Fluorescence detection can be a strong alternative to ultraviolet or other detectors for some compounds. 

	This month we continue our “LC Troubleshooting” series examining how different components of the liquid chromatography (LC) system work. In the first three installments we looked at pumps, mixing, degassing, and autosamplers (1–3). Last month we began looking at detectors, covering the variable-wavelength and diode-array ultraviolet (UV) detectors (4). In the present discussion, we begin looking at detectors that are less widely used than UV detectors. The first of these is the fluorescence detector. 

	UV detectors rely on the characteristics of many molecules to absorb energy when exposed to certain wavelengths of light. For some molecules, the wavelength of absorption may be quite specific (for example, anthracene has a strong absorbance at 251 nm). Some classes of molecules have a common absorbance region, such as proteins and peptides. The aromatic rings of some of the amino acids in these molecules absorb strongly at 280 nm, making this wavelength a common choice for proteins and peptides. Many organic molecules exhibit strong UV absorbance at wavelengths <210 nm, thus allowing a detector set at 200 nm to act somewhat as a “universal” detector.

	In some cases, when the light energy is absorbed by a molecule, it raises some of the electrons to an excited state. When these electrons return to the ground state and light is emitted, the process is referred to as 

. Fluorescence detectors rely on this molecular property for detection.

	Fluorescence is much less common than UV absorbance and because for a given molecule, both the excitation and emission wavelengths are specific, fluorescence detectors can be very selective. This selectivity sometimes allows the detection of a molecule by fluorescence when interferences present in the chromatogram would not allow detection by UV absorption. An example of this selectivity is shown in the chromatograms of Figure 1, which compares response of riboflavin (Rbfl) with UV and fluorescence detection. The chromatograms in Figures 1a and 1b were obtained using a UV detector set at 365 nm. The chromatograms in Figures 1c and 1d show the same samples by fluorescence with an excitation wavelength of 365 nm and emission at 530 nm. (It should be noted that fluorescence is not completely efficient, so less energy is emitted than is absorbed. This means that excitation must be at a lower, more energetic wavelength than the higher, less energetic emission wavelength.) On the left of Figure 1 are chromatograms of injection of 5 ng of riboflavin reference standard. The peak is clearly present in both chromatograms, but it can be seen that the fluorescence detector gives a peak approximately 7x as large (although the signal-to-noise ratio is approximately half of this increase). There is a response benefit for the fluorescence detector in this case, but a greater benefit is in selectivity when a real sample is run. On the right side of Figure 1, riboflavin is determined in a dog food extract. The riboflavin peak is clearly visible by fluorescence (Figure 1d). With UV detection (Figure 1b), on the other hand, one cannot say with confidence that riboflavin is present, and it certainly cannot be quantified. Here the selectivity of the fluorescence detector allows riboflavin to be determined in the presence of interferences that have strong UV absorbance but do not fluoresce at the selected wavelengths.

					Figure 1: Comparison of UV and fluorescence response for riboflavin (Rbfl): (a,b) UV detection at 365 nm; (c,d) fluorescence detection at 365 nm excitation and 530 nm emission. (a,c) Injection of 5 ng riboflavin standard; (b,d) riboflavin in dog food extract. Cation-exchange separation on a 100 mm x 2.1 mm Zipax-SCX column with water as mobile phase at 1.0 mL/min. Adapted from DuPont data presented as Figure 4.14 in reference 5.

	When UV detection is used, most workers avoid derivatization of the analyte because of inconvenience and possible increase in variability of the analytical results. However, it is much more common to use derivatization with fluorescence detection. Derivatization can generate fluorescent analyte molecules, increasing method sensitivity and selectivity. One popular reaction is with 

-phthaldialdehyde (OPA), which reacts with primary amines to form fluorescent isoindole derivatives of compounds such as proteins and peptides. This reaction can be done as a precolumn derivatization or on-line as a postcolumn reaction.

	The conceptual schematic of a generic fluorescence detector is shown in Figure 2. An intense lamp is needed to provide excitation energy in the UV and visible ranges. A xenon flash lamp is the typical choice, providing light at wavelengths ranging from 200 nm up to 700–900 nm. If the detector flow cell were a linear design, as is used in the UV detector (for example, Figure 3a in reference 4), the photodetector would have to filter out the light at the excitation wavelength. However, the light emitted in the fluorescence process is emitted in all directions, so monitoring emission at a 90° angle to the excitation wavelength is just as effective-plus it eliminates the problem of interference and minimizes scattered light. (Although not a common design, it is possible to configure the detector with a different cell design so that UV absorbance can be monitored in a straight-through optical path and fluorescence at right angles.) The construction of the flow cell is not as robust as that for a UV detector, and typical pressure limits are 3–35 bar (40–500 psi), so be sure to check the specifications for your detector and ensure that the pressure limits are not exceeded. The photons generated during fluorescence are monitored by a photosensitive device, which could be a photodiode, but more commonly is a photomultiplier tube (PMT).

					Figure 2: Schematic of a generic fluorescence detector.

	The excitation and emission wavelengths typically are selected using a pair of monochromators. The diffraction grating in the excitation monochromator is rotated to select the desired excitation wavelength and direct it through the flow cell. In a similar manner the desired emission wavelength is selected to allow the PMT to monitor the appropriate wavelength. Either or both monochromators can be used in a scanning mode. This process has the same weakness of using scanning with a variable-wavelength detector (see reference 4) because of changing concentration of the analyte during scanning. However, scanning techniques are useful when trying to identify the best excitation and emission wavelengths if these characteristics of the sample molecule are not already known. Alternate designs of detectors have been made in the past in which less-expensive optical filters are used for one or both monochromators, but the most popular instruments today are the dual-monochromator units.

	The key to effective use of the fluorescence detector is to find the best excitation and emission wavelengths. The excitation maximum for fluorescence and the UV absorbance maximum are usually very close to each other. For this reason, a UV spectrum obtained from a diode-array detector or the literature is a very useful tool to identify the desired excitation wavelength. Typically, a 20-nm bandwidth is specified for the monochromators, so a wavelength setting close to the ideal wavelength should provide an acceptable setting. After the excitation wavelength is selected, the emission wavelength can be scanned to identify the desired setting. For routine operation, the excitation-emission wavelength pair can be set for each peak in a chromatogram and switched during the run to get the best response for each peak.

	The process of optimizing the detector settings will depend a bit on the design of the detector, so be sure to consult the operator’s manual that comes with your particular detector-there usually are excellent instructions on how to set up the detector and use its features. I found a free e-book produced by one manufacturer (6) that includes a very useful discussion of fluorescence detection and some of its nuances.

	After the excitation and emission wavelengths and other chromatographic conditions are chosen, it is wise to check for any background response, just like you would do for an LC method with UV detection. An example of how background response can differ is shown in Figure 3 (6). In this case, three water sources were chosen-deionized water, water purified with a laboratory water purification system (HPLC-grade), and commercially prepared LC–mass spectrometry (MS)-grade water. For the test of Figure 3, water was run at a constant flow rate and following a short autozero time, the response was recorded using the excitation and emission wavelengths chosen for each of five polynuclear aromatic hydrocarbons in the sample. You can see that the background was much higher for the first and third conditions than for the remaining conditions. Also there was noticeably higher background in all cases when deionized water was used when compared to high performance liquid chromatography (HPLC)-grade or LC–MS-grade water. These results are analogous to those observed for UV detection in that higher-purity reagents give lower background signals. Fluorescence of the analytes as well as background fluorescence can be influenced by the molecular structure of the analyte as well as the chromatographic conditions and column temperature.

					Figure 3: Comparison of background fluorescence of different water sources under various conditions. Blue: deionized water; red: lab-prepared HPLC-grade water; green: commercial LC–MS grade water. In each case, the baseline is autozeroed, then allowed to collect data at the excitation–emission wavelength pair shown above each plateau. Wavelength pairs are those used for analysis of a five-component sample of polynuclear aromatic hydrocarbons. Figure data adapted from Figure 26 in reference 6.

	When the chromatograms of Figure 1 were examined earlier, it was seen that background components in the dog food sample interfered with detection by UV (Figure 1b), but the selectivity of fluorescence allowed detection of riboflavin in the presence of these interferences. In some cases, however, interferences in the sample can suppress the fluorescence of an analyte in an analogous manner to ion suppression in LC–MS detection. An interesting example of fluorescence suppression, or quenching, is shown in Figure 4 (7). In this example, naphthalene is injected approximately every 2 min and detected by UV (254 nm) or fluorescence (250-nm excitation, 340-nm emission). Initially the mobile phase is sparged with helium (He) and the baseline is steady with a strong naphthalene response. Next, the mobile phase is sparged with air while injections continue. The UV baseline rises a bit because of UV absorbance of oxygen and the signal rises the same amount. In the case of fluorescence, however, the baseline stays steady, but the signal drops gradually to about 20% of the original signal. This drop results from oxygen quenching the fluorescence of naphthalene under these conditions. When helium sparging is resumed, oxygen is displaced from the mobile phase and both chromatograms return to normal. In practical application, it is unlikely that the changes in the UV chromatogram would be noticed, because it is common to autozero at the beginning of each run, which would mask the observed offset. The fluorescence problem could be a very vexing one to solve. At the time the study was done, it was common to degas solvents with helium, either in a batch mode or continuously. Imagine if the mobile phase were sparged in the morning with helium, then placed on the system. Over the course of the day, air would be absorbed gradually into the mobile phase and the signal would gradually drop. The next day, the pattern would repeat itself. It might take a long time to troubleshoot that problem!

					Figure 4: Example of fluorescence quenching of naphthalene because of oxygen dissolved in the mobile phase. Upper chromatogram: UV detection at 254 nm; lower chromatogram: fluorescence with 250-nm excitation and 340-nm emission. Mobile phase is sparged with helium (He) or air, as indicated. Adapted from data of reference 7.

	Today most systems use in-line vacuum degassing instead of helium sparging. Vacuum degassing drops the air burden of the mobile phase enough that outgassing problems in the pump or mixer are not important, but it does not selectively remove oxygen from the mobile phase. Oxygen makes up approximately 20% of air, so dissolved oxygen will still remain in the mobile phase, although at a reduced level. It would be interesting to repeat the experiment of Figure 4 using an in-line degasser to see if enough oxygen remained to cause fluorescence suppression. If suppression were observed, it would be a good idea to sparge the mobile phase with helium or nitrogen to displace the oxygen before the in-line degassing took place.

	Fluorescence detection can be a powerful tool to increase sensitivity and selectivity of LC analysis for compounds that fluoresce. It can be a good alternative to other detectors for such compounds. I remember a method we developed in our laboratory for a porphyrin drug that strongly fluoresced. Our client insisted that we develop a LC-MS method for it. The method was a nightmare-sample preparation was difficult, ion suppression was prominent, and the signal was weak. When we received clinical samples, the fluorescence of the drug made the plasma visibly green. We had explained to the client that fluorescence with emission at 440 nm gave a stronger and interference-free signal with nothing more than protein precipitation for sample preparation. But the client wanted LC–MS . . . and the client is always right!

Introduction to Modern Liquid Chromatography,

 2nd edition (Wiley, Hoboken, New Jersey, 1979).	

 (Thermo Fisher Scientific, Germering, Germany, 2013) 

http://www.dionex.com/en-us/webdocs/113591-Handbk-FLD-3000-Method-Develop-AN70302_E.pdf

		S.R. Bakalyar, M.P.T. Bradley, and R. Honganen, 

“LC Troubleshooting” Editor John Dolan has been writing “LC Troubleshooting” for 

 for more than 30 years. One of the industry’s most respected professionals, John is currently the Vice President of and a principal instructor for LC Resources in Lafayette, California. He is also a member of 

’s editorial advisory board. Direct correspondence about this column via e-mail to 

LC Troubleshooting

Fluorescence detection can be a strong alternative to ultraviolet or other detectors for some compounds.

How Does It Work? Part V: Fluorescence Detectors

This is part II of a series of tutorials that explain, in the simplest manner, how statistics can be useful, even to chromatographers who normally find statistics difficult, with a minimal understanding of its features. This installment explains linear regressions.

	Here in part II, we focus on linear regressions. They are especially useful in developing quantitative structure–retention relationships (QSRR) (1). Linear regression is the method most often used by chromatographers, although it is often a simple linear regression with a unique variable (in cases of calibration curves for quantitation, for instance) (2), while the linear regressions used in QSRR are most often multiple, and thus comprise several variables.

	Let me start with the simplest form-a single variable. Before a linear regression, it is necessary to eliminate any problem in the distribution of variables. So you must first examine your data to identify potential issues. First, be careful with extreme values on one feature because they may act as lever points. For instance, in a calibration curve based on ultraviolet (UV) absorbance, the largest concentration values must not exceed the linearity range otherwise Beer’s law would not be respected anymore. Looking at Figure 1, it is easy to understand that an extreme value (heavy weight in this figure) has little influence on the slope (of linear trend or seesaw in this figure) when it is in the middle of the distribution, but will make the slope vary (or seesaw move up or down) when it is positioned at the edges.

					Figure 1: Lever points in linear regressions.

	Second, the range covered by each feature must be sufficiently wide otherwise the contribution of this feature cannot be assessed. For instance, if you want to investigate the effect of temperature on extraction conditions, you should make temperature vary in a significant range to cause changes in extraction yield.

	Let us suppose that you have measured chromatographic retention data for 

 analytes, of which the structures are known to you. The analytes may be described by several features. One that is often known or of which the knowledge is often desired, is octanol–water partition coefficient (log P). For instance, in a reversed-phase high performance liquid chromatography (HPLC) system, there should be a more-or-less linear tendency existing between retention (log 

 values) and log P. All HPLC chromatographers use this notion intuitively, as they expect polar analytes (low log P values) to be eluted first from a reversed-phase HPLC system, and nonpolar analytes (high log P values) to be eluted last.

	Similarly, gas chromatography (GC) analysts expect that elution order will be somewhat in accordance with volatility. An example is shown in Figure 2a: The correlation is not perfect (as indicated by the regression coefficient, which you would prefer to be closer to 1), but a linear tendency clearly appears. The regression line is the one that minimizes the sum of distances between the line and all points considered to calculate this line. The line indicates the theoretical model, or the positions where all points should fall, if the model was perfect. The correlation coefficient provides some indication on the quality of the fit, but the significance of this term is clearly overestimated by many users: correlation coefficients are strongly affected by lever points.

					Figure 2: (a) Simple linear regression to correlate log P values and retention data measured in reversed-phase HPLC; (b) normalized residuals.

	Other interesting figures resulting from linear correlation are provided by a Student 

-test of significance (indicating the probability that a significant correlation exists between 

) and standard error of the regression (or standard error in the estimate). For instance, in the present example, the standard error was 1.0. To simplify, this value can be interpreted as the average error in predicted log P value, which would be about one log P unit (plus or minus). It is also useful to look at the graph of residuals (Figure 2b) where possible outliers will appear: residuals show the deviation between measured value (log 

) and predicted value based on perfect correlation to log P. Widely scattered points, with normalized residuals preferably comprised between -2 and +2 are a good sign that the data were initially well distributed and that no point is acting too strongly on the final result.

	In the case of Figure 2, the linear regression is a simplification of the chromatographic process. Clearly, retention in this HPLC system cannot be fully explained by hydrophobic partitioning, and other factors should be considered. That is why more complex models have been developed, based on multiple linear regressions with several variables instead of one, to achieve a more accurate description.

	Let me explain this concept with a very trivial example. Imagine a shopping center with two main corridors. In Figure 3, you will see that corridor A was definitely designed to retain women, with shops that are bound to attract their attention, while corridor B was arranged to retain male shoppers. Each of these corridors has different “interaction sites” to retain different sorts of “analytes” (shoppers). The retention of each “analyte” may also be affected by “operating conditions” like temperature (also called shopping fever). To fully understand the retention capability of each corridor, you must make sure that the people crossing them are varied. In other words, the “descriptors” to qualify the shoppers must cover a wide range (gender, age, social origin, amount available in their bank account, and so forth). If you now calculate a multiple linear regression of “time spent in the shopping center” versus “shopper descriptors,” you will obtain an equation that will allow you to compare your two corridors in terms of “strength of interactions” you have managed to establish with the shoppers (attractive to some, repulsive to others).

					Figure 3: Two corridors in a shopping center to retain women (corridor A) or men (corridor B).

	With chromatographic systems, we can proceed similarly: The shopping center is your column oven and the two corridors are two different chromatographic columns. Instead of considering only hydrophobicity, one might take into account the number of hydrogen-bond donors or polar surface area. These features may be known or measured, or calculated with modeling software. With a multiple linear regression analysis of log 

 on these molecular descriptors, you will thus establish an equation relating chromatographic retention to analyte features:

	It would tell you in which way retention is related to these features: positive or negative relation, and the extent of this relation. Note that in a reversed-phase HPLC system, hydrophobicity logically results in increased retention (positive term) while polarity results in reduced retention (negative terms). Possibly, if the descriptors are accurate enough and sufficient to explain all possible mechanisms, the model may have some predictive capability. In such a case, we should favor partial least squares (PLS) regression (which will be discussed in a future installment).

	The adequate features may be selected through a variety of processes. The one I tend to favor is to have a preselection of features, based on the chromatographer’s experience, but other methods exist to select features from large sets of molecular descriptors. The significance of each term initially included in the equation can be assessed with statistical tests (like Student’s 

-test) to reveal those features that may (or may not) be significantly related to the retention process.

	For multiple linear regressions, the variables should not be correlated. This is obviously inherent to the selection of analytes, and may occur when analytes belonging to only one structural family are included (like only polynuclear aromatic hydrocarbons). The retention range (extension of log 

 values) must also be wide enough, with analytes little retained, and analytes strongly retained. When all analytes are eluted close to the dead volume, it is impossible to measure what causes retention.

	Overall quality of the model obtained may be assessed with a number of factors, among which the most often used are Fisher F statistic, standard deviation, homogeneity of the residual values, and adjusted correlation coefficient.

	Any QSRR can only be validated if it is in accordance with good chemical sense and with the initial observations on the chromatographic data.

	One of the most famous QSRRs is the linear solvation energy relationship with Abraham descriptors, which I have used a lot myself. A lot of useful information (also pertaining to QSRR in general) can be found in the comprehensive review in reference 3.

	In the next session, I will discuss an important method you have all heard about: principal component analysis (PCA).

 is an Assistant Professor at the University of Orléans, in Orléans, France. Direct correspondence to: 

Special Feature

This is part two of a series of tutorials that explain, in the simplest manner, how statistics can be useful, even to chromatographers who normally find statistics difficult, with a minimal understanding of its features. This part explains linear regressions.

Statistics for Analysts Who Hate Statistics, Part II: Linear Regression and Quantitative Structure–Retention Relationships

In this installment of “Biotechnology Today,” we discuss the use of capillary electrophoresis–mass spectrometry (CE–MS) (with a sheath flow interface) for the analysis of intact proteins as well as protein digests. We discuss the unique aspects that the user needs to be aware of while testing biotherapeutics versus small-molecule drugs. We also highlight that the optimization of CE and MS parameters together results in the creation of a more robust and reproducible protein analysis approach. Finally, we list some of the most common errors that are likely to occur during CE–MS analysis and suggest ways to overcome them.

	The rapid advancements in the field of biopharma have increased the demand for analytical techniques that can accurately fingerprint the analyte (1). This challenge takes on a whole new dimension when we apply it to analytical characterization of 

, which are a set of complex molecules with diversity ranging from cytokines, growth factors, hormones, peptides, and proteins including monoclonal antibodies (mAbs) and fusion proteins (2,3). The recent spurt of patent expiries, together with the high cost of biotherapeutics have fueled the interest in biosimilars (4). The success of biosimilars depends on achieving the optimum trade-off between the need to demonstrate equivalence of quality, safety, and efficacy with the innovator product and achieving a reduction in the extent of clinical examination that the drug has to undergo. The latter is the primary contributor to the lower cost of development and commercialization of a biosimilar. A key component in achieving this balance is that of performing a thorough analytical characterization using a multitude of orthogonal, high resolution, analytical tools (5). Apart from the sophistication and orthogonality of the analytical methods used, it is also important that the analytical method be such that it allows the characterization of biologics while conserving any potentially present modifications and without changing the heterogeneity of the sample (6,7).

	As highlighted by several researchers, biotech products, as well as processes, are complex (8). The process complexity essentially pertains to the often uncharacterized feed material and raw materials that are used and the complex interplay of metabolic dynamics in the host organism, that makes it difficult to predict the influence of one, or a combination of variables, on process performance. The presence of such a large number of product related variants, including oxidized, reduced, deamidated, phosphorylated, and glycosylated forms in a typical biotherapeutic, can be attributed to the high complexity of these products. For accurate identification and quantification of these attributes, it is necessary to have high selectivity and sensitivity of the analytical methods used for their characterization (3,5,9). Conventional techniques, such as liquid chromatography (LC), size-exclusion chromatography (SEC), and gel electrophoresis have been quite successful in terms of their utility in characterization of protein therapeutics, and they still continue to be essential components of a typical, characterization armory of proteins. However, these techniques suffer from one or more limitations, with respect to analysis time, sample processing requirements, feasibility of automation, selectivity, and sensitivity for deciphering the minute changes in the overall protein architecture.

	During the past several decades, capillary electrophoresis (CE) has emerged as an analytical tool for characterization of large molecules (for example, protein therapeutics) (10). The major reason for this emergence is the high separation efficiency that it offers, due to minimal longitudinal diffusion, which otherwise contributes to the peak broadening during separations in other, contemporary analytical tools like high performance liquid chromatography (HPLC). In addition, since CE analysis is performed in a fused-silica capillary or in a coated capillary, under aqueous conditions (for minimizing protein adsorption), protein analysis can typically be achieved with minimal conformational changes. Finally, since CE separation is based on differences in charge, shape, and size of the analyte, it can be advantageously capitalized for differentiating among various post-translationally modified (PTM) forms of the proteins, which are likely to have different electrophoretic mobilities during separation.

	Mass spectrometry (MS) has also gained wide popularity as one of the most useful detection devices in proteomics research. For biotherapeutic characterizations, MS is widely used to gain information about the primary sequence of the protein, often via peptide mapping (also known as bottom up methods). MS is also used to better understand any possible modifications, higher order structure, glycosylation, and host cell protein (HCP) analysis (11,12). However, the real power of MS is truly harnessed once it is used in conjunction with a powerful separation technique such as LC or CE. However, interfacing CE with MS is technically not as straightforward as it is to couple LC to MS. The approaches developed so far to couple CE with MS can be divided into three categories (Figure 1). First, the use of an interface based on direct spray with a metal or metallized needle for electrical contact (Figure 1a). A second approach uses a liquid junction interface, in which the background electrolyte (BGE) in the capillary is hydraulically connected to the reservoir where an electrical contact is made (Figure 1b). The third generation of CE–MS coupling makes use of a coaxial sheath flow interface, whereby a conductive solvent flows around the outside of the capillary, and it provides a hydraulic connection to an electrical contact (Figure 1c). A comprehensive description of the various CE–MS coupling approaches is available in the literature (13–16). It should also be noted here that although the sheathless-low sheath flow interfaces were developed with superior sensitivity as compared to the sheath flow interface, the commercial use of the latter far exceeds the former (17).

					Figure 1: Illustration of the three commonly used interfaces. (a) Coaxial sheath liquid interface, (b) porous tip sheathless interface, and (c) electroosmotic flow driven sheath liquid Interface used for coupling CE to MS.

	In this article, we discuss the use of CE–MS with a sheath flow interface for the analysis of intact proteins as well as protein digests. We discuss the unique aspects that users need to be aware of while testing biotherapeutics versus small-molecule drugs. We also highlight that the optimization of CE and MS parameters together results in the creation of a more robust and reproducible protein analysis. Finally, we list some of the most common errors that are likely to occur during CE–MS analysis and suggest ways to overcome them.

	The therapeutic protein sample used in this study was an IgG1 mAb (derived from CHO cell culture). The mAbs were donated to us by a major domestic biotech manufacturer. Sample preparation reagents were of reagent grade, and CE–MS buffer reagents were of LC–MS grade.

	An Agilent capillary electrophoresis (G7100) system coupled to an Agilent 6230B time-of-flight (TOF) mass spectrometer through Agilent’s CE–MS adapter kit (G1603A) was used for performing CE–MS analysis. An Agilent 1260 Infinity II binary pump (G4220A) was used for sheath liquid delivery employing a splitter with a 99:1 ratio, effectively producing a sheath flow rate of 4 µL/min.

		Intact mAb: mAb sample (2 mg/mL) was buffer exchanged in 50 mM ammonium acetate, pH 4.

		mAb digest: mAb sample (1 mg/mL) was denatured, reduced, and alkylated according to the standard protocol (18). The reduced mAb was digested using trypsin (trypsin–protein = 1:50) and incubated overnight at 37 °C. Before analysis, the digested mAb was buffer exchanged to 50 mM ammonium acetate, pH 4.0. The digested sample was desalted before analysis using ZipTip according to the manufacturer’s protocol.

	The CE and MS parameters used for the mAb (intact and digest) analysis are summarized in Table I.

	CE–MS analysis of mAbs requires more thorough optimization of both separation (CE) as well as ionization (MS) parameters. This more thorough optimization is needed primarily because of the difficulty in creating the gas phase ions using electrospray ionization (ESI) for such a large compound without compromising the separation. Also, as discussed earlier, a reliable analytical method for protein analysis should be such that it preserves all the modifications that are potentially present in the molecule without introducing any artifacts during sample preparation and analysis. Thus, it would be desirable to perform the analysis in the native form of the protein. However, for achieving good electrophoretic separations, the molecule has to be sufficiently charged. The choice of buffer should also take into consideration the compatibility of the buffer and ionization of the analyte in MS.

	In LC–MS analysis of proteins, the sample is generally reconstituted or buffer exchanged in 0.05–0.1% trifluoroacetic acid or formic acid (5). However, in our experience better results were obtained when using ammonium acetate buffer of suitable strength and pH (depending on target analytes pI). A clean mass spectrum was obtained while maintaining the protein’s native structure. For the present case, the mAb sample reconstituted in 50 mM ammonium acetate (pH 4.0) was found to give satisfactory results.

	A large number of publications have highlighted the issue of adsorption of proteins onto the walls of the fused-silica capillary. This adsorption problem has been generally tackled using coated capillaries where interaction between the analytes and the capillary wall is minimal. For basic proteins, positively charged coatings in combination with a low-pH BGE minimize the adsorption on the wall. Similarly, for acidic proteins, negatively charged coatings in combination with a high-pH BGE have been found to minimize electrostatic and hydrophobic interactions of the proteins with the capillary walls (19). Coatings such as polyvinyl alcohol also help in eliminating electroosmotic flow (EOF) independent of the pH of the BGE. This helps in overcoming cumbersome washing procedures and significantly improves reproducibility of the migration time. For our study, we used a polyvinyl alcohol-coated capillary for the analysis of the intact and digested protein.

	Before analyzing samples by CE–MS, it is essential to properly condition the capillary so as to achieve reproducible migration of charged analytes under the applied field. For uncoated capillaries (for example, bare fused-silica capillaries) a flush of 1 N sodium hydroxide at 900 mbar for 20 min, followed by flushing with water and the method’s BGE for 10 and 30 min, respectively, at 900 mbar. For most of the coated capillaries-coated with polyvinyl alcohol, for example-a flush with sodium hydroxide is not required. It should be noted that the nebulizer pressure should be set at 0 psig and sheath liquid flow kept on while flushing the capillaries with sodium hydroxide because of the incompatibility of MS toward nonvolatile buffers. Alternatively, the flushing can be done without inserting the capillary into the mass spectrometer’s main body and setting the capillary voltage to 0 V. In either case, the frequency of MS source cleaning (generally with 60% acetonitrile) and MS capillary maintenance should be increased for retaining acceptable performance of the instrument. For mAb analysis, we found that flushing at 5 bar (high-pressure flushing) for 15–20 min greatly enhanced the results obtained. A stable current profile was maintained during the entire analysis period, as shown in Figure 2.

					Figure 2: Current profile during intact mAb analysis after conditioning the capillary (a) at 900 mbar and (b) 5 bar.

	The results of CE–MS with a sheath liquid interface greatly depend on the optimization of the sheath liquid composition as well as flow rates. Also, an important factor that needs to be kept in mind is that the polyimide coating of the capillary at the sprayer tip (MS system end) should be carefully removed. If the coating is not removed, the sheath liquid causes swelling and expansion of the polyimide coating to the sprayer tip, which can result in improper nebulization, reduced sensitivity, and current abnormalities. The sheath liquid composition that is widely reported in the literature is 50% methanol as the base solution with a suitable acid or base with a volatility comparable to that of the base solution (20). Isopropanol and ethanol are also used as the base solution. Acetonitrile, though well suited for ionization in MS, is known to deteriorate the polyimide coating of the capillaries and is therefore generally not used. Acids of choice are acetic acid and formic acid, and ammonium acetate and ammonium formate of suitable strengths are recommended as the base of choice (20). The optimization of sheath liquid flow rate is aimed at reaching the right balance between sensitivity and stability of measurements. Although high flow rates enhance the stability of the measurements, they also lead to multifold sample dilution, which affects the sensitivity. Keeping these aspects in mind, the sheath liquid flow rate is typically kept between 0.4 µL/min to 0.8 µL/min depending on the demands of the analysis (20). Most sheath liquid interfaces use an external pump with a recycling function that splits the flow in predefined ratios. If the same solution is used for several days, recycling increases the risk of dirt accumulation in the sheath liquid. As a rule of thumb, the sheath liquid should always be prepared fresh daily. Failing to prepare fresh sheath liquid is likely to result in significant difficulties during analysis (Figure 3).

					Figure 3: CE–MS results using (a) fresh sheath liquid and (b) recycled sheath liquid on third day. No spectra corresponding to the mAb can be observed in (b) because the analyte signal is suppressed by the relatively high amount of the contaminant in the CE effluent.

	The applied voltage for electrophoretic separation varies according to the properties of the analytes. However, the maximum applied voltage should not exceed 300–350 V/cm. Also, when using high-concentration acid BGEs such as 1 M formic acid, the applied voltage should be kept around 250–300 V/cm because such buffers cause sudden application of the voltage and thereby result in the cracking of the capillaries. The developed cracks, in most cases, are not visible to the naked eye but do result in unstable current profiles during analysis.

	At the MS level, optimization achieves two objectives. One is to obtain stable spray and the other is optimal ionization and transportation of the analyte into the MS system. For optimization of spray conditions, the nebulizer pressure and drying gas temperature need to be optimized. For ionization of the analyte, ion introduction voltage (capillary voltage) and fragmentation voltage need to be optimized (for TOF-MS). The parameters for obtaining stable spray conditions depend on the effluent from the upstream separation device. It should be noted here that because of the difference in flow characteristics of liquid chromatography (milliliters per minute) and capillary electrophoresis (nanoliters per minute), the parameters of LC cannot be translated to CE for obtaining equivalent response for a particular analyte. It was observed that a nebulizer pressure greater than 10 psig during CE–MS analysis of mAbs resulted in suction pressure at the mass spectrometer end that caused backflow of the analyte to the CE. This problem was overcome by using nebulizer pressure (5–6 psig) with a drying gas flow rate of 5–10 L/min. Also, switching off the nebulizer gas during injection by creating a time segment in the acquisition software prevented entry of the bubble into the capillary during injection and thereby helped achieve a stable current profile throughout the run. Gas temperatures of 150 °C (for digest) and 200 °C (for intact mAb) yielded satisfactory results. Fragmentor and capillary voltages of 400 V and 3500 V, respectively, were found to be optimal for intact mAb, and 200 V and 3000 V, respectively, were found to be optimal for the digest mAb analysis. It should be noted that while the optimal values of capillary and fragmentor voltages depend on the nature of compound (intact or digest), parameters such as drying gas temperature and flow, nebulizer pressure, and so forth, are flow dependent parameters and therefore are linked to the separation conditions used.

	Figure 4 illustrates the effect of sheath liquid composition on the analysis of mAb digests.

					Figure 4: Total ion chromatogram (TIC) corresponding to mAb digest analysis using the conditions specified in Table I. Sheath liquid: (a) 0.1% acetic acid in 50% methanol, (b) 0.3% acetic acid in 50% methanol, and (c) 0.5% acetic acid in 50% methanol. The sequence coverage of 0%, 13%, and 98%, respectively, could be achieved in the three cases.

	CE–MS has gradually become a gold standard in mAb analysis and is applied routinely from product development to release testing. Nevertheless, the robustness of CE–MS continues to be a concern in the minds of analytical chemists. However, over the past two decades, researchers have published a number of interlaboratory studies that have belied these speculations and have provided ample evidence of the robustness, efficacy, specificity, and reliability of CE–MS. As demonstrated in this column installment, adapting an existing analytical method (for example, LC–MS) to CE–MS is nontrivial and requires optimization of the individual parameters that may impact separation as well as detection. Another major issue is the limitations in the availability of CE–MS interfaces, particularly across vendors. Despite these challenges, the superiority of CE–MS for testing the various attributes of proteins including charge heterogeneity, size distribution, and glycan profiles will ensure that it remains a potent tool in the analytical arsenal.

		S.A. Berkowitz, J.R. Engen, J.R. Mazzeo, and G.B. Jones, 

		G. Narula, V. Hebbi, S.K. Singh, A.S. Rathore, and I.S. Krull, 

		N. Nupur, S.K. Singh, G. Narula, and A.S. Rathore, 

DOI: 10.1016/j.jchromb.2016.05.027 (2016).	

		S.A. Berkowitz, I.S. Krull, A.S. Rathore, 

 (Springer Berlin Heidelberg, 2016), pp. 1–27. DOI: 10.1007/7355_2015_5004	

		A. Beck, H. Diemer, D. Ayoub, F. Debaene, E. Wagner-Rousset, C. Carapito, A.V. Dorsselaer, and S. Sanglier-Cianférani, 

		A. Beck, S. Sanglier-CianfeÌrani, and A.V. Dorsselaer, 

		R.D. Smith, J.A. Olivares, N.T. Nguyen, and H.R. Udseth, 

		M.D. Jones, L.A. Merewether, C.L. Clogston, and H.S. Lu, 

		R. Haselberg, G.J. De Jong, and G.W. Somsen, 

		H.H. Lauer and G.P. Rozing, Agilent Technologies Primer on CE-MS, P/N 5990_3777EN. (

http://www.agilent.com/cs/library/primers/Public/5990_3777EN.pdf)

 is a graduate student in the Department of Chemical Engineering at the Indian Institute of Technology.

 is a professor in the Department of Chemical Engineering at the Indian Institute of Technology in Delhi, India.

 is a Professor Emeritus with the Department of Chemistry and Chemical Biology at Northeastern University in Boston, Massachusetts, and a member of 

Biotechnology Today

In this article, we discuss the use of CE-MS (sheath flow interface) for analysis of intact proteins as well as of protein digests. We discuss the unique aspects that the user needs to be aware of while testing biotherapeutics versus small molecule drugs. We also highlight that the optimization of CE and MS parameters together result in the creation of a more robust and reproducible protein analysis approach. Finally, we list some of the most common errors that are likely to occur during CE-MS analysis and suggest ways to overcome them.

Practical Considerations of Capillary Electrophoresis–Mass Spectrometry for Analysis of Biotherapeutics

Reversed-phase liquid chromatographic columns can be compared quantitatively for differences in selectivity by means of the hydrophobic-subtraction model. This approach allows selection of columns that are either equivalent or different in selectivity. The present article presents a summary of this model and shows in detail how to use it in practice.

	For the practical user of reversed-phase liquid chromatography (LC) columns, two cases are of particular importance when comparing column selectivity. We’ll refer to these as equivalent and orthogonal columns, with examples given in Figure 1 for three columns under identical operating conditions. For two columns to be equivalent, ideally they would give identical separation, but for practical use they can differ somewhat as long as they subjectively give the same separation and enable the same quantitative results to be obtained. The separations for the two columns shown in Figures 1a and 1b are almost identical, so these two columns would qualify as equivalent. The quantitative results of analysis on the two columns would be expected to be the same. We might desire equivalent columns so that we could specify an alternate column for a method or to substitute for a column that is no longer available to us. Often columns do not need to be as similar as those of Figures 1a and 1b to give the same analytical results, especially if the separations are not too challenging.

					Figure 1: Simulated chromatograms for equivalent and orthogonal columns. (a) ACE C18, (b) XTerra MS C18, and (c) Zorbax Bonus RP.

	Identifying an equivalent or orthogonal column is much more difficult than looking at the label on the column box. For example, a C18 and phenyl column from one manufacturer may be more similar than two C18 columns from different manufacturers. At one time, it was thought that all columns with the same bonded phase chemistry were equivalent. This is reflected in the somewhat outdated classification by the United States Pharmacopeia (USP), where different “L” classifications are used (1). For example, all C18 columns are classified as L1. Today it is well known that such classification of column chemistry is oversimplified, and that great differences in the appearance of the chromatogram are possible with different L1 columns. It is likely that there are more than 1000 reversed-phase LC columns on the market today (2), so tools are needed to help determine which of these columns are equivalent or orthogonal.Sometimes, particularly during method development, we desire an orthogonal column. An orthogonal column should give a distinctly different separation from the original column, including significantly different resolution among adjacent peaks in the original separation, as well as reversals in elution order. The separation of Figure 1c is very different from that of Figure 1a or 1b, so this column could be considered orthogonal. The peaks in Figures 1a and 1b are eluted in numeric order, whereas the order for Figure 1c is 2, 1, 5, 3, 7, 4, 6, 9, 10, and 8. Note especially that peak pairs 2–3 and 4–5 were marginally separated on the first two columns, but easily separated on the column of Figure 1c. Finding an orthogonal column does not mean that all the peaks will be completely resolved, but instead that the separation is quite different from the starting column. Additional method development efforts often will be required to get an acceptable separation (for example, to improve separation of peak pairs 2–1 and 5–3 of Figure 1c). Orthogonal columns may be desired to resolve hard-to-separate peak pairs, to help assure us that additional peaks are not “hidden” under existing peaks, or to otherwise make changes to an existing separation. Orthogonal columns are also essential for the two columns used in two-dimensional (2D) separation.

	The purpose of the present article is to describe the hydrophobic-subtraction (H-S) model of reversed-phase selectivity as it applies to identifying equivalent and orthogonal columns. We briefly describe the main interactions important in reversed-phase selectivity and we show how the H-S model has been incorporated into free web-based tools (3,4) that help users choose equivalent or orthogonal columns from the myriad choices available.

	When we started work on column selectivity, our goal was this: Is it possible to test reversed-phase LC columns with a simple mixture of test solutes under standardized conditions and quantitatively predict column equivalency or orthogonality for other samples under other conditions? With the chromatograms of Figure 1, we could use a visual comparisons of columns. A simple way to quantitatively compare column selectivity is shown in Figure 2. Here retention, expressed as log retention factor 

, is compared between two columns. Retention factors of the 10 solutes used in Figure 1 are shown as diamonds and the best fit line is shown in red. In Figure 2a retention on the columns of Figure 1a and 1b is compared; all the points lie on or very close to the trend line, which means that the relative retention is very close to the same for each solute on each column. We would say that these two columns are equivalent. Figure 2b shows a comparison of the retention of the same solutes on the columns of Figure 1a and 1c; the scatter about the trend line is quite apparent. We can see peak reversals take place for every adjacent peak pair except peaks 9–10. There is a much greater difference between these two columns than there is between the columns of Figure 2a. The larger the scatter about the best fit line, the greater the difference in the columns. This scatter is quantitatively expressed as the standard error (SE) of the curve, and can be related to the average difference in selectivity, α, between the two columns.

					Figure 2: Comparing column selectivity by comparing retention (log 

) of solutes on two columns. (a) Column of Figure 1a versus 1b (ACE C18 and XTerra MS C18); (b) column of Figure 1a versus 1c (ACE C18 and Zorbax Bonus RP). Solutes: 1. 4-nitrophenol, 2. acetophenone, 3. benzonitrile, 4. 5-phenylpentanol, 6. 4-n-butylbenzoic acid, 7. toluene, 8. cis-chalcone, 9. ethylbenzene, and 10. 

	To compare columns, a set of solutes needs to be chosen that adequately represents the important interactions between the solute and column. We determined the retention of 87 solutes on 10 high-purity, type-B silica C18 columns under two isocratic conditions (acetonitrile–phosphate buffer at pH 2.8 and 7.0). Based on this data set, a multiple regression technique was used to derive equation 1, which estimates retention, 

, for a given solute within ±1% (1 standard deviation [SD]).

 is the retention factor for a particular solute, 

 is the retention factor for ethylbenzene, the Greek letters (η′, σ′, β′, α′, and κ′) are characteristics of the solutes, and the capital letters (

) are column parameters. The five terms describing selectivity are 

, the hydrophobic interaction between the solute and the column; 

, the resistance of the bonded phase to penetration by bulky molecules; 

, hydrogen bond (H-B) interactions between basic solutes and acidic sites (silanols) in the column; 

, H-B interactions between acidic solutes and a vicinal-silanol pair on the column surface (shown as “x” in Figure 3); and 

, cation exchange between ionized bases and ionized silanol groups in the column. These five interactions are illustrated in the representations of Figure 3. The retention factor for ethylbenzene 

 is used to compensate for differences in packing surface area, which have no effect on column selectivity. The column parameters (

) are discussed further later in this article.

					Figure 3: Conceptual illustration of the five major solute-column interactions of the hydrophobic-subtraction model.

	For the characterization of column parameters (

), 16 suitable test solutes were chosen based on their known solute parameters (η′, σ′, β′, α′, and κ′). Experimental runs for these 16 test solutes allow the calculation of column parameters (

) for any column. After the column parameters are known for two columns, the quantitative difference in selectivity between the columns can be determined using the technique and web-based comparison tools (3,4) described later.

	The current discussion is intended as a recapitulation of other publications on the H-S model, and describes the proposed molecular interactions. For a more detailed discussion of the H-S model, readers are encouraged to read the review (5) plus later reference (6); the cross-references contained in these two articles contain detailed discussions of all aspects of the model.

 correspond to the five most important chemical interactions between solute molecules and the stationary phase (as discussed below) and define column selectivity.

	It is not surprising that the dominant contributor to retention with reversed-phase LC columns is hydrophobic interaction between solutes and the stationary phase, because this is the mechanism of retention usually assigned to reversed-phase columns (7). In fact, 

 is such a dominant factor that its contribution to retention overshadows the minor-but from a selectivity standpoint more important-factors 

. (This dominance of hydrophobic retention accounts for the relatively strong correlation, 

 = 0.73, in Figure 2b even when the selectivity is quite different.) Because of its dominant influence, the value of 

 must be mathematically removed during the calculation of 

hydrophobic-subtraction model of reversed-phase retention

	Hydrophobicity accounts for the general interaction of hydrophobic solute molecules with the hydrophobic stationary phase. The more hydrophobic the solute, the larger its value of η. Hydrophobic interaction is illustrated in the cartoon of Figure 3. 

 defines the hydrophobic nature of the column.

	The density and ligand length of the bonded phase on the silica surface of reversed-phase LC columns will vary from one column to another. As a result, a bulky molecule will have more difficulty penetrating into the bonded phase of a densely bonded phase than a less densely bonded one. Or for the same bonded phase, a more bulky molecule will have a harder time penetrating the bonded phase than a less bulky one. From an oversimplified viewpoint (see reference 8 for more details), this interaction is somewhat analogous to size-exclusion chromatography (SEC) in that both SEC and 

* depend on the hydrodynamic diameter of the solute molecule and relate to the ease of that molecule entering a cavity in the column. However, SEC describes solute access to the column pores, whereas 

* describes the resistance to penetrating the bonded phase.

	Steric selectivity is illustrated in Figure 3 for two molecules of different hydrodynamic radius. The less bulky molecule can more easily penetrate the bonded phase than the more bulky one. With steric selectivity, σ′ is related to the hydrodynamic radius of the solute molecule and 

* is related to the bonded phase density and other bonded phase characteristics that result in poorer penetration by bulky molecules than less bulky ones.

	The interaction of neutral proton acceptors (nonionized bases) with acidic silanol sites on the column is responsible for the column 

 value. As expected, solute molecules that are better proton acceptors (greater hydrogen-bond basicity) are more strongly retained. Conversely, factors that reduce the availability of silanol groups will tend to reduce the value of 

. Thus, endcapping and more dense bonding of the stationary phase will result in smaller A values. Column hydrogen-bond acidity is illustrated in Figure 3 by the interaction of a nonionized basic substituent, B, with an acidic silanol group, Si-OH.

	A silica surface usually is thought to be acidic in nature, yet our studies observed that there is a significant interaction of proton donors, such as nonionized acids or phenols, with the stationary phase. We use the term “basicity” to refer to the hydrogen-bond acceptor strength of such interactions. Whereas the other column variables (

) are fairly easy to describe in terms of the simple interactions shown in Figure 3, 

 is somewhat less obvious. One of the key sources of column basicity is interaction with vicinal silanol groups (X) on the silica surface. Vicinal silanols comprise two adjacent silanol groups that are internally hydrogen bonded. This enables them to doubly hydrogen-bond with carboxylic acids to form a stable ring structure as seen in Figure 4a. Phenolic solutes are less strongly retained by vicinal silanols, which can be explained by the single hydrogen-bond interaction of Figure 4b. The α′

 interaction of Figure 4 is observed for all silica columns. Low purity, type-A silica columns have an additional α′

 interaction, attributed to the presence of metal impurities, particularly Al(III) and Fe(III). Embedded-polar-group (EPG) columns show a third type of α′

, which involves the selective retention of both acids and phenols by proton donor sites (amide, carbamate, or urea) present in such columns. These additional interactions are the reason for generalizing the column interaction site in Figure 3 as “X.” The α′

 interactions are described further in reference 6.

					Figure 4: Illustration of column “basicity.” Interaction of (a) carboxylic acids and (b) phenols with vicinal silanols.

	The final term from equation 1 is the cation-exchange term, 

, cation exchange on silica is a well-known phenomenon in reversed-phase LC. Processes that increase the ionization of silanol groups, such as increased metal contamination of type-A columns and increased mobile phase pH, result in increased values of 

. This is characterized by increased retention of ionized basic solutes. Values of 

 are the result of solute interactions with free column silanols; see reference 9 for further details. A graphic representation of the cation-exchange interaction is shown in Figure 3.

	Equation 1 lists the five primary interactions 

 that allow prediction of retention factors within ±1% (1 SD) for type-B alkylsilica columns. An obvious question arises: Are there other unaccounted for interactions? Indeed, there are (10). Equation 1 might be expanded to take into account π-π interactions between aromatic solutes and alkyl phenyl or cyanopropyl columns (11) as well as dipolar interactions with cyanopropyl columns (12). Consequently, equation 1 is somewhat less reliable for these two column types. Ignoring these additional interactions will not often affect the usefulness of the values of 

 for identifying equivalent or orthogonal columns.

Comparison of Column Selectivity by Means of the Parameter 

 are determined for a column, based on experimental testing, these column parameters can be used to help identify columns that are either equivalent or orthogonal to each other. There are two ways to compare columns in this context: graphical and mathematical. We’ll look at these next, then we’ll examine a web-based tool that can be used for practical column comparison.

	Figure 5 provides a graphical illustration of the concept of column comparison for just three column properties (

). The difference in selectivity can be measured by the distance in three-dimensional (3D) space between points for the two columns (red dashed line).

					Figure 5: Graphical comparison of two columns based on column steric (

	This distance for three variables can be obtained from the Pythagorean theorem:

 is the difference in selectivity, comparable to the distance between the two points in Figure 5. This technique of comparison can be mathematically (but not visually) extended to all five column parameters:

	Now we can compare any two columns for which we have measured the column parameters 

 is large, the columns are different. For the best results, a set of loading coefficients (not shown, see reference 13) are added to equation 3 to take into account the different influence of 

 contributes only a little to selectivity, although it is the main factor in determining retention, 

 ≤ 3: columns are equivalent for most separations

 < ~10: columns are likely to be equivalent, especially for samples that contain only a few solutes with resolution, 

 ≥ 100: columns are orthogonal when ionized solutes are present in the sample

 ≥ 35: columns are orthogonal when the sample comprises predominantly nonionized analytes and when values of 

	These are generalizations to be used as guidelines. In a collaborative study between several laboratories, the guidelines above were found to be reasonable for identification of equivalent (14) and orthogonal (15) columns when applied to a variety of pharmaceutical samples under different chromatographic conditions.

 have been measured for ~700 commercially available reversed-phase columns. These have been tabulated into two databases that are available on the internet (3,4). Here we use the USP database (3) for illustrative purposes. First, we look at how to use the database to identify equivalent columns and then how to find orthogonal ones.

	The USP database for column comparison looks similar to Figure 6 (the figure has been simplified somewhat for clarity in the current presentation). Note that on the USP website, there are two different databases available. The “USP Database” is limited to a smaller number of C18 columns and has not been validated to the extent of the following “PQRI Database”-we recommend the latter database for use in reversed-phase column comparison. The database features are discussed briefly below and an example is presented.

					Figure 6: PQRI database comparing column selectivity (3). See text for description.

	The drop-down box at the top of Figure 6 gives access to all the columns in the database. In the present example, we have selected an ACE 5 C18 column from Advanced Chromatography Technologies Ltd. (ACT) from the ~700 columns listed for comparison with other columns.

	The statement, “You are viewing similar columns,” indicates you are in the equivalent-column mode. If you would rather be in the orthogonal mode, click the “View Different” button to select that option (see red arrows).

	At the top of the data table, you have the option to indicate if your sample contains acids or bases, and the pH of the mobile phase (see blue arrows). These entries adjust the loading factors noted for equation 3 so that the 

 values calculated best reflect results that will be relevant to your sample. If you do not know the compounds present in the sample, check both acids and bases. When you have made the appropriate entries, click the “Update” button (upper right in Figure 6, see black arrow) and the results will be updated to reflect comparisons with your column and conditions.

	Once the Update button is pressed, the table below will display your selected column as the first entry and the 10 most similar columns in order below it. The data columns are described below.

		Rank: This is the numeric order of column similarity, beginning at 0 (the selected column) and continuing with 1 for the most similar column, and so forth.

 will increase down the table, corresponding to less similar columns.

		H, S, A, B, C (2.8), C (7.0): These are the 

 values measured for the column. Note that values of 

 at pH 2.8 and pH 7.0 are shown. The calculation algorithm interpolates between these two values depending on the pH value selected when the conditions are entered (pH 2.8 in the current example).

		Type: This indicates the classification of the column type, mostly based on the column manufacturer’s information. “A”: older, lower purity, type-A silica; “B”: modern, high purity, type-B silica; “EP”: embedded polar or AQ phase; “phenyl”: phenyl phase; “CN”: cyano phase; “F”: fluoro phase; “other”: phases that don’t fit the preceding classifications, such as zirconia. As a general rule, when initially selecting a column for a separation, it is strongly recommended to choose a type-B column over a type-A column for improved reproducibility and reduced peak tailing.

		USP Description: This is the USP’s classification of the column. L1: C18; L7: C8; see reference 1 for a complete listing of the USP “L” classifications.

		Manufacturer: This is a hyperlink to the column manufacturer’s website, where more information about the column can be found.

		Previous/Next (circled at bottom left): Click on the desired word, and the columns displayed below the reference column will be updated with the previous or next 10 columns in the database. For example, if you are viewing columns 1–10, clicking on “Next” will update the display to show column 0 (your selected reference column) and columns 11–20.

Peer-Reviewed Article

Reversed-phase liquid chromatographic columns can be compared quantitatively for differences in selectivity by means of the hydrophobic-subtraction model. This allows selection of columns that are either equivalent or different in selectivity. The present paper both presents a summary of this approach and shows in detail how to use it in practice.

The Hydrophobic-Subtraction Model for Reversed-Phase Liquid Chromatography: A Reprise

The 44th International Symposium of High Performance Liquid Phase Separations and Related Techniques (HPLC 2016), chaired by Professor Robert Kennedy, was held June 19–24 in San Francisco, California, at the Marriott San Francisco Marquis. This installment of “Column Watch” covers some of the highlights observed at the symposium, including stationary-phase developments, particle technology, and areas of growing application of high performance liquid chromatography (HPLC). In addition, trends and perspectives on future developments in HPLC noted from the conference are presented. 

HPLC 2016, chaired by Professor Robert Kennedy, was held June 19–24 in San Francisco, California, at the Marriott San Francisco Marquis. This installment of "Column Watch" covers some of the highlights observed at the symposium including stationary-phase developments, particle technology, and areas of growing application of HPLC. In addition, trends and perspectives on future developments in HPLC noted from the conference are presented.

LCGC North America-09-01-2016