most biotechnology products are not in commercial databases, they are genetically engineered and not naturally occurring, in most instances. A combination of top-down and bottom-up protein sequencing, on baseline resolved glycoprotein variants, will provide the correct, complete structure of each variant, if the proper MS system is employed. Again, there are very few, if any, examples of such complex analytical techniques being routinely used in the biotech industry today. It is not that it cannot be done, it is just that it is not being routinely used, as yet (17,19,30–36,45–50).  

We should hasten to add that complete identification of individual, glycopeptides coming from a glycoprotein enzymatic digest is quite common today, using a variety of MS approaches (30,33,45–50). This allows for assigning the specific location of each attached glycan, the amino acid sequence of the peptide, and the exact structure of the glycan, just using MS methods. However, TDS does not appear to be 100% successful for glycoproteins yet, but can provide N-terminal and C-terminal sequences for many proteins (

). At times, such TDS methods can also identify other (internal) peptides coming from collision-induced dissociation (CID) in the MS system. However, there are no total TDS glycoprotein sequences yet demonstrated in the open literature. These TDS methods are really of several (at least three) basic designs, one using MALDI-TOFMS with in-source decay (ISD) and post-source decay (PSD) of the original glycopeptides or, in some instances, proteins (Bruker Daltonics, Billerica, Massachusetts) (35,36,42–44). Another involves a combination of CID with electron transfer dissociation (ETD) or electron capture dissociation (ECD), using ESI-MS, usually with reversed-phase LC interfacing, often described using an Orbitrap instrument (Thermo Fisher Scientific, Waltham, Massachusetts) (42–48). Yet another viable approach, has used a combination of ion mobility spectrometry (IMS) together with TOFMS, as in the Synapt instrument (Waters Corp., Milford, Massachusetts), using high and low energy dissociation of the intact glycopeptides to denote amino acid sequence, location of each glycan, and the specific glycan structure.  

An approach that will work, though not yet fully demonstrated for a complex mixture of glycoproteins in a biotech drug substance, would be MDLC or 2DGE together with just TDS. This has been demonstrated for a variety of simpler proteins (Figure 4), but it has not yet been demonstrated for even a simple mixture of glycoproteins with complete characterizations. Figure 4 illustrates the c- and y-fragment ions coming from the intact protein, with confirmation of the N- and C-termini peptides, using ISD on a MALDI-TOFMS type MS system (43). However, the upper MW range of TDS is today perhaps too low for most glycoprotein variants of biotech interest (31,42–50). It is undoubtedly a matter of time before these limitations will be overcome, and then there will be no reason whatsoever that mixtures of intact glycoproteins cannot be fully characterized by TDS alone (42–50). 

Figure 4: Top-down protein sequencing of proteins without bottom-up analysis (43). Sample A: reISD spectrum annotated by fragment ions after successful assignment of the established protein sequence. BioTools confirms the N- and C-termini on the intact protein (43).			

So, then, how does TDS really work? It works similarly to doing complete glycopeptide sequencing, via a combination of CID, ECD, or ETD and other fragmentation methods (for example, ISD), that will provide three things: amino acid sequence, location of each glycan on specific amino acids, and the precise structure of each glycan. This has been described several times by various groups for glycopeptides, but remains to be described for intact glycoproteins, using only TDS. This is basically because it is not yet possible using MS-MS alone, whatever the fragmentation mechanism, to totally sequence the entire amino acid backbone. The protein fragments in the MS and basically the N- and C-termini are sequenced, but putting back together the other peptides is often impossible. The coverage possible appears to vary, protein-to-protein and method-to-method. Thus, in most instances, a combination of TDS and bottom-up peptide mapping is still needed to obtain the total sequence, location, and nature of each modification. As one of our colleagues has so aptly put it: "Top-down only workflows do not deal with the fact that proteins have low levels of microheterogeneities that, left unresolved, are not accessible by top-down methodologies. Top-down can tell you where something of high occupancy resides on a protein, but seldom addresses the necessary quantitative or dynamic range requirements needed for biotherapeutic characterization work today" (51).  

In bottom-up glycoprotein analyses, the intact protein is enzymatically digested, as usual, to form a mixture of glycopeptides (40,41,45–50). These are then usually resolved by reversed-phase LC–ESI-MS, and using a combination of CID and ETD or ECD (or other MS methods), depending on the specific MS system employed, information is derived for the peptide sequence, location of each glycan attachment, and the specific glycan structure (45–50). It still remains to put those glycopeptides together with other peptides, nonglycosylated, back together again into the original protein's correct sequence. Figure 5 then illustrates the type of information obtained by both CID and ETD or ECD, for typical glycopeptides, now showing mainly the glycan sequences and structures observed for this single species (47). However, this figure does not show the actual glycan sequencing data but only shows the final, intact mixture of glycopeptide variants detected. Glycan sequencing routines are common today, and there are various software platforms and databases (for example, GlycoSuite [Tyrian Diagnostics Ltd., Sydney, Australia] and others) that assist in such delineations. The amino acid sequence and location of glycan was done in a separate MS experiment. Each MS vendor has a somewhat different approach to doing glycopeptide characterization, as above, depending on the nature of their MS products. Some use IMS-TOF–TOFMS (Waters), others may use an Orbitrap instrument with CID/ETD(ECD) (Thermo Fisher Scientific), while still others may use an ion-trap with ETD and proton transfer reaction (PTR) (Bruker) methods. 

Figure 5: Complete sequencing of a typical glycopeptide in a peptide map, using CID with ETD (47). Deconvoluted full MS spectrum of bovine alpha1-acid glycoprotein bi-antennary peptide 91CVYNCSFIK99.			

Let us consider another, quite recent, such approach for doing top-down posttranslational modification (PTM) characterization, that described by Bruker, which uses its HCUltra ion trap instrument with ECD and PTR reactants (42,43,52,53). In this approach, intact proteins can be identified and sequenced without any prior enzymatic digestion. However, the sequence coverage is not 100%, and the specific sites of phosphorylation for the parent protein, pH4 histone, were not fully and absolutely identified. However, fragments all across the 100–11,331 Da of the protein are visible, with a sequence coverage of >80%. It does not appear that other approaches can yet 100% characterize complex glycoproteins, using top-down methods alone, no matter what MS instrument is available. When that does come about, it remains to be seen how the regulatory agencies will respond in future characterization requirements for complex mixtures of glycoproteins in new filings? 

However, some very recent posters by Bruker at the ASMS 2010 meeting describe complete or very nearly complete, amino acid sequencing of fairly large proteins (52,53). In the case of a 1.36-kDa protein, camelid nanobody, it was completely sequenced, N- to C-terminal (52). This protein was not present in any known database, and thus did not lend itself to conventional, bottom-up methods. This is perhaps the most complete large-protein sequencing described to date, but again there was no indication of modifications, their location or specific nature.  

Finally, What About Doing Just Shotgun Proteomics or Bottom-Up Methods for Glycoprotein Characterization?

For many years, still today, most proteomics researchers have tried to identify individual proteins or glycoproteins based on shotgun methods, also known as bottom-up alone (57,58). Indeed, we would conjecture that most of the existing literature in proteomics areas has relied and still relies on solely bottom-up methods. That is, most researchers do not perform both top-down and bottom-up proteomics at the very same time on the same proteins (40,41). The world today relies almost entirely on bottom-up peptide mapping methods, together with database searching, Mascot or MOWSE scores and percent coverage numbers to "identify" proteins. This is nonsensical. Bottom-up methods rely almost entirely on identifying "all" the peptides derived from the original, intact protein of interest, be this after 2DGE or MDLC separation methods. However, it is impossible to know when one has indeed detected "all" the possible peptides coming from an unknown parent protein. The absence of evidence is not evidence of absence.  

And thus, when the database assigns a protein as a hit, it gives a score or number, which only suggests that may or may not be the right parent protein structure. That confidence can vary from 20% to 99%, but even a scoring of 99% can still be misleading or downright wrong. Without measuring the intact protein's true molecular weight by top-down MS methods, one never has true confirmation of what the database is suggesting as the "most likely" hit. Hence, it is very possible, if not likely, that much of the existing literature in proteomics areas may have incorrectly and inadvertently misidentified the true protein structure. As a speaker at the recent ISPPP 09 meeting cleverly put it: "Shotgun proteomics is full of holes." This would be especially true for glycoproteins, given the diversity and complexity of the possible component glycopeptides in a large parent glycoprotein. Hopefully, this situation should or shall change for the better in the near future.  

And, to finish (at last!), clearly to characterize a biotechnology-derived glycoprotein using only bottom-up methods is also a doubtful approach to come up with the correct structure. It is a very risky and questionable approach, even when we know the original DNA sequence that was installed in the cell expression system to generate a known sequence of amino acids for the glycoprotein of interest. That is not very helpful once there are PTMs that occur between the protein's translation and final expression for identification, often involving assimilation (combining peptides from differently expressed proteins in the cell), truncation or clipping (enzymatic removal of pieces of the originally expressed protein structure) or amino acid changes, just to mention a few possible PTMs. 

Table I summarizes a variety of methods used in the detection, purification, and structural analysis of glycoproteins (63). However, this is far from complete, as there are numerous other electrophoretic methods, for example, that have become extremely useful, such as HPCE, 1DGE, 2DGE, and others. 

Table I: Some important methods used to study glycoproteins			

 Co-Editor Ira S. Krull is an Associate Professor of chemistry at Northeastern University, Boston, Massachusetts, and a member of 

 is a biotech CMC consultant and an associate professor with the Department of Chemical Engineering,  Indian Institute of Delhi, India. He is also a member of 

(1) The Future of Biotech. The 2010 Guide to Emerging Markets and Technology, BioWorld Today Publishers, 2010.  

(2) Massachusetts Biotechnology Council, Cambridge, MA, 

(3) BIO, Biotechnology Industry Organization, 

BioTerminology, a Guide to the Biopharmaceutical Lexicon, First Edition

 (Waters Corporation, Milford, Massachusetts, February, 2009). 

(6) Scientific American WorldView, A Global Biotechnology Perspective, 

The Future of Biotech, The 2010 Guide to Emerging Markets and Technology, 

BioWorld Today (Scientific American, Inc., Publishers, 2009.  

Posttranslational Modification of Proteins. Expanding Nature's Inventory

 (Roberts and Company, Publishers, Englewood, Colorado, 2006). 

Biopharmaceuticals, Biochemistry and Biotechnology, Second Edition

 (John Wiley & Sons, Ltd., West Sussex, UK, 2003), Chapter 10. 

The Challenge of CMC Regulatory Compliance for Biopharmaceuticals

 (Kluwer Academic/Plenum Publishers, New York, 2004), Chapters 1-2, Table 2, Chapter 1.  

Biotechnology Today

Glycoproteins appear to have become the most common biopharmaceutical product today, and they also seem to be increasing in popularity and importance with time.

Glycoproteins and Biotechnology — Recombinant Biopharmaceutical Products: Identification, Characterization, and Chemical Equivalency?

John Dolan has been writing "LC Troubleshooting" for 

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

In the 1970s, Rohrschneider made an extensive study of the properties of various solvents (1). Among other things, he showed that, in some cases, when two solvents were mixed, properties of a third solvent could be obtained.

This suggested that some subset of all possible solvents could be used for chemical purposes. Around the same time, Lloyd Snyder was interested in the various solvent properties that were important in liquid chromatography (LC) separations, especially reversed-phase separations (2,3). From other studies, he knew that three properties were of particular significance in terms of selectivity: the acidic, basic, and dipolar nature of the solvents. Snyder recalculated Rohrschneider's data based on these three properties and plotted them in what is now known as the solvent selectivity triangle. A subset of these data is shown in Figure 1, where each solvent is plotted according to its proportion of acidic, basic, and dipolar properties. The heavy dots show the central tendency of several solvent classes, some of which I have labeled.

In principle, three solvents — one with 100% acidity, one with 100% basicity, and one with 100% dipolarity — would be located at the extreme corners of the triangle. If these solvents existed, we should be able to blend them in different proportions and get the properties of solvents lying within the boundries of the triangle. Of course, such solvents do not exist. Most solvents have a mixture of acidic, basic, and dipolar properties, so they lie away from the corner and usually away from the edges of the triangle. Since the desired solvents don't exist at the corners of Figure 1, we might consider other solvents that are close to the corners — solvents that exhibited primarily acidic, basic, or dipolar properties. For example, carboxylic acids (RCOOH in Figure 1) are quite acidic, amines (for example triethylamine) are basic, and the chlorinated solvents (for example, dichloromethane, CH

) have dipole properties. If we were to blend these three solvents together, we might expect to get intermediate properties. But again, we are foiled, because the chlorinated solvent doesn't mix with the others and two phases result — not a desirable mobile phase characteristic! We can continue working our way toward the center of the triangle, searching for solvents that are miscible and have a dominance of one of the desired properties. For example, alcohols (such as methanol) have some acidic properties, ethers (for example, tetrahydrofuran) have basic properties, and nitriles (for example, acetonitrile) have dipole properties. These three solvents are mutually miscible and form a subset (shaded area) of the total triangle. It would be nice to conclude that because of Snyder's work, these three solvents were selected as the solvents of choice for reversed-phase separations, but this would be a bit presumptuous. By the time this work was completed, these three solvents were already the most popular solvents, but now we know why they had been found to be most effective. The enhanced acidic, basic, and dipolar properties of one solvent can assist certain chemical interactions better than another mobile phase solvent, so that difficult-to-separate solutes can now be separated.

Figure 1: Classification of solvent properties according to their acidic, basic, and dipolar nature. Adapted from references 2 and 3.			

One popular application of the solvent selectivity triangle is based on a set of seven experiments using various solvent blends. This strategy formed the basis of the Sentinal system introduced by DuPont when they were in the LC business and was adapted as the Perkin-Elmer Solvent Optimization System (PESOS). These products enabled automated or semiautomated selection of the best solvent to separate a particular sample. The concept of this approach is shown in Figure 2. The methanol–acetonitrile–tetrahydrofuran solvent triangle of Figure 1 is adjusted to an equilateral triangle for convenience. One solvent is chosen to start with and its concentration in water or buffer is adjusted to get retention in the desired range in an effort to achieve a separation, for example, 50:50 methanol–buffer. Next, a second solvent is adjusted to give similar overall retention, for example, a 40:60 acetonitrile–buffer mixture gives similar retention. Then the third solvent is blended with buffer to get similar retention — for example, a 30:70 tetrahydrofuran–buffer mixture in the present situation. By appropriate adjustment of the organic–buffer ratio, the three mobile phases should give very similar retention times when the entire separation is considered — individual retention times are expected to vary between solvent systems. These are experiments 1, 2, and 3 at the corners of the triangle in Figure 2. Next, 1:1 blends are made of the mobile phases from each corner to generate the runs 4, 5, and 6 on the edges of the triangle. Finally a 1:1:1 blend formed the central experiment, 7, to complete the seven-experiment set. Retention data from each of these experiments could be entered into the optimization program and results for conditions between the experimental points could be calculated to find the best possible separation, often a quaternary blend of methanol, acetonitrile, tetrahydrofuran, and buffer.

Figure 2: The solvent-selectivity triangle approach to method development.			

Alternatively, the seven chromatograms could be printed and set out in the same triangle pattern with visual interpolation to identify the best separation conditions. An example of this approach is shown with the simulated chromatograms of Figure 3. The methanol–buffer run (run 1) separates only four peaks of the five-component mixture — the last two peaks are merged. The acetonitrile–buffer run (run 2) nearly separates the last two peaks, but the second two have merged and reversed. The tetrahydrofuran–buffer run (run 3) has its problems, too — the last two peaks are nicely separated, but the second two are still a problem. It is reasonable to assume that intermediate mobile phase conditions will give intermediate separations when a mobile phase condition is varied. So, if we were making these runs one at a time and examining the results before proceeding, we might skip run 5, because it is obvious from runs 1 and 2 that an intermediate separation of the last two peaks will be no better than for the binary mobile phases. The same prediction could be made for the second two peaks in run 6 based on runs 2 and 3. In run 4, a 50:50 blend of the mobile phases used in runs 1 and 3 is able to separate all five peaks to baseline by giving up part of the excess separation of the first three peaks observed in run 1 and part of the excess separation of the last two peaks from run 3. The quaternary blend of 1:1:1 of runs 1, 2, and 3 to form run 7 is unsatisfactory too. 

Figure 3: Simulated partial chromatograms for various mixtures of methanol, acetonitrile, and tetrahydrofuran with buffer. Chromatograms positioned to correspond to Figure 2.			

This solvent triangle approach of method development was quite popular in the 1970s and 1980s. It was fairly intuitive to use and generated useful methods. It was equally good for developing isocratic and gradient methods. However, because the mobile phases often comprised all three solvents plus water, they tended to be complex and the more components contained in a mobile phase, the more things can go wrong. This, coupled with the less reproducible nature of the columns of the day, resulted in methods that often needed constant tweaking to keep running. As we'll see in next month's "LC Troubleshooting

 further understanding of reversed-phase retention behavior introduced the solvent-strength optimization technique that largely supplanted the solvent selectivity triangle as the most popular approach to method development. But just because it isn't as popular today doesn't mean that we should discard the solvent triangle approach — it is a powerful tool that can help to solve difficult problems, especially after simpler techniques have failed. 

We've seen how the solvent selectivity triangle helps us understand why acetonitrile, methanol, and tetrahydrofuran have become complementary and preferred solvents for reversed-phase LC mobile phases and how the triangle can serve as an organizational tool for using solvent type to adjust peak spacing in a chromatogram. There are other uses for the solvent triangle as well. 

We can use the solvent triangle to help us make wise choices for the selection of additional solvents to test. For example, if we find that methanol in the mobile phase does not give us the peak spacing we need, we might be tempted to try another alcohol — perhaps ethanol or propanol. But the properties of all of these alcohols cause them to be located in the lower left corner of the triangle. This suggests that it is unlikely that we'll find dramatically different chromatographic properties by changing to a different alcohol. We'd be much better moving to an entirely different part of the triangle if our goal is to use solvent selectivity to move peaks around in the chromatogram. On the other hand, there may be other reasons why we might switch from one alcohol to another. For example, larger alcohols, such as propanol, tend to be less denaturing to biomolecules than methanol — it would be reasonable to expect similar, but not identical, chromatographic properties when making this change. As another example, we might find that the basic properties of tetrahydrofuran are favorable for a particular separation, but its other properties, such as relative instability and high UV absorbance, make it less than ideal for some applications. It is possible to successfully substitute another ether, such as methyl 

-butyl ether (MTBE), for tetrahydrofuran in some cases. (MTBE, however, is not fully miscible with water, so it may require a cosolvent in the mobile phase.) 

Still another application of the solvent triangle is to identify alternate mobile phase solvents if the supply of solvent is limited. In the alcohol example above, we saw that the properties of the different alcohols were sufficiently similar that a change in alcohol type is unlikely to be an advantage when trying to change selectivity. Conversely, this suggests that we could change from one alcohol to another if necessary for other reasons. For example, if the availability of high performance liquid chromatography (HPLC)-grade methanol became scarce for some reason, we might reasonably switch to ethanol with the hope of obtaining a similar separation. For various economic, political, and practical reasons (see reference 4) HPLC-grade acetonitrile has become quite expensive in the last few years. It would be nice to find another solvent that had similar properties to use as an alternative. Unfortunately, there aren't any readily available solvents in the dipole corner of the triangle that fulfill this requirement, so we are stuck with acetonitrile. If acetonitrile is necessary to get the desired separation and no alternative solvents are available, we have to approach the problem from a different angle. We could change from using a 150 mm × 4.6 mm column to a 150 mm × 2.1 mm column. The reduction in column diameter will allow us to lower the flow rate by a factor of 5 and get the same retention time — for example, from 1 mL/min to 0.2 mL/min. This would reduce the acetonitrile consumption by fivefold, as well. Or we could switch from a 150 mm × 4.6 mm column packed with 5-µm particles to a 100 mm × 4.6 mm column filled with 3-µm particles. Both of these columns will generate approximately 10,000 plates with real samples, so the 100-mm column should reduce the run time by one third — a 15-min run would drop to 10 min, with a corresponding saving of solvent. Or we could combine the two changes and switch from the 150 mm × 4.6 mm, 5-µm column to a 100 mm × 2.1 mm, 3-µm column and reduce the solvent consumption by 5 × 3 = 15-fold. Of course, if these changes were made, we might have to make some other adjustments, such as reducing the amount of injected sample or adjusting the system plumbing to reduce extracolumn volume.  

The solvent selectivity triangle helps to simplify the number of solvent choices that are likely to be beneficial to explore during LC method development. It also gives us an organizational framework when exploring solvent selectivity as a means of resolving difficult-to-separate peaks. Finally, we can get some insight into the pros and cons of alternate solvents if we want to make a change. Next month's "LC Troubleshooting" will consider an alternative to solvent-type selectivity: solvent-strength selectivity. 

"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 

LC Troubleshooting

Methanol and acetonitrile are the two most popular organic solvents to use in the mobile phase for reversed-phase liquid chromatography (LC) separations. Tetrahydrofuran is also used to a lesser extent. Why are these solvents the most popular ones? Are there alternative solvents that should be tried? This month's "LC Troubleshooting" examines these and some other aspects of solvent selection for LC.

Selectivity in Reversed-Phase LC Separations, Part I: Solvent-Type Selectivity

Issue PDF

Click the title above to open the LCGC North America December 2010 regular issue, Vol 28 No 12, in an interactive PDF format.

LCGC North America-12-01-2010