Comparison of two column characterisation systems based on pharmaceutical applications

Available online at www.sciencedirect.com

Journal of Chromatography A, 1189 (2008) 59–71

Comparison of two column characterisation systems based on pharmaceutical applications Erik Haghedooren a , Tam´as N´emeth a,b , Sanja Dragovic a , B´ela Nosz´al c , Jos Hoogmartens a , Erwin Adams a,∗ a

Katholieke Universiteit Leuven, Laboratorium voor Farmaceutische Analyse, O&N 2, PB 923, Herestraat 49, B-3000 Leuven, Belgium b National Institute of Pharmacy, Zr´ınyi u. 3, H-1051 Budapest, Hungary c Semmelweis University, Department of Pharmaceutical Chemistry, H˝ogyes E. u. 9, H-1092 Budapest, Hungary Available online 8 February 2008

Abstract A useful column characterisation system should help chromatographers to select the most appropriate column to use, e.g. when a particular chromatographic column is not available or when facing the dilemma of selecting a suitable column for analysis according to an official monograph. Official monographs of the European Pharmacopoeia and the United States Pharmacopeia are not allowed to mention the brand name of the stationary phase used for the method development. Also given the overwhelming offer of several hundreds of commercially available reversedphase liquid chromatographic columns, the choice of a suitable column could be difficult sometimes. To support rational column selection, a column characterisation study was started in our laboratory in 2000. In the same period, Euerby et al. also developed a column characterisation system, which is now released as Column Selector by ACD/Labs. The aim of this project was to compare the two existing column characterisation systems, i.e. the KUL system and the Euerby system. Other research groups active in this field will not be discussed here. Euerby et al. developed a column characterisation system based on 6 test parameters, while the KUL system is based on 4 chromatographic parameters. Comparison was done using a set of 63 columns. For 7 different pharmaceutical separations (fluoxetine, gemcitabine, erythromycin, tetracycline, tetracaine, amlodipine and bisacodyl), a ranking was built based on an F-value (KUL method) or Column Difference Factor value (Euerby method) versus a (virtual) reference column. Both methods showed a similar ranking. The KUL and Euerby methods do not perfectly match, but they yield very similar results, allowing with a relatively high certainty, the selection of similar or dissimilar columns as compared to a reference column. An analyst that uses either of the two methods, will end up with a similar ranking. From a practical point of view, it must be noted that the KUL method only includes 4 parameters and 3 chromatographic methods compared to 6 parameters and 4 methods for the Euerby method. Hence, the time needed to determine the chromatographic properties of a column is shorter for the KUL approach. Access to the KUL method also requires no download procedures. © 2008 Elsevier B.V. All rights reserved. Keywords: Reversed-phase liquid chromatography; Test parameters; Column characterisation system; Pharmaceutical separations

1. Introduction There are several hundreds of different reversed-phase liquid chromatography (RPLC) stationary phases commercially available, the manufacturers of which vary and so the chromatographic performance may differ considerably. Official compendia like the European Pharmacopoeia (Ph. Eur.) [1] or United States Pharmacopeia (USP) [2] mostly prescribe C18

∗

Corresponding author. Tel.: +32 16 323444; fax: +32 16 323448. E-mail address: [email protected] (E. Adams).

0021-9673/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.chroma.2008.02.012

RPLC columns for analysis. In chromatographic conditions prescribed in a monograph, the mobile phase is exactly described whereas the stationary phase is only vaguely identified in terms like chain length, end-capping, base-deactivation, particle size and sometimes pore size and specific surface. The European Pharmacopoeia does not report the names of the suitable columns in the monographs. Instead, various descriptions have been employed in the attempt to differentiate C18 stationary phases with different properties (Table 1). Unfortunately, these descriptions are not very helpful nor have they been applied consistently. It has become evident over the years with the innovations introduced by the column

60

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

Table 1 Reversed-phase (C18 ) stationary phases listed in the European Pharmacopoeia (octadecylsilyl silica gel for chromatography) Description

Property

R R1 R2 Base-deactivated Endcapped End-capped & base deactivated

3–10 ␮m Ultra pure <20 ppm metals Ultra pure, 15 nm pore size, 20% carbon load 3–10 ␮m, pre-treated 3–10 ␮m, to minimise any interaction with basic compounds it is carefully endcapped to cover most of the remaining silanol groups 3–10 ␮m, 10 nm pore size, 16% carbon load, pre-treated. To further minimise any interaction with basic compounds it is carefully end-capped to cover most of the remaining silanol groups

manufacturers and with the increasing need for higher selectivity to separate impurities that there is need for a better description of the properties of the stationary phases. This has been recognised by the Ph. Eur. and to help the analyst, the commercial names of suitable columns from the proposed monograph in Pharmeuropa were included in the “knowledge” database. A similar situation exists in the USP, where the description of L-types for some packing is so broad that many columns meet the specifications. However, all columns that meet the specifications are not necessarily alike and only a subset of those complying may actually achieve the desired separation. L1, defined as the group including “octadecyl silane chemically bonded to porous silica or ceramic micro-particles, 3–10 ␮m in diameter”, includes all C18 columns. However, all C18 columns are not alike due to the varying column manufacturing procedures, including, e.g. various types of ligands and end-capping procedures. So, the statement “a C18 is not a C18 ” holds true [3]. The USP provides information on the development column in its Chromatographic reagents book and in the on-line version of USP. For older monographs, this information is not available. Even when the column on which the method was developed is known, the separations are not necessarily reproducible. In order to achieve the required separation, some adjustments are permitted by the Ph. Eur. for the prescribed parameters of the stationary phase: column length: ±70%, column internal diameter: ±25% and the particle size may be reduced upto maximum 50% [1]. Nonetheless, adjustment of the various parameters will not always result in satisfactory chromatography. In that case, it may be necessary to replace the column with another of the same type (e.g. octadecylsilyl silica) which may exhibit the desired chromatographic behavior. It is also possible that the required column is not easily available on the market or simply is not present in the laboratory. The average control laboratory does not have available all the stationary phases used in official monographs. It should also be considered that column properties may change over longer periods of production or that previous use of a column changes its properties. Control laboratories cannot use a new column for each analysis. Thus, it is important to have the means to identify columns with equivalent selectivity, particularly for the control of impurities. The advantage of having columns with various properties is that they can be used to solve diverse separation problems. A disadvantage is that analysts do not always recognise that nominally identical materials show very different chromatographic properties. On the other hand, during method development it may be needed to identify columns with quite different proper-

ties in order to select a column that best solves the separation problem. In this situation also, a column ranking system, based on chromatographic performance, is helpful. An appropriate system should be easily applicable and a helpful tool for analysts when confronted with the choice of a similar or dissimilar column when performing pharmaceutical separations.The availability of a good characterisation and classification system is important for several reasons. Often, one has to find a column similar to one that is described in an existing method or in a paper because the prescribed column is not available in the laboratory. Sometimes, the column that was used for method development is simply not available anymore on the market. It is also possible that column properties differ between batches or that previous use of a column changed its properties. Many laboratories, e.g. control laboratories, do not use a new column for each separation they have to perform. In each of these cases, it is desirable to be able to identify an alternative column of similar selectivity, so that the replacement column will provide an “equivalent” separation as the original column. This column characterisation system allows the identification of an alternative column, with similar selectivity as the prescribed reference column, so that the alternative column will provide a separation “equivalent” to that of the prescribed column. It also allows the follow up of column ageing, so that analysts can easily check whether the characteristics of their columns have changed over time. When needed, the column classification system can also help to indicate columns with different (orthogonal) selectivity, which is useful during method development. Here, a comparison of two existing characterisation and ranking methods was done. A set of 63 RPLC C18 columns was characterised according to both methods. First, the columns were characterized with the Katholieke Universiteit Leuven (KUL) method, using four parameters: the retention factor of amylbenzene (k amb ), the relative retention factor benzylamine/phenol at pH 2.7 (rk ba/ph pH 2.7 ), the retention factor of 2,2 -dipyridyl (k 2,2 -dip ) and the relative retention factor triphenylene/oterphenyl (rk tri/o-ter ). Then, the columns were characterised according to the Euerby method, using 6 parameters: the relative retention time of pentylbenzene (kPB ), the selectivity factor between amylbenzene and butylbenzene (αCH2 ), the selectivity factor between triphenylene and o-terphenyl (αT/O ), the selectivity factor between caffeine and phenol (αC/P ), the selectivity factor between benzylamine and phenol at pH 2.7 and 7.6 (αB/P pH 2.7 and pH 7.6 ). Next, principal component analysis (PCA) was performed onto both datasets. Evaluation of the scores and loading plots gave more information about the columns

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

and the parameters. Both systems were then critically evaluated using 7 different pharmaceutical separations (fluoxetine, gemcitabine, erythromycin, tetracycline, tetracaine, amlodipine and bisacodyl). A ranking was built, based on an F-value (KUL method) or Column Difference Factor (CDF) value (Euerby method) versus a (virtual) reference column. Ranges were defined to find a suitable column, with a high, moderate or low probability. Based on the number of suitable columns in the respective ranges, the KUL and Euerby method showed similar patterns. 2. Column classification systems 2.1. KUL method In recent years, a simple chromatographic test procedure to characterise and rank RPLC C18 columns has been developed by Hoogmartens, and co-workers. First, a procedure, allowing the measurement of a number of parameters, reflecting chromatographic characteristics was developed. In order to make the procedure as simple as possible, the number of test parameters was kept minimal. The second part intended to group RPLC columns with closely related characteristics. The third part consisted in performing pharmaceutical separations to check the usefulness of the developed ranking system of the columns in practice. First of all, 36 test parameters were carefully selected from literature and determined on 69 RPLC columns [4]. When developing a method like this, it is very important to check the repeatability and reproducibility. Out of the 36 test parameters, 24 proved to be repeatable and reproducible [5]. PCA was used as chemometric tool to learn more about the columns and the parameters. This approach resulted in a reduction of 24 parameters to four final column parameters [6–9]. These four parameters are: the retention factor of amylbenzene (k amb ), the relative retention factor benzylamine/phenol at pH 2.7 (rk ba/ph pH 2.7 ), the retention factor of 2,2 -dipyridyl (k 2,2 -dip ) and the relative retention factor triphenylene/o-terphenyl (rk tri/o-ter ). Next, a ranking system based on F-values was introduced, starting with the selection of 4 reference parameters corresponding to a freely chosen reference column. The F-value for a column i is calculated as:     F = (kamb,ref − kamb,i ) + (rkba/ph pH 2.7,ref − rkba/ph pH 2.7,i ) 2

2

    +(rktri/o-ter,ref −rktri/o-ter,i ) +(k2,2  -dip,ref −k2,2 -dip,i ) 2

2

61

where xij is the value of parameter j on column i, x¯ j is the mean of parameter j on all tested columns and sj is the standard deviation for parameter j [10]. The next step was to check the correlation of the characterisation system and real separations. Previous studies included the separation of acetylsalicylic acid (ASA) [11], and seven other pharmaceutical substances from their respective impurities (clindamycin hydrochloride, buflomedil hydrochloride, chloramphenicol sodium succinate, nimesulide, phenoxymethylpenicillin, dihydrostreptomycin sulphate and vancomycin) [12–14]. After testing 69 columns to check the applicability of the KUL method, a nice relationship between the ranking of the columns and the selectivity in the separations of the pharmaceuticals was demonstrated and it was concluded that the column characterisation system is an interesting tool for analysts in the selection of a suitable RPLC C18 column. The parameters of the columns used are freely accessible on a website [15], where anyone can freely define a reference column or reference parameters and the column ranking is given based on the F-values. In order to evaluate the separation on the stationary phases, the chromatographic response function (CRF), which is a measure for the overall selectivity, was applied [15]. The CRF was calculated as: CRF =

n−1  i=1

fi gi

(3)

where n is the total number of solutes, g the interpolated peak height, i.e. the distance between the baseline and the line connecting the two peak tops, at the location of the valley and f the depth of the valley, measured from the line connecting the two peak tops, as shown in Fig. 1a and b shows a specific case where a large peak was separated from a much smaller peak. This was resolved by using a horizontal line to find the f and g values, instead of a line connecting the peak tops. It follows that a baseline separated peak pair has an f/g ratio of 1.00, a non-separated pair has a value of 0.00 and in case of partial co-elution the separation has an intermediate value. Columns with CRF = 1.00 show baseline separation for all peaks, but this does not necessarily mean that the separation is identical or column properties are exactly the same. It only indicates that these columns are suitable for that separation. In the previous studies, a virtual reference column was calculated starting with the selection of all the columns yielding a CRF equal to 1.00 for the concerned separation. Each of these columns was characterised

(1)

The F-value of a column i equals the sum of squares of the differences between each parameter value of the reference column and of a column i. The smaller the F-value, the more similar is column i to the reference column and the higher is column i found in the ranking (high ranked columns). Before being introduced in Eq. (1), the parameters are autoscaled: xij − x¯ j sj

(2)

Fig. 1. (a) Determination of CRF value for peaks of almost equal size; (b) Determination of CRF value for a small and a large peak.

62

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

by 4 parameters. In order to define a virtual column, which can be considered as ideal for the given separation, the mean value for each of the 4 parameters was calculated after removing outlying values using a Grubbs’ test (α = 0.05) [16]. First, all columns giving a CRF = 1 in a given separation were grouped and a Grubb’s test was performed onto the data to trace and remove outliers for each parameter. If a column had one or more outliers, all of its four parameters were omitted for the calculation of the virtual column parameters. Ideal column parameters were calculated by taking the average of the parameters of the remaining columns with a CRF = 1. Finally, F-values for all columns were calculated versus the reference parameters obtained for this virtual, ideal column. Although the F-value does not allow exact characterisation of columns, it is an interesting tool to facilitate the selection of a suitable RPLC C18 column. 2.2. Euerby method In 1989, Tanaka and co-workers started a column characterisation study on silica C18 packing materials [17]. Their goal was to develop a method without destroying the packing material and without using any expensive instrumentation. So, instead of using stationary phase destructive methods or spectroscopic methods, test compounds were selected to show the differences in chromatographic properties among the packing materials: hydrophobic properties, steric selectivities, the extent of hydrogen bonding and of ion exchange properties. The final test procedure included the relative retention time of pentylbenzene (amount of alkyl chains, kPB ), the selectivity factor between amylbenzene and butylbenzene (hydrophobicity, αCH2 ), the selectivity factor between triphenylene and o-terphenyl (steric selectivity, αT/O ), the selectivity factor between caffeine and phenol (hydrogen bonding capacity, αC/P ), the selectivity factor between benzylamine and phenol at pH 2.7 and 7.6 (ion exchange capacity at pH <3 and >7, αB/P pH 2.7 and pH 7.6 ). The attributed nomenclature for the test parameters was kept as described in publications. The mobile phase for the first three parameters consisted of water:methanol (20:80, v/v), for the fourth parameter water:methanol (30:70, v/v) and for the last two methanol:0.02 mol/l phosphate buffer pH 2.7 or 7.6 (30:70, v/v). Based on these results, a hexagon was drawn by plotting the determined parameter values. When the first three parameters showed high values, this was an indication of a higher retention of the hydrocarbons, hydrophobicity and steric selectivity, respectively. For the last three parameters, the smaller the value, the lower the number of silanols or ion exchange sites. Extending these parameters with a parameter to study the effect of surface metal contamination on column selectivity and the efficiency, Euerby and co-workers performed their column characterisation system on 30 commercially available RPLC columns [18]. The additional parameter, the 2,3- and 2,7-dihydroxynaphtalene efficiency ratio test (DERT), was afterwards omitted as the effect of very low concentrations of surface complexed metals on the characterization parameters was small and the value was reported to be highly dependent on the column history. Based on the above mentioned 6 parameters and additionally the number of theoretical plates per meter of the

compound pentylbenzene (NPB ), PCA, cluster analysis and radar plots were performed on the data set, in order to reduce the number of parameters and to indicate the correlated parameters. In 2000, Euerby et al. extended the number of tested columns to 85 columns [19]. Based on the earlier mentioned 7 parameters, PCA was applied again and 7 groups of columns were observed: non-endcapped C18 columns with poor surface coverage based on acidic silica (type A), polar embedded C8 and C18 columns, other C8 and C18 columns with different degrees of endcapping and different silicas, some specific C8 columns, C18 columns based on acidic, type A, silica and/or with poor surface coverage, modern C18 columns based on less acidic, type B silica and relatively hydrophobic C18 columns. Based on the PCA plots on 135 stationary phases, NPB was found to be highly correlated with kPB and therefore removed. The study of separations of a very hydrophobic steroid and amino alcohol pointed out that columns situated close to each other on the PCA plot gave similar separations. The number of columns was further extended to 135, including cyano, phenyl, perfluorinated, enhanced polar selectivity and aqua columns [20]. In 2005, the comparison of columns with polar embedded or amino endcappings using PCA was published [21]. Also using PCA, the comparison of commercially available RPLC columns containing phenyl moieties was investigated [22]. The final procedure contained the six parameters of Tanaka et al. and PCA was used to provide a graphical comparison of the phases within the database. PCA proved to be useful in discriminating between C18 and C8 , and between the polar embedded groups and polar/hydrophilic endcappings. To identify stationary phases with equivalent chromatographic properties, all 6 variables were autoscaled and the differences between the autoscaled variables of the column of interest and all other columns were calculated. Next, the 6-dimensional space was reduced to one single value, the CDF value based on the Euclidean distance, showing the difference between a certain reference column and all other ones: √ CDF = [(xnt1 − xn1 )2 + (xnt2 − xn2 )2 + (xnt3 − xn3 )2 +(xnt4 − xn4 )2 + (xnt5 − xn5 )2 + (xnt6 − xn6 )2 ] (4) where xn1 –xn6 are normalized values, calculated for each of the six parameters (xnx = (xx –μx )/SD) and xnt1 –xnt6 are the normalized values of the six chromatographic parameters for the target column. By sorting or ranking these distances, it is possible to identify the most similar or dissimilar columns in the database. The robustness of this final procedure was assessed using a fractional factorial design and was deemed to be acceptable [22]. This system was released as ACD/Column Selector by Advanced Chemistry Development, Inc. (ACD/Labs). 3. Experimental 3.1. Column examination In this study, a set of 63 RPLC C18 columns was investigated. Specifications of the different columns examined are given in

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

63

Table 2 List of RP–LC columns examined and their properties as provided by the manufacturer Column number

Name of the column

Length (mm)

Internal diameter (mm)

Particle size (␮m)

˚ Pore size (A)

Manufacturer/Supplier

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63

Acclaim 3 ␮m Acclaim 5 ␮m ACE 5 C18 Alltima AQ Alltima C18 Alltima HP C18 Alltima HP C18 Amide Brava BDS C18 Capcell Pak C18 ACR Capcell Pak C18 AQ Capcell Pak C18 MG Capcell Pak C18 UG120 Chromolith Performance Discovery C18 Discovery HS C18 Exsil ODS 5 ␮m Hamilton HxSil C18 Hydrospher C18 HyPURITY Advance HyPURITY Aquastar HyPURITY C18 Inertsil ODS-2 Inertsil ODS-3 Inertsil ODS-80A Inertsil ODS-P Kromasil KR100-5C18 LiChrosorb RP-18 LiChrospher 100RP-18 MP-Gel ODS-5 Omnispher 5 C18 Platinum C18 Platinum EPSC18 Polaris 5 C18-A Prevail Amide Prevail C18 Prevail Select C18 Prontosil 120 5 C18 AQ Prontosil 120 5 C18 AQPLUS Prontosil 120 5 C18 ace EPS Prontosil 120 5 C18 H Prontosil 1205C18SH Prontosil 60 5 C18H Purospher RP-18e Purospher Star RP-18 Pursuit 5 C18 Restek Allure C18 Restek Pinnacle DB C18 Restek Pinnacle II C18 Restek Ultra C18 Supelcosil LC-18 Supelcosil LC-18 DB Superspher 100 RP-18 Uptisphere 5 ODB-25QS Wakosil II 5 C18 RS Xterra MS C18 Xterra RP C18 YMC-Pack Pro 3 C18 YMC-Pack Pro 5 C18 YMC-Pack Pro C18 RS Zorbax Eclipse XDB-C18 Zorbax Extend-C18 Zorbax SB-Aq Zorbax SB-C18

150 250 250 250 250 250 250 250 250 250 250 250 100 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250 250

4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.0 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.0 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6 4.6

3 5 5 5 5 5 5 5 5 5 5 5

300 120 100 100 117 100 100 145 80 80 90 120 20000/130* 180 120 80 312 120 190 190 190 150 100 80 100 100 100 100 120 110 100 100 180 190 110 120 120 120 120 120 120 60 90 120 180 60 140 110 100 100 100 100 120 120 125 125 120 120 80 80 80 80 80

Dionex Dionex Achrom Alltech Alltech Alltech Alltech Alltech Shiseido Fine Chemicals Shiseido Fine Chemicals Shiseido Fine Chemicals Shiseido Fine Chemicals Merck Supelco Supelco SGE Hamilton YMC Thermo Electron Corp. Thermo Electron Corp. Thermo Electron Corp. GL Sciences Inc. GL Sciences Inc. GL Sciences Inc. GL Sciences Inc. EKA Chemicals Merck Merck YMC/OmniChrom Varian Alltech Alltech Varian Alltech Alltech Alltech Bischoff Bischoff Bischoff Bischoff Bischoff Bischoff Merck Merck Varian Restek Restek Restek Restek Supelco Supelco Merck Interchrom/Achrom SGE Waters Waters YMC YMC YMC Agilent Agilent Agilent Agilent

5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 3 5 5 5 5 5 5

64

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

Table 3 Description of the methods used to characterise columns and the samples analysed in the KUL column characterisation methods Method

Mobile phase 2.7a

Sample

Column parameter

Benzylamine Phenol

rk ba /ph pH 2.7

A

Methanol–water—0.2 mol/l KH2 PO4 at pH (34:90:1 0 w/w)

B

Methanol–water—0.2 mol/l KH2 PO4 at pH 6.5a (34:90:1 0 w/w)

2,2 Dipyridyl

k 2,2 -dip

C

Methanol–water (317:100 w/w)

Uracil Amylbenzene o-terphenyl triphenylene

k amb rk tri/o-ter

Method

Sample composition

A B C

5 mg of benzylamine and 5 mg of phenol in 10 ml of mobile phase A 3 mg 2,2 -dipyridyl in 10 ml of mobile phase B 0.1 mg uracil, 7 mg amylbenzene, 0.2 mg o-terphenyl and 0.02 mg triphenylene in 10 ml mobile phase C a

The pH adjustments were performed before adding the organic compound of the mixture.

Table 2. All columns were gifts from either the manufacturer or the supplier. After the characterization of all columns by the KUL test procedure and subsequently by the Euerby test procedure, 7 pharmaceutical separations were performed.

3.2. KUL method 3.2.1. Reagents and samples Uracil was purchased from Janssen Chimica (Geel, Belgium) and o-terphenyl from Aldrich (Bornem, Belgium). Amylbenzene, benzylamine, 2,2 -dipyridyl, phenol and triphenylene were acquired from Acros Organics (Geel, Belgium). All solvents and reagents were of European Pharmacopoeia (Ph. Eur.) quality. Methanol (Prolabo, Paris, France) was of LC grade, other chemicals were of AR grade. Potassium dihydrogen phosphate was obtained from Fluka (Buchs, Switzerland) and phosphoric acid from Sigma–Aldrich (Seelze, Germany). Water was purified by a Milli-Q water-purification system (Millipore, Milford, MA, USA).

3.2.2. Instrumentation and liquid chromatographic conditions The LC apparatus consisted of a Varian (Walnut Creek, CA) 9010 LC pump, a 9100 autosampler equipped with a 20 ␮l loop and a 9050 UV–vis detector, set at 254 nm. ChromPerfect 4.4.0 software (Justice Laboratory Software, Fife, UK) was used for data acquisition. The flow rate was 1 ml/min and the column temperature was maintained by immersion in a water bath heated by a Julabo EC thermostat (Julabo, Seelbach, Germany) at 40 ◦ C. The pH of the buffers was adjusted using a Consort C831 pH meter (Consort, Turnhout, Belgium), equipped with a Hamilton (Bonaduz, Switzerland) combination glass electrode.

3.2.3. Column characterisation To characterise the columns, three chromatographic methods were carried out in a defined order (A-B-C) (see Table 3) to determine the 4 column parameters. Each column was first conditioned with the mobile phase of method A after which rk ba/ph pH 2.7 was determined in triplicate. Then, method B was performed on that column and finally method C. Next, the col-

Table 4 Description of the methods used to characterise columns and the samples analysed in the Euerby column characterisation methods Method

Mobile phase

Sample

Column parameter

A

Methanol–water (8:2, v/v)

Pentylbenzene Butylbenzene triphenylene o-terphenyl

kPB αCH2 αT/O

B C D

Methanol–water (3:7, v/v) Methanol–0.02 mol/l KH2 PO4 at pH 7.6 (3:7, v/v) Methanol–0.02 mol/l KH2 PO4 at pH 2.7 (3:7, v/v)

Caffeine phenol Benzylamine.HCl phenol Benzylamine.HCl phenol

αC/P αB/P pH 7.6 αB/P pH 2.7

Method

Sample composition

A B C D

0.6 ␮g/ml pentylbenzene, 0.3 ␮g/ml butylbenzene and 0.5 ␮g/ml triphenylene and o-terphenyl 1 mg/ml caffeine and 0.5 mg/ml phenol 0.5 mg/ml benzylamine.HCl and 0.5 mg/ml phenol 0.5 mg/ml benzylamine.HCl and 0.5 mg/ml phenol

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

65

umn was conditioned with the storage solution prescribed by the manufacturer. See Table 3. As compared to the originally proposed chromatographic methods, some of the conditions were slightly adapted, as was explained elsewhere [23]. The relative retention factor benzylamine/phenol at pH 2.7 (rk ba/ph pH 2.7 ) from method A, the retention factor of 2,2 -dipyridyl (k 2,2 -dip ) from method B, the retention factor of amylbenzene (k amb ), and the relative retention factor triphenylene/o-terphenyl (rk tri/o-ter ) from method C were calculated using the dead volume obtained in method C with uracil. All measurements were executed in triplicate, resulting in RSD values lower than 1%. 3.3. Euerby method 3.3.1. Reagents and samples In addition to the reagents and samples from Section 3.1.1., butylbenzene and coffein were obtained from Acros Organics. 3.3.2. Instrumentation and liquid chromatographic conditions The equipment used was the same as in Section 3.2.2.

Fig. 2. Loadings plot of the KUL test parameters.

4. Results and discussion 3.3.3. Column characterisation The characterisation of all columns was performed as reported in his previous publication [20]. To characterise the columns, four chromatographic methods were carried out in a defined order (A-B-C-D) to determine the 6 column parameters. Each column was first conditioned with the mobile phase of method A after which rk ba/ph pH 2.7 was determined in triplicate. Then, methods B and C were performed on that column and finally method D. Next, the column was conditioned with the storage solution prescribed by the manufacturer. See Table 4. As dead time marker, the first disturbance of the baseline on the injection of methanol was used. All measurements were executed in triplicate, resulting in RSD values lower than 1%. 3.4. Chromatographic conditions of separations Chromatographic conditions and chromatograms were already published elsewhere for fluoxetine and gemcitabine [24], erythromycin and tetracycline [25], tetracaine [26] and amlodipine and bisacodyl [1]. 3.5. Data treatment For each parameter, the average value and standard deviation from the three chromatograms was calculated. Next, each single value was autoscaled as shown in Eq. (1) for both the KUL and Euerby data. F-values and CDF values were calculated as shown earlier. PCA was performed on the autoscaled data of both datasets, resulting in scores plots and loadings plots. These calculations were executed with Statistica 7 (StatSoft, Tulsa, OK, USA).

4.1. Comparison of column parameters When comparing the two methods, three common parameters were found: kPB (amount of alkyl chains) and k amb (hydrophobicity); αB/P pH 2.7 (ion exchange capacity at pH <3) and rk ba/ph pH 2.7 (silanol activity); αT/O (steric selectivity) and rk tri/o-ter (steric selectivity). When comparing the respective chromatographic circumstances, it was observed that only the composition of the mobile phase differed slightly in the ratio water–organic modifier (Tables 2 and 3). To check the correlation among these parameters, the values for both methods per column were plotted versus each other and the coefficient of determination (R2 ) was calculated. For the test parameters describing hydrophobicity, the R2 value was 0.99 with an equation of y = 1.006x − 0.027. For the silanol activity parameter at pH 2.7, the R2 was 0.93 with an equation of y = 0.934x + 0.001. For the steric selectivity parameter, the R2 was 0.99 with an equation of y = 1.032x − 0.046. So, in general the common test parameters matched very well. 4.2. Comparison based on principal component analysis The use of PCA for the elucidation of the similarities and dissimilarities among the HPLC stationary phases was already described [27–29]. Here, PCA was used to compare the obtained loadings and scores plots of both approaches. The loadings plots for both systems can be seen in Figs. 2 and 3. By plotting the loadings for PC1 and PC2, it is possible to see which of the variables are the most important (longest distance from the origin) and if any variables are correlated (the same directions from the origin). First of all, these plots were compared to earlier published articles. When comparing the KUL loadings

66

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

Fig. 4. Scores plot according to the KUL test procedure.

Fig. 3. Loadings plot of the Euerby test parameters.

plot (Fig. 2) with that from Iv´anyi et al. [6], it was seen that the four parameters were positioned in the same way towards each other; rk tri/o-ter (steric selectivity) and k amb (hydrophobicity) in opposite directions, and between these two, rk ba/ph pH 2.7 (silanol activity) and k 2,2 -dip (silanol activity and metal impurity), showing some correlation. This can easily be explained on the basis of similar property they represent, i.e. silanol activity. The metal impurity factor obviously prevents entire correlation. Also, when comparing Fig. 3 with the loadings plot from Euerby et al. [19], similar patterns were observed. Here, kPB (surface area and surface coverage) and αT/O (shape selectivity) are in opposite direction towards each other, and between them are situated αB/P pH 2.7 and 7.6 (ion exchange capacity at pH <3 and >7) and αC/P (hydrogen bonding capacity). Comparison between Figs. 2 and 3 shows that the common parameters (hydrophobicity, silanol activity and steric selectivity) are positioned towards each other in the same way (corresponding points are 90◦ clockwise rotated). Also, about 70% of the original information was used for both systems as can be calculated from the contribution of PC1 and PC2. Moreover, Fig. 3 indicates that kPB (amount of alkyl chains) and αCH2 (hydrophobicity) are correlated, which makes sense since the amount of alkyl chains is related to hydrophobicity. From here, it can be concluded that one of these parameters is superfluous and that omitting one of these will give similar results. By plotting the scores on PC1 and PC2, it is possible to graphically interpret similarities and differences between objects (here columns). Comparison of the scores plots (Figs. 4 and 5) reveals that the same groups of points which correspond to columns are formed. For example, column nos. 42, 46 and 59 are grouped for both systems, in the direction of the hydrophobicity parameters. This can be explained since all three columns (Prontosil 60-5-C18 -H, Restek Allure and YMC-Pro Pack RS) possess high carbon contents (21%, 27% and 22%, respectively). Other

groups could be distinguished with similar properties, as defined by the test parameters. Columns no. 25 and 27 (both considered as polymeric bonded C18 columns with no base-deactivation) are closely related for both systems in the direction of the silanol parameters. A group of columns no. 7, 34, 36, 56 is dissociated (all are polar embedded columns) and column no. 19 (also polar embedded) is positioned close to this group. Two columns with polar endcapping (no. 4 and 35) are next to each other in both scores plots. It can be concluded that most columns outside the central cluster show similar grouping and have the same properties based on the test parameters. The central cluster, positioned for both systems between the hydrophobicity and steric selectivity parameters, also shows that the two systems result in similar groups and relative positions. The loadings and scores plots were considered to be similar in the KUL and Euerby system. PC1–PC2 plots describe 70% and 69% (respectively for the KUL and Euerby’s method) of the total variation in the data set. Examination of PC3 (explaining 14% for Euerby’s method) and PC4 (explaining 10%) did not yield additional useful informa-

Fig. 5. Scores plot according to the Euerby test procedure.

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

67

Table 5 Column ranking obtained with the F-values (KUL method), relative to the ideal parameter values (k amb : 0.317, rk ba/ph 2.7 :−0.141, k 2,2 -dip : −0.106, rk tri/o-ter : −0.138) for the separation of amlodipine No

Column name

k amb

rk ba/ph pH2.7

k 2,2 -dip

rk tri/o-ter

9 52 30 22 41 61 48 15 17 53 44 40 29 57 60 11 37 58 2 26 3 12 54 63 47 5 49 45 10 18 39 23 14 55 51

Capcell Pak C18 ACR Superspher 100 RP-18 Omnispher 5 C18 Inertsil ODS-2 Prontosil 120 5 C18 SH ZorbaxExtend-C18 Restek Pinnacle II C18 Discovery HS C18 Hamilton HxSil C18 Uptisphere 5 ODB-25QS Purospher Star RP-18 Prontosil 120 5 C18H MP-Gel ODS-5 YMC-Pack Pro3 C18 Zorbax Eclipse XDB-C18 Capcell Pak C18 MG Prontosil 120 5 C18 AQ YMC-Pack Pro 5 C18 Acclaim 5 ␮m Kromasil KR100-5 C18 ACE 5 C18 Capcell Pak C18 UG120 Wakosil II5 C18 RS Zorbax SB-C18 Restek Pinnacle DB C18 Alltima C18 Restek Ultra C18 Pursuit 5 C18 Capcell Pak C18AQ Hydrospher C18 Prontosil 1205 C18ace EPS Inertsil ODS-3 Discovery C18 Xterra MS C18 Supelcosil LC-18 DB

0.385 0.679 0.667 0.434 0.392 0.635 0.148 0.803 0.329 0.579 0.641 −0.092 0.225 0.554 0.334 0.700 −0.133 0.277 0.975 1.157 −0.278 −0.094 0.313 −0.188 −0.386 0.747 1.346 −0.406 −0.571 −0.226 0.045 1.164 −0.705 −0.436 −0.614

−0.282 0.026 −0.151 −0.523 −0.091 −0.292 0.135 −0.192 −0.063 −0.136 −0.249 −0.069 −0.285 −0.385 −0.175 −0.165 −0.024 −0.486 −0.133 −0.080 −0.078 −0.358 −0.340 0.141 0.171 −0.150 −0.165 −0.135 −0.404 −0.435 −0.560 −0.426 −0.159 −0.145 0.330

−0.278 0.000 −0.230 −0.241 0.327 −0.370 −0.425 −0.186 0.382 −0.123 0.318 −0.288 0.231 −0.105 −0.449 0.016 −0.215 −0.359 −0.002 0.022 −0.734 −0.578 −0.059 −0.262 −0.712 0.843 0.076 −0.699 0.139 −0.323 −0.512 0.437 −0.856 −0.459 −0.620

−0.154 0.019 0.150 0.077 −0.328 −0.333 −0.310 −0.255 −0.305 −0.618 −0.039 −0.501 0.423 −0.846 −0.847 −0.841 −0.797 −0.840 −0.644 −0.187 −0.242 −0.774 −1.037 −0.864 −0.254 −0.097 −0.181 −0.647 −0.653 −1.088 0.832 −0.764 −0.358 −1.124 −0.730

0.055 0.194 0.221 0.225 0.231 0.232 0.236 0.259 0.272 0.300 0.306 0.337 0.456 0.618 0.623 0.657 0.662 0.679 0.701 0.728 0.763 0.844 0.850 0.887 0.972 1.088 1.094 1.135 1.182 1.331 1.356 1.485 1.657 1.665 1.703

1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.96 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

21 33 8 6 43 38 13 24 56 16 1 42 59 46

Hy PURITY C18 Polaris 5 C18-A Brava BDS C18 Alltima HP C18 Purospher RP-18e Prontosil 120 5 C18 AQ PLUS Chromolith Performance Inertsil ODS-80A Xterra RP C18 Exsil ODS 5 ␮m Acclaim 3 ␮m Prontosil 60 5 C18 H YMC-Pack Pro C18 RS Restek Allure C18

−0.847 −0.853 −1.026 −0.808 1.036 0.268 −1.162 1.632 −1.151 −0.148 −1.299 2.185 2.383 2.561

−0.114 −0.147 0.200 −0.068 −0.318 −0.410 −0.289 −0.378 −0.316 0.532 −0.005 −0.391 −0.502 −0.222

−0.929 −0.856 −0.530 −0.856 0.969 1.275 −0.880 0.831 −0.978 1.595 −1.147 0.725 0.359 0.836

−0.160 0.276 −0.077 −0.683 0.557 0.451 −0.332 −0.731 0.732 0.520 −0.762 −0.609 −0.648 −0.310

2.032 2.102 2.105 2.131 2.186 2.327 2.845 3.016 3.702 3.993 4.104 4.465 4.875 5.958

1.00 1.00 1.00 1.00 1.00 1.00 0.96 1.00 1.00 0.82 0.82 1.00 1.00 1.00

62 34 28 31 36 35 4 7 19 32 27 20 50 25

Zorbax SB—Aq Prevail Amide LiChrospher 100 RP-18 Platinum C18 Prevail Select C18 Prevail C18 Alltima AQ Alltima HP C18 Amide Hy PURITY Advance Platinum EPS C18 LiChrosorb RP-18 HyPURITY Aquastar Supelcosil LC-18 Inertsil ODS-P

−1.799 −1.503 0.513 −1.502 −0.878 −0.299 −0.372 −1.370 −1.964 −1.620 −0.458 −1.589 −0.352 1.024

0.283 −1.271 0.811 1.662 −1.130 −0.094 −0.035 −0.757 −2.248 2.975 3.307 0.438 5.096 −0.277

−0.396 −0.877 2.260 −0.687 −0.881 1.867 1.680 −1.221 −1.466 0.072 1.766 −0.386 −0.581 4.728

−1.414 1.048 0.531 −0.584 2.020 2.002 2.146 2.438 1.231 1.384 0.731 3.577 −0.527 2.348

6.369 6.588 6.989 7.092 7.663 8.853 8.887 11.101 13.369 15.811 16.745 17.845 28.253 30.063

0.00 1.00 0.95 0.00 0.98 0.00 1.00 1.00 0.36 0.00 0.77 0.00 1.00 0.35

F-value

CRF

68

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

Table 6 Column ranking obtained with the CDF-values (Euerby method), relative to the ideal parameter values (kPB : 0.395, αCH2 : 0.377, αT/O : −0.372, αC/P : −0.308, αB/P pH 7.6 : −0.373, αB/P pH 2.7 : −0.140) for the separation of amlodipine No

Column name

kPB

αCH2

αT/O

αC/P

αB/P pH 7.6

αB/P pH 2.7

41 53 48 44 9 52 60 15 22 61 11 12 58 54 30 5 2 29 3 57 37 17 45 47 23 40 18 49 24 26 43 14 55 51 63 38 21 6 10 33 39 13

Prontosil 120 5 C18 SH Uptisphere 5 ODB-25QS Restek Pinnacle II C18 Purospher Star RP-18 Capcell Pak C18 ACR Superspher 100 RP-18 Zorbax Eclipse XDB-C18 Discovery HS C18 Inertsil ODS-2 Zorbax Extend—C18 Capcell Pak C18 MG Capcell Pak C18 UG120 YMC-Pack Pro5 C18 Wakosil II5 C18 RS Omnispher 5 C18 Alltima C18 Acclaim 5 ␮m MP-Gel ODS-5 ACE5 C18 YMC-Pack Pro3 C18 Prontosil 120 5 C18 AQ Hamilton Hx Sil C18 Pursuit 5 C18 Restek Pinnacle DB C18 Inertsil ODS-3 Prontosil 120 5 C18 H Hydrospher C18 Restek Ultra C18 Inertsil ODS-80A Kromasil KR100-5 C18 Purospher RP-18e Discovery C18 Xterra MS C18 Supelcosil LC-18 DB ZorbaxSB-C18 Prontosil 120 5 C18 AQ PLUS Hy PURITY C18 Alltima HP C18 Capcell Pak C18 AQ Polaris 5 C18-A Prontosil 120 5 C18 ace EPS Chromolith Performance

0.322 0.654 0.093 0.650 0.484 0.645 0.403 0.871 0.384 0.552 0.794 −0.042 0.370 0.276 0.669 0.807 0.934 0.271 −0.374 0.812 −0.197 0.310 −0.361 −0.363 1.134 −0.112 −0.052 1.257 1.180 1.310 0.965 −0.704 −0.434 −0.584 −0.262 0.260 −0.821 −0.830 −0.463 −0.838 0.032 −1.146

0.298 0.634 0.552 0.232 0.052 0.450 0.659 0.555 0.166 0.915 0.426 0.265 0.764 0.384 0.615 0.395 0.845 0.340 0.241 0.768 0.115 0.220 0.314 0.395 0.428 −0.163 0.221 0.637 0.468 0.713 0.598 0.245 0.182 0.108 0.627 0.013 0.178 0.260 −0.764 −0.303 −0.461 0.127

−0.333 −0.622 −0.318 −0.061 −0.183 −0.002 −0.829 −0.276 0.045 −0.299 −0.833 −0.771 −0.823 −1.012 0.148 −0.082 −0.635 0.364 −0.209 −0.834 −0.764 −0.320 −0.634 −0.274 −0.750 −0.501 −1.074 −0.186 −0.665 −0.221 0.513 −0.363 −1.105 −0.729 −0.982 0.405 −0.142 −0.678 −0.662 0.276 0.819 −0.347

−0.149 −0.330 −0.506 −0.373 −0.556 −0.443 −0.304 −0.497 −0.507 −0.528 −0.353 −0.452 −0.362 −0.290 −0.510 −0.013 −0.427 −0.172 −0.314 −0.300 −0.053 0.263 −0.420 −0.561 −0.314 −0.093 −0.149 −0.522 −0.282 −0.506 −0.308 −0.462 −0.430 −0.274 0.376 0.344 −0.363 −0.307 0.046 −0.507 −0.747 −0.119

−0.236 −0.481 −0.421 −0.503 −0.503 −0.439 −0.404 −0.510 −0.280 −0.491 −0.535 −0.536 −0.521 −0.383 −0.430 0.084 −0.482 −0.189 −0.402 −0.520 −0.236 0.156 −0.537 −0.497 −0.457 0.000 −0.472 −0.510 0.125 −0.508 −0.465 −0.496 −0.538 −0.368 −0.017 0.111 −0.410 −0.321 −0.306 −0.533 −0.524 0.160

−0.116 −0.151 0.080 −0.309 −0.228 −0.038 −0.070 −0.231 −0.454 −0.218 −0.229 −0.248 −0.352 −0.255 −0.058 −0.067 −0.189 −0.208 −0.108 −0.425 −0.065 0.094 0.011 0.122 −0.373 −0.020 −0.369 −0.215 −0.205 −0.125 −0.346 −0.152 −0.215 0.442 0.114 −0.519 −0.007 −0.177 −0.411 −0.014 −0.524 −0.413

0.240 0.456 0.463 0.482 0.485 0.487 0.542 0.574 0.604 0.623 0.641 0.650 0.651 0.662 0.672 0.745 0.779 0.785 0.800 0.802 0.814 0.834 0.841 0.856 0.867 0.875 0.897 0.956 0.981 1.015 1.098 1.125 1.145 1.224 1.236 1.249 1.263 1.270 1.525 1.576 1.617 1.684

1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.96 1.00 1.00 1.00 1.00 1.00 1.00 0.96

8 28 42 1 59 46 16

Brava BDS C18 LiChrospher 100 RP-18 Prontosil 60 5 C18 H Acclaim 3 ␮m YMC-Pack ProC18 RS Restek Allure C18 Exsil ODS 5 ␮m

−1.005 0.514 1.975 −1.442 2.446 2.738 −0.111

−0.228 0.235 1.226 0.828 1.370 0.863 0.202

−0.061 0.483 −0.596 −0.693 −0.623 −0.326 0.478

0.176 0.529 0.006 −0.453 −0.510 −0.428 1.032

−0.012 0.693 0.050 −0.516 −0.527 −0.248 1.776

0.358 0.828 −0.317 −0.187 −0.418 −0.303 0.625

1.743 1.881 1.890 1.930 2.323 2.405 2.830

1.00 0.95 1.00 0.82 1.00 1.00 0.82

56 36 35 4 7 34 25 31 27 20 50 19 62 32

Xterra RP C18 Prevail Select C18 Prevail C18 AlltimaAQ Alltima HP C18 Amide Prevail Amide Inertsil ODS-P Platinum C18 LiChrosorb RP-18 Hy PURITY Aquastar Supelcosil LC-18 HyPURITY Advance Zorbax SB—Aq Platinum EPS C18

−1.193 −0.815 −0.232 −0.419 −1.411 −1.467 0.986 −1.512 −0.524 −1.646 −0.362 −1.909 −1.799 −1.669

−2.128 −1.170 −0.367 −0.294 −1.046 −2.546 0.302 −1.042 −0.088 −0.866 0.383 −4.243 −3.257 −1.851

0.600 1.937 1.973 2.168 2.395 0.958 2.511 −0.517 0.687 3.710 −0.534 0.920 −1.225 1.705

−0.659 −0.716 1.298 1.485 −0.765 −0.468 0.777 1.714 1.258 1.069 −0.388 −0.670 5.850 2.636

−0.590 −0.499 0.938 1.046 −0.404 −0.328 2.464 1.874 2.227 0.565 0.389 −0.497 0.782 5.641

−0.485 −1.175 0.038 −0.010 −1.015 −1.353 −0.131 1.682 3.016 0.344 5.261 −2.287 −0.084 2.853

3.167 3.232 3.282 3.579 3.731 3.909 4.230 4.258 4.620 5.038 5.509 5.752 7.617 8.206

1.00 0.98 0.00 1.00 1.00 1.00 0.35 0.00 0.77 0.00 1.00 0.36 0.00 0.00

CDF value

CRF

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

69

Table 7 Overview of the number of suitable columns for all seven separations within three different ranges based on a virtual, ideal column

KUL

Separation Criterion

Amlodipine (CRF = 1)

Erythromycin (CRF = 1)

Tetracycline (CRF ≥ 0.8)

Tetracaine (CRF = 1)

Bisacodyl (CRF = 1)

Fluoxetine (CRF ≥ 0.8)

Gemcitabine (CRF = 1)

F<2

34/35 (97%) 11/14 (79%) 4/14 (29%)

20/33 (61%) 0/14 (0%)

3/36 (8%)

38/39 (97%)

9/34 (26%)

35/38 (92%)

4/13 (31%)

9/10 (90%)

5/15 (33%)

3/11 (27%)

3/16 (19%)

1/14 (7%)

11/14 (79%)

5/14 (35%)

10/14 (71%)

39/39 (100%) 10/10 (100%) 12/14 (86%)

40/42 (95%) 4/7 (57%) 5/14 (36%)

21/39 (54%) 2/9 (22%) 0/15 (0%)

4/43 (9%)

42/43 (98%)

12/43 (28%)

37/43 (86%)

2/8 (25%) 2/12 (17%)

4/6 (67%) 12/14 (86%)

5/9 (56%) 2/11 (18%)

3/7 (43%) 8/13 (62%)

44/44 (100%) 5/5 (100%) 12/14 (86%)

0.88

0.90

0.82

0.86

0.64

0.84

0.87

26 Euerby

CDF < √

√

3

3 < CDF < 3 CDF > 3 CoD

For comparison, coefficients of determination values are added.

tion. The plots still indicated that kPB and αCH2 provide similar information. 4.3. Comparison based on seven pharmaceutical separations To compare both methods in a practical way, useful for analysts, 7 different pharmaceutical separations were used. These were performed after applying the KUL and the Euerby characterization methods on 63 RPLC C18 columns. These separations will be used for further comparison. A distinction between similar and dissimilar columns was made based on a ranking of all columns towards a reference column or reference parameters. First, all parameters were autoscaled and the suitable columns were selected, based on a minimum CRF value. This was for most of the separations when baseline separation of all peaks was observed (CRF = 1), but for the separations of fluoxetine and tetracycline, two partially co-eluting peaks were almost always present. Therefore, for those two separations, CRF ≥ 0.8 was used as criterion [24,25]. After detecting and omitting outliers based on the Grubbs’ test, a virtual, ideal column was calculated and used as reference column for a separation. This approach was already applied successfully onto several separations [12–14,24]. The columns were ranked using the F-value and CDF-value, respectively. Next, the ranked columns were placed in ranges and the number of suitable columns in the respective ranges was compared. Additionally, for fluoxetine and gemcitabine, besides the use of a virtual reference column, a second approach was tested, which was based on a single reference column with high or low efficiency, as was published earlier [24]. Based on the relative ranking of both methods towards the virtual, ideal column or a reference column with high or low efficiency, a coefficient of determination (R2 ) was always calculated. These strategies were used to approach an objective comparison between both systems. The above mentioned calculations were applied onto the chromatographic data of fluoxetine, gemcitabine, erythromycin, tetracycline, tetracaine, amlodipine and bisacodyl. The complete comparison for the separation of amlodipine is illustrated in Tables 5 and 6, whereas all results are summarised in Table 7. Data that are not shown, can be provided on request.

As can be seen from Table 5, the ranges used were F < 2, 2 < F < 6 and F > 6. The highest possibility to find a suitable column (CRF = 1.00) based on a virtual, ideal reference column is observed for F < 2, with a percentage of 97%. The possibility lowers to 79% for 2 < F < 6 and is the lowest when F > 6 (29%). For the Euerby (Table 6), the ranges √ method √ were recalculated to CDF < 3, 3 < CDF < 3 and CDF > 3, since the Euerby method uses 6 parameters (versus 4 with the KUL method) and the CDF is calculated as the root of the sum of the squared parameter differences. The descending trend to find a suitable column when the CDF-value increases was confirmed: 95% of √ suitable columns can be found when √ CDF < 3, 57% when 3 < CDF < 3, and finally 36% when CDF > 3. In Fig. 6, the relative order of all columns for the Euerby method is plotted against that of the KUL method. The R2 is 0.88 (y = 0.940x + 1.900). For the other separations also, similar rankings were obtained, as can be seen in Table 7, showing R2 values between 0.64 and 0.90. For Tables 5–7, a virtual ideal column was used as the reference column. This needs the availability of results on several columns, which is not the most common situation. Therefore, rankings were

Fig. 6. Correlation of the KUL and Euerby test parameters for the separation of amlodipine.

70

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71

Table 8 Overview of the number of suitable columns for the separation of fluoxetine and gemcitabine within three different ranges based on a reference column Separation Reference column Criterion

KUL

Euerby

CoD

Fluoxetine

F<2 26 √ CDF < 3 √ 3 < CDF < 3 CDF>3

Gemcitabine

Alltima C18 N = 3474 (CRF ≥ 0.8)

Prontosil 120 5 C18 AQ N = 2127 (CRF ≥ 0.8)

Inertsil ODS-3 N = 66503 (CRF = 1)

Acclaim 5 ␮m N = 24921 (CRF = 1)

26/30 (87%) 18/23 (78%) 4/10 (40%)

35/41 (85%) 10/18 (56%) 3/4 (75%)

27/27 (100%) 25/25 (100%) 9/11 (82%)

30/30 (100%) 24/24 (100%) 7/9 (78%)

35/42 (83%) 9/12 (75%) 4/9 (44%)

38/46 (83%) 6/9 (67%) 4/8 (50%)

39/39 (100%) 11/11 (100%) 11/13 (85%)

39/39 (100%) 11/11 (100%) 11/13 (85%)

0.79

0.72

0.86

0.90

For comparison, coefficients of determination values are added.

also performed, using a single column as the reference column. Table 8 gives an overview of the number of suitable columns for the separation of fluoxetine and gemcitabine for the three different ranges. As reference column, two different columns were chosen, one with a higher and one with a lower plate number (N). Again, both the KUL and Euerby methods performed in a similar way. The observed R2 values were between 0.72 and 0.90. It can be concluded that the KUL and Euerby methods do not perfectly match, but they yield very similar results, allowing with a relatively high certainty the selection of similar or dissimilar columns towards as compared to a reference column. 5. Conclusion The KUL method uses four chromatographic column parameters, while the Euerby method uses six parameters where of at least one is superfluous. Comparison of the common parameters used in both methods showed a good correlation by calculation of R2 and corresponding loadings and scores plots obtained by principal component analysis. A ranking was built based on an F-value (KUL method) or CDF value (Euerby method) versus a virtual or real reference column. The KUL and Euerby method show similar patterns, R2 values ranged from 0.64 to 0.90. Therefore, it was concluded that the methods are quite similar and both are helpful in the selection of a suitable column. From a practical point of view, it must be noted that the KUL method only includes 4 parameters and 3 chromatographic methods as compared to 6 parameters and 4 methods for the Euerby method. Hence, the time needed to determine the chromatographic properties of a column is shorter for the KUL approach. Acknowledgments The authors thank the manufacturers and the suppliers for the gift of columns. T. N. thanks the Ministry of the Flemish Community for financial support. S. D. enjoys a scholarship of the Government of Serbia, “Fund of Young Talents”. E. A. is a post-doctoral fellow of the Fund for Scientific Research (FWO)-Flanders, Belgium.

Financial support to this project was given by a Research Grant of the Fund for Scientific Research-Flanders (Belgium). References [1] European Pharmacopoeia, 5th edition, Council of Europe, Strasbourg, France, 2005. [2] United States Pharmacopeia 28, The United States Pharmacopeial Convention, Rockville, MD, 2005. [3] R.E. Majors, LC–GC Eur. 20 (2007) 13. [4] D. Visky, Y. Vander Heyden, T. Iv´anyi, P. Baten, J. De Beer, B. Nosz´al, E. Roets, D.L. Massart, J. Hoogmartens, Pharmeuropa. 14 (2002) 288. [5] D. Visky, Y. Vander Heyden, T. Iv´anyi, P. Baten, J. De Beer, Zs. Kov´acs, B. Nosz´al, E. Roets, D.L. Massart, J. Hoogmartens, J. Chromatogr. A 977 (2002) 39. [6] T. Iv´anyi, Y. Vander Heyden, D. Visky, P. Baten, J. De Beer, I. L´az´ar, D.L. Massart, E. Roets, J. Hoogmartens, J. Chromatogr. A 954 (2002) 99. [7] D. Visky, Y. Vander Heyden, T. Iv´anyi, P. Baten, J. De Beer, Z. Kov´acs, B. Nosz´al, P. Dehouck, E. Roets, D.L. Massart, J. Hoogmartens, J. Chromatogr. A 1012 (2003) 11. [8] D.V. McCalley, R.G. Brereton, J. Chromatogr. A 828 (1998) 407. [9] K. Le Mapihan, J. Vial, A. Jardy, J. Chromatogr. A 1030 (2004) 135. [10] P. Dehouck, D. Visky, G. Van den Bergh, E. Haghedooren, E. Adams, A. Kerner, Y. Vander Heyden, D.L. Massart, Zs. Kov´acs, B. Nosz´al, J. Hoogmartens, LC–GC Eur. 17 (2004) 592. [11] P. Dehouck, D. Visky, Y. Vander Heyden, E. Adams, Z. Kov´acs, B. Nosz´al, D.L. Massart, J. Hoogmartens, J. Chromatogr. A 1025 (2004) 189. [12] D. Visky, E. Haghedooren, P. Dehouck, Zs. Kov´acs, K. K´oczi´an, B. Nosz´al, J. Hoogmartens, E. Adams, J. Chromatogr. A 1101 (2006) 103. [13] E. Haghedooren, J. Diana, B. Nosz´al, J. Hoogmartens, E. Adams, Talanta 71 (2007) 31. [14] E. Haghedooren, D. Visky, P. Dehouck, K. K´oczi´an, J. Diana, Zs. Kov´acs, B. Nosz´al, J. Hoogmartens, E. Adams, LC–GC Eur. 20 (2007) 82. [15] www.pharm.kuleuven.ac.be/pharmchem/columnclassification. (2007). [16] D.L. Massart, B.G.M. Vandeginste, L.M.C. Buydens, S. De Jong, P.J. Lewi, J. Smeyers-Verbeke, Handbook of Chemometrics and Qualimetrics: Part A, 3rd ed., Elsevier, Amsterdam, 2003. [17] K. Kimata, K. Iwaguchi, S. Onishi, K. Jinno, R. Eksteen, K. Hopsoya, M. Araki, N. Tanaka, J. Chromatogr. Sci. 27 (1989) 721. [18] E. Cruz, M.R. Euerby, C.M. Johnson, C.A. Hackett, Chromatographia 44 (1997) 151. [19] M.R. Euerby, P. Petersson, LC–GC Eur. 13 (2000) 665. [20] M.R. Euerby, P. Petersson, J. Chromatogr. A 994 (2003) 13. [21] M.R. Euerby, P. Petersson, J. Chromatogr. A 1088 (2005) 1. [22] M.R. Euerby, P. Petersson, W. Campbell, W. Roe, J. Chromatogr. A 1154 (2007) 138.

E. Haghedooren et al. / J. Chromatogr. A 1189 (2008) 59–71 [23] E. Haghedooren, A. Kerner, B. Nosz´al, J. Hoogmartens, E. Adams, J. Pharm. Biomed. Anal. 44 (2007) 634. [24] K. K´oczi´an, E. Haghedooren, S. Dragovic, B. Nosz´al, J. Hoogmartens, E. Adams, J. Pharm. Biomed. Anal. 44 (2007) 894. [25] E. Haghedooren, K. K´oczi´an, S. Huang, S. Dragovic, B. Nosz´al, J. Hoogmartens, E. Adams, J. Liq. Chromatogr. Related Technol., in press.

71

[26] Pharmaeuropa, Council of Europe, Strasbourg, France 18 (2006) 494. [27] C. West, E. Lesellier, J. Chromatogr. A 1110 (2006) 191. [28] T. B˛aczeka, R. Kaliszana, K. Novotn´a, P. Jandera, J. Chromatogr. A 1075 (2005) 109. [29] P. Forlay-Frick, J. Fekete, K. H´eberger, Anal. Chim. Acta 536 (2005) 71.