Corona-Charged Aerosol Detection in Supercritical Fluid

To reduce the noise and increase the reproducibility, the transfer line (1 m × 0.25 .... Data acquisition for the CAD was carried out using the PRONT...
0 downloads 0 Views 231KB Size
Anal. Chem. 2007, 79, 2472-2482

Corona-Charged Aerosol Detection in Supercritical Fluid Chromatography for Pharmaceutical Analysis C. Brunelli,† T. Go´recki,†,‡ Y. Zhao,§ and P. Sandra*,†

Pfizer Analytical Research Centre, Ghent University, Krijgslaan 281 S4-bis, B-9000 Ghent, Belgium, and Analytical R&D-Pfizer Inc., Eastern Point Road, Groton, Connecticut 06340

In this paper we describe the coupling of a corona-charged aerosol detector (CAD) to packed column supercritical fluid chromatography (pSFC). The CAD can be coupled to pSFC by connecting the outlet of the back pressure regulator (BPR) directly to the inlet of the detector. To reduce the noise and increase the reproducibility, the transfer line (1 m × 0.25 mm i.d. stainless steel) was placed in a thermostatic bath at 45 °C. Limits of detection (LODs) ranged from 3 to 11.5 ng loaded on column, with an average value of 4.5 ng (from 0.6 to 2.3 mg/L, with an average value of 0.9 mg/L for 5 µL injection). To reduce differences in response at different mobile-phase compositions, mobile-phase flow compensation was performed by placing a T-piece before the BPR. In this way, the differences in response were significantly reduced from a factor of 2-3 to a factor of 1.2-1.7. Compared to CAD application without flow compensation, the average LOD was higher by a factor of ∼1.8. However, the nearly uniform response in gradient analysis with mobile-phase flow compensation far outweighed the slight increase in the LOD. The performance of the pSFC-CAD combination was illustrated by the analysis of selected pharmaceutically related compounds. In addition, a comparison with UV detection was made. Measurement of analytes at a relative concentration of 0.05% versus the main component was demonstrated. Packed column supercritical fluid chromatography (pSFC) has been regaining popularity in the past decade, especially in the pharmaceutical industry. The orthogonality of the normal-phase SFC separation mechanism compared to reversed-phase HPLC (RP-HPLC) makes pSFC a powerful technique that complements the toolbox of separation scientists. For example, pSFC is a technique of primary significance in the characterization of targeted combinatorial libraries, impurity profiling, and quality control, owing to its high efficiency achievable in a short analysis time.1-4 We recently described a pSFC “generic” approach for the analysis of pharmaceuticals using an ethylpyridine silica column * To whom correspondence should be addressed. E-mail: pat.sandra@ richrom.com. † Ghent University. ‡ On sabbatical leave from the Department of Chemistry, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada. § Analytical R&D-Pfizer Inc. (1) Berger, T. A.; Wilson, W. H. J. Biochem. Biophys. Methods 2000, 43, 77-85.

2472 Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

with methanol/acetonitrile mixtures as organic modifiers and formic acid (FA) and isopropylamine (IPA) as acidic and basic additives.5 In all these applications, high sensitivity of detection and a wide dynamic range are important requirements. In principle, pSFC is compatible with a wide variety of detection systems (both GC- and LC-like), yet few of them are commercially available. The most popular pSFC detector these days is the UV detector. It is characterized by good sensitivity and a wide dynamic range of over 4 orders of magnitude. In addition, its response is generally not affected by the composition of the mobile phase. On the other hand, it is not suitable for compounds with no chromophores. Lower limits of detection for some compounds (down to the picogram level) can often be achieved by selective detectors. Examples include the chemiluminescent nitrogen detector (CLND) and chemiluminescent sulfur detector (CLSD), which are becoming increasingly popular in the pharmaceutical industry owing to their high sensitivity and selectivity, as well as uniform response to compounds containing nitrogen and sulfur, respectively.6,7 Other examples of selective detectors used in pSFC include the electron capture detector (ECD)8 and fluorescence detector.9 The performance of most of these detectors suffers in the presence of organic modifiers in the mobile phase. For example, the response of the CLND decreases as the amount of methanol used as the modifier increases up to 15%.6 Combs et al. optimized the conditions to increase the maximum concentration of methanol in the mobile phase to 20%, but no signal was observed at 30%.10 Methanol concentrations not exceeding 10% were used with the CLSD, while a maximum concentration of 5% was reported for the ECD.8 (2) Zhao, Y.; Sandra, P.; Woo, G.; Thomas, S.; Gahm, K.; Semin, D. LC-GC Eur. 2004, 17, 224-238. (3) White, C.; Burnett, J. J. Chromatogr., A 2005, 1074, 175-185. (4) Zhao, Y.; Woo, G.; Thomas, S.; Semin, D.; Sandra, P. J. Chromatogr., A 2003, 1003, 157-166. (5) Brunelli, C.; Zhao, Y.; Ye, Q.; Sandra, P. Development of a Generic Method for the High Resolution Separation of Pharmaceuticals by SFC on Ethyl Pyridine Silicagel. Presented at the 29th International Symposium on Capillary Chromatography, Rive del Garda, Italy, May 29 to June 2, 2006; Posters G3 and G4. (6) Shi, H.; Taylor, L. T.; Fujinari, E. M. J. Chromatogr., A 1997, 757, 183191. (7) Shi, H.; Taylor, L. T.; Fujinari, E. M.; Yan, X. J. Chromatogr., A 1997, 779, 307-313. (8) Strode, III J. T. B.; Taylor L. T. J. Chromatogr., A 1996, 723, 361-369. (9) Smith, R. M.; Chienthavorn, O.; Dansk, N.; Wilson I. D. J. Chromatogr., A 1998, 798, 203-206. (10) Combs, M. T.; Ashraf-Khorassani, M.; Taylor, L. T. Anal. Chem. 1997, 69, 3044-3048. 10.1021/ac061854q CCC: $37.00

© 2007 American Chemical Society Published on Web 02/16/2007

One of the most desirable detector characteristics in the pharmaceutical industry, especially in impurity profiling, is universal and uniform response. Unfortunately, the selection of universal pSFC detectors is very limited. Flame ionization detection (FID) is only suitable for applications requiring no more than 1% organic modifier in the mobile phase,8 which severely limits its usefulness. Mass spectrometry is a quasi-universal detection method providing additional structural information about the analytes; therefore, it is highly desirable in pharmaceutical analysis. The coupling of SFC and mass spectrometry was recently reviewed by Pinkston et al.11 The main limitations of the pSFCMS coupling are relatively poor sensitivity and non-uniform response, strongly dependent on the nature of the analyte. An evaporative light scattering detector (ELSD), which responds to all nonvolatile compounds, can be easily coupled to pSFC, but its response is nonlinear (especially at low analyte concentrations) and depends on the composition and flow rate of the mobile phase.12-15 A promising universal detector suitable for pSFC analyses using mobile phases containing organic modifiers was recently described by Xia and Thurbide.16 The so-called acoustic flame detector (AFD) gives a uniform mass response for all analytes independently of their volatility and mobile-phase composition (although an increase in the baseline is observed when a gradient of the modifier is used). Unfortunately, the AFD is not yet commercially available. The corona-charged aerosol detector (CAD) is a new type of detector recently introduced for HPLC applications.17,18 In the CAD, the effluent from the HPLC column is nebulized and dried in a stream of nitrogen, which acts also as an ionizing gas. Aerosol particles formed by nonvolatile analytes become charged in a corona discharge source and are then detected by a sensitive electrometer. In HPLC applications, the CAD provides a quasiuniform response to nonvolatile analytes independently of their nature18-20 with sensitivity higher than that of the ELSD. However, just like for the ELSD, the response of the detector depends on the composition of the mobile phase, with higher response observed at higher organic content. An empirical solution to this problem, so-called mobile-phase compensation, was recently proposed by Go´recki et al.21 In this approach, the composition of the mobile phase reaching the detector is kept constant by mixing (11) Pinkston, J. D. Eur. J. Mass Spectrom. 2005, 11, 189-197. (12) Strode, J. T. B., III; Taylor, L. T. J. Sep. Sci. 1996, 34, 261-271. (13) Herbreteau, B.; Salvador, A.; Lafosse, M.; Dreux, M. Analusis 1999, 27, 706-712. (14) Lesellier, E.; Gaudin, K.; Chaminade, P.; Tchapla, A.; Baillet, A. J. Chromatogr., A 2003, 1016, 111-121. (15) Deschamps, F. S.; Baillet, A.; Chaminade, P. Analyst 2002, 127, 35-41. (16) Xia, Z.; Thurbide, K. B. J. Chromatogr., A 2006, 1105, 180-185. (17) Paschlau, J. GIT Labor-Fachz. 2005, 49 (1), 32-33. (18) Gamache, P. H.; McCarthy, R. S.; Freeto, S. M.; Asa, D. J.; Woodcock, M. J.; Laws, K.; Cole, R. O. LC-GC Eur. 2005, 18, 345-354. (19) Gallaghar, R. T.; Goodall, E. Evaluation of Charged Aerosol Detection for Use as a Relative or Absolute Purity Indicator. Presented at the 29th International Symposium on High Performance Liquid Phase Separations and Related Techniques, Stockholm, Sweden, June 26-30, 2005; abstract p 426. (20) Christensen, J.; Goodall, E. HPLC with Charged Aerosol Detection for the Measurement of Different Lipid Classes. Presented at the 29th International Symposium on High Performance Liquid Phase Separations and Related Techniques, Stockholm, Sweden, June 26-30, 2005; abstract p 733. (21) Go´recki, T.; Lynen, F.; Szucs, R.; Sandra, P. Anal. Chem. 2006, 78, 31863192.

the effluent from the column with a second mobile-phase stream of exactly reverse composition, delivered by an additional HPLC pump. Uniform mass response was demonstrated across the entire mobile-phase gradient with this technique. In principle, the charged aerosol detector should be easy to couple with pSFC as long as there is an organic modifier in the mobile phase; however, to the best of our knowledge, such coupling has not been reported in the literature yet. The CAD can be coupled to pSFC by connecting the outlet of the back pressure regulator (BPR) directly to the inlet of the detector. In this paper we report on the performance of the charged aerosol detector in coupling with pSFC with both constant mobile-phase composition and using an organic modifier gradient. EXPERIMENTAL SECTION Materials. The mobile phase was CO2, N45 quality, purchased from Air Liquide (Lie`ge, Belgium). LC-MS grade methanol (MeOH) used as the modifier was purchased from Biosolve (HA Valkenswaard, The Netherlands). Nitrogen 4.0 (Messner, Mechelen, Belgium) was used as the makeup gas for the CAD. All standards were purchased from Sigma-Aldrich GmbH (Steenheim, Germany). Instrumentation. A Minigram SFC system (Berger Instruments, Mettler-Toledo, Newark, DE), equipped with an FCM1100/1200 dual-pump fluid control module, a TCM-2250 heater control module, and an ALS-3100/3150 autosampler, was used in all experiments. The system was controlled and data acquisition was carried out using PRONTO software (Berger Instruments) running on a primary computer (PC1). The analyses were carried out using an ethylpyridine column, 25 cm × 4.6 mm i.d., 3 µm particle size, from Princeton Chromatography (Cranbury, NJ). The injection volume was 5 µL in all experiments. The outlet of the BPR was directly connected to the CAD via a stainless steel transfer line (1 m × 0.25 mm i.d.) bypassing the Minigram phase separator. A Julabo F10 thermostatic bath (Julabo Labortech GmbH, Seelbach, Germany) kept the temperature of the transfer line constant at 45 °C. The CAD was from ESA Analytical (Aylsbury, Buckinghamshire, England). The following parameters were used: acquisition range 500 pA, digital filter set to low, N2 pressure 35 psi. Data acquisition for the CAD was carried out using the PRONTO software. UV detection was performed using a Hewlett-Packard series 1050 diode array detector (Agilent Technologies, Waldbronn, Germany) equipped with a highpressure 6 mm flow cell installed between the column and the BPR. The UV signal was acquired using Chemstation software (Agilent Technologies) running on a second computer (PC2). Data acquisition was triggered automatically via a remote cable using the start signal from the Minigram pSFC oven module and terminated after a predefined time. In the experiments with mobile-phase flow compensation, a Hewlett-Packard series 1050 HPLC pump controlled by the PC2 was used to deliver MeOH via a 1 m × 0.12 mm i.d. PEEK tubing to a zero dead volume T-piece installed before the BPR. The pump was operated in controlled flow mode. Sample Preparation. Individual stock solutions of theophylline, testosterone, cortisone, naproxen, sulfadimidine, sulfamerazine, sulfamethoxazole, sulfaquinoxaline, and sulfamethizole were prepared in methanol at a concentration of 5000 mg/L (except for sulfaquinoxaline, for which the stock solution concentration Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

2473

was 1000 mg/L because of limited solubility). The stock solutions were stored in a refrigerator. New solutions were prepared weekly. To establish the linear range, mixtures of the nine compounds were prepared at concentrations of 10, 50, and 100 mg/L by appropriate dilutions of the stock solutions. A solution of all analytes except sulfaquinoxaline was prepared separately at a concentration of 500 mg/L. The solution of sulfaquinoxaline at 500 mg/L was prepared and analyzed individually. Two mixtures were prepared at 1000 mg/L. Mixture A included theophylline, cortisone, sulfadimidine, and sulfamethoxazole, while mixture B included testosterone, naproxen, sulfamerazine, and sulfamethizole. The stock solution of sulfaquinoxalin (1000 mg/L) was analyzed individually. The steroid mixture containing 17-methyltestosterone, testosterone, progesterone, cortisone, estrone, estradiol, and estriol was prepared in methanol at a concentration of 100 mg/L. The 0.05% pharmaceutical mixture was composed of sulfamethoxazole as the main compound at a concentration of 10000 mg/L, with sulfadimidine, sulfamerazine, and sulfaquinoxaline at 5.0 mg/L (0.05% with respect to sulfamethoxazole). Methods. pSFC analyses were run at 40 °C at a mobile-phase flow rate of 2.0 mL/min and an outlet pressure (Poutlet) of 100 bar. Isocratic analyses were performed at 10%, 20%, 30%, 35%, and 40% methanol in CO2. Gradient analyses were carried out using the following methanol program: 10% for 1 min, increasing to 35% at 2% per minute. All analyses were run in triplicate. For flow compensation of the organic modifier, a secondary stream of MeOH was introduced to the mobile phase before the BPR. The flow rate of the secondary stream was adjusted to keep the ratio of CO2 to MeOH nearly constant in the detector. Each sample was analyzed at 10%, 20%, 30%, and 35% MeOH in CO2, using compensating stream flow rates of 1.2, 0.8, 0.4, and 0.2 mL/min, respectively. Gradient analyses were run using the same methanol program as in the non-compensated experiments. The compensation flow was programmed from 1.2 mL/min (held for 3.35 min to account for the delay in the system) to 0.2 mL/min in 12 min. Analysis start was triggered automatically using the start signal from the pSFC oven module via a remote cable and finished after a predefined time. All analyses were run in triplicate. Limits of detection (LODs) were estimated using the EPArecommended procedure.22 A standard solution containing naproxen at 10 mg/L and the remaining analytes at 3 mg/L was prepared. These concentrations corresponded to 5 times the estimated limit of detection at a signal-to-noise ratio of 3, as determined from the analysis of a 100 mg/L standard. The mixture was analyzed eight times, and the LODs were calculated by multiplying the standard deviations of the peak areas by 3.

Figure 1. Configuration of the pSFC-UV-CAD system.

RESULTS AND DISCUSSION The nearly universal response to nonvolatile compounds, higher sensitivity compared to that of the ELSD and dynamic range extending over ∼4 orders of magnitude, were the main features of the CAD that determined its success as a detector in HPLC. The main drawback of this detector, i.e., strong dependence of its response on the mobile-phase composition, could be easily circumvented by using the mobile-phase compensation technique.21

The CAD belongs to the category of so-called “open cell” detectors, in which the mobile phase is converted to gas before detection of the solutes.23 For such detectors, two approaches are generally followed: a pre-BPR and a post-BPR connection. To connect an open cell detector between the column outlet and the BPR while maintaining the correct pressure in the SFC system, the mobile phase should be split, with only a fraction of it delivered to the detector and the rest passing through the BPR. The mobilephase pressure can be maintained and actively controlled by the BPR in such a configuration provided that the transfer line offers sufficient restriction. However, delivering only a fraction of the mobile phase to the detector results in lower sensitivity, and low amounts of impurities might be impossible to detect. On the other hand, when an open cell detector is connected after the BPR, transferring the entire sample to the detector is straightforward. Considering that high sensitivity and a wide dynamic range are some of the most important features required by pharmaceutical analysis, only the post-BPR configuration was considered in this study. After the BPR, when pressure in the transfer line becomes sufficiently low, pressurized fluid CO2 turns into gas, and only the CO2-saturated organic modifier present in the mobile phase remains in the liquid form. As long as the amount of the modifier in the mobile phase is sufficient, nonvolatile and semivolatile analytes should remain dissolved in it and not precipitate. In this way, complete transfer of the analyte to the detector can be accomplished while the stability of the pressure and flow conditions in the SFC system are preserved. In HPLC, the response of the CAD depends strongly on the composition of the mobile phase. Higher organic content in the mobile phase leads to an increase in the transport efficiency of the nebulizer, which results in a greater number of particles reaching the detector chamber and in a higher signal.24 In SFC, on the other hand, only the CO2-saturated organic modifier reaches the detector in the liquid form, independently of the current composition of the mobile phase. Thus, one could reasonably expect a constant response of the detector for different compositions of the mobile phase, which would be very advantageous. CAD-pSFC Coupling. Preliminary experiments were performed using the instrumental setup illustrated in Figure 1. The UV detector was used in line with the CAD for comparative purposes. The results of these experiments proved that CAD can be a useful detector in pSFC, as it produced good response to different

(22) U.S. EPA. Definition and Procedure for the Method Detection Limit, Revision 1.11; Environmental Protection Agency: Washington, DC, 1984; 40 CFR Part 136, Appendix B.

(23) Hill, H. H., Jr.; Gallagher, M. M. J. Microcolumn Sep. 1990, 2, 114-119. (24) Cobb, Z.; Barret, D.; Shaw, P.; Meehan, E.; Watkins, J.; O’Donohue, S.; Wrench, N. J. Microcolumn Sep. 2001, 13, 169-175.

2474 Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

compound classes. However, the baseline noise was higher than expected, and the response was poorly reproducible on a day-today basis. The root cause of both problems was traced back to a lack of temperature control in the system. When CO2 turns from a pressurized fluid state into a gas, its adiabatic expansion leads to cooling. Visual observations of the spray leaving the outlet of the transfer line in the absence of temperature control demonstrated that the spray was nonuniform, and frequent sputtering occurred. After the BPR, as the density, pressure, and temperature drop below a certain point, the one-phase binary mobile phase may produce two separate phases, a CO2-rich gaseous phase and a modifier-rich liquid phase,25 resulting in segmented flow compromising the quality of the chromatogram; this, in fact, caused instability and spikes in the baseline. The problem was circumvented by replacing the short transfer line used in the preliminary experiments with a longer line, which was coiled and immersed in a thermostated bath. Different temperatures were tested (results not shown). On the basis of these experiments, 45 °C was selected as the optimum temperature. The length of the transfer line provided sufficient restriction and heat transfer to prevent the occurrence of segmented flow. The spray leaving the transfer line under these conditions was uniform, and sputtering was eliminated. Higher temperatures did not produce any further improvement. Apart from reducing the noise, thermostating of the transfer line also improved the reproducibility of the results. This was most likely related to the reduction of the internal temperature variations in the detector. The version of the CAD used in the study did not have internal temperature control. As the temperature in the laboratory varied somewhat from day to day (and also during the day when it was sunny), the temperature of the detector changed, and so did its response. With the thermostating, these internal temperature variations were most likely significantly reduced, leading to more reproducible results. It should be pointed out that a recently introduced new CAD model (unavailable to us for the study) does include temperature control of the nebulizer; therefore, poor reproducibility caused by detector temperature variations should no longer be an issue. For further experiments, a series of pharmaceutically relevant compounds with different structures and chemical nature, namely, theophylline, testosterone, cortisone, naproxen, sulfadimidine, sulfamerazine, sulfamethoxazole, sulfaquinoxaline, and sulfamethizole, were chosen to evaluate the CAD response after separation on an ethylpyridine column under different conditions. Calibration curves were determined for each analyte in the concentration range from 10 to 1000 mg/L. Higher concentrations were not included because the limited solubility of some analytes could cause their precipitation in the transfer line and/or distortion of peak profiles. Figure 2 presents the calibration curves of the analytes determined for 20% MeOH in the mobile phase. The curves in Figure 2a, plotted in linear scale, show that the response was not linear, which was expected considering that in HPLC the CAD has a quadratic response. When plotted in logarithmic coordinates, which is the standard way of presenting the calibration curves for the CAD in HPLC, the curves were nearly linear (Figure 2b). The response factors for the individual compounds differed somewhat more than expected, with the (25) Chester, T. L.; Pinkston, J. D. J. Chromatogr., A 1998, 807, 265-273.

differences close to a factor of 2 for some compounds. The response factors reported for the CAD in HPLC were more uniform, although for different solutes.19,20 Nevertheless, the result was still considered good taking into account the vastly different natures of some of the analytes. Figure 3 illustrates the dependence of the response factors of the detector on the mobile-phase composition for two analytes, cortisone (Figure 3a) and sulfamethoxazole (Figure 3b). The dependences for the remaining analytes looked similar. Contrary to expectations, the responses changed significantly when the percentage of the organic modifier in the mobile phase changed. In a stark contrast to HPLC, the response factors generally decreased when the amount of the organic modifier increased, with the exception of cortisone, for which an initial increase was observed between 10% and 20% modifier in the mobile phase at lower concentrations. This finding could not be explained by changes in nebulizer transport efficiency brought about by varying organic content in the mobile phase (as is the case in HPLC24), considering that the liquid fraction reaching the CAD in pSFC was always CO2-saturated methanol of nearly constant composition. It seems much more likely, instead, that the response changes were caused by changes in the flow rate of the gaseous CO2 accompanying the change in the organic modifier percentage in the mobile phase. In HPLC, the liquid mobile phase is converted to aerosol in the nebulizer of the CAD. In pSFC, on the other hand, as the pressure drops at the outlet of the BPR, the mobile phase begins its conversion into the gas-liquid aerosol in the transfer line and enters the detector in this form already. Consequently, the CAD nebulizer plays only a marginal role in the generation of the aerosol and subsequent formation of analyte particles. On the other hand, changes in the CO2/organic modifier ratio directly affect the generation of the aerosol at the inlet of the detector and therefore affect the response. Apparently, the nebulization efficiency was higher at a lower percentage of the organic modifier (thus a higher flow rate of CO2), which would explain the generally higher responses obtained under these conditions. Figure 4 presents the averaged calibration curves (linear and logarithmic) for all analytes used in the study determined for five different isocratic compositions of the mobile phase and for gradient elution. Quasi-linearity of the response was maintained under all conditions (Figure 4b), but the response varied by a factor of slightly more than 2 when going from 10% to 40% of the organic modifier, with the response in gradient elution similar to that obtained at 40% MeOH. Such response variation, while significant, was much lower than that observed in HPLC (where it reached nearly 1 order of magnitude21). Reproducibility of the response was estimated by running each analysis in triplicate. The estimated average relative standard deviations (RSDs) of the results ranged from 7% at 1000 mg/L to 9% at 5 mg/L (the latter value was obtained in the limit of detection study; therefore, it was based on eight rather than three results). Interestingly, the RSDs of the results did not follow a monotonous trend; a minimum in RSD was observed for analyte concentrations of 100 mg/L instead. Poorer precision at higher analyte concentrations was most likely caused by limited solubility of some of the analytes, which could have led to their partial precipitation in the transfer line followed by resolubilization in the pure MeOH. Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

2475

Figure 2. Calibration curves for the standard mixture analyzed in isocratic mode at 20% MeOH: (a) linear scale, (b) logarithmic scale.

The sensitivity of the CAD in pSFC was evaluated by determination of the LODs of the analytes in gradient elution. The LODs for the individual compounds ranged from 0.6 to 2.3 mg/L (5 µL injection), with an average value of 0.9 mg/L (corresponding to 3-11.5 ng loaded on column, with an average value of 4.5 ng). In the case of UV detection at 254 nm, the LODs for the same compounds were slightly lower, ranging from 0.3 to 0.9 mg/L, with an average value of 0.4 mg/L (from 1.5 to 4.5 ng on column, with an average of 2.0 ng). At 210 nm, the LODs ranged from 0.6 to 1.2 mg/L, with an average value of 0.8 mg/L (from 3.0 to 6.0 ng on column, average 4.0 ng). Interestingly, in the concentration range examined (with the highest concentrations limited by analyte solubility), the slope of the averaged calibration curve obtained with the CAD was significantly larger than the slope of the averaged calibration curve obtained with the UV detector (Figure 5). This did not translate into lower LODs because of higher baseline noise observed with the CAD. 2476

Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

Overall, it can be concluded that the LODs obtained with the CAD were comparable to those achievable with the UV detector for the analytes used in the study. It should be kept in mind, though, that the CAD produces a response also for compounds with no chromophores, which is a significant advantage of this detector. Mobile-Phase Flow Compensation. In spite of the clear advantages of the pSFC-CAD coupling described above, the dependence of the detector response on the mobile-phase composition is a significant drawback, especially in applications where nearly uniform response is crucial (e.g., impurity profiling in pharmaceutical preparations). In HPLC, a similar problem was solved by the mobile-phase compensation approach,21 in which the composition of the mobile phase reaching the detector was kept constant in gradient runs by mixing the column effluent with an additional stream of the mobile phase of reverse composition

Figure 3. CAD response factor vs the percentage of MeOH in the mobile phase for cortisone (a) and sulfamethoxazole (b).

before the detector. It seemed interesting, therefore, to try an analogous approach in pSFC. Postcolumn mixing of the chromatographic effluent with an additional stream of a fluid is not new on the SFC scene. For example, low flow of CO2 was added to the chromatographic effluent by a secondary pump to control the outlet pressure before SFC was interfaced with APCI-MS;25 in another application, a mixture of MeOH/water was added to the effluent after the column to deliver the solute to the mass spectrometer when the mobile phase contained only small amounts of the organic modifier.26 Lesellier et al. used postcolumn addition of a solution containing formic acid and triethylamine to increase the ELSD sensitivity.14 Sandra et al. delivered a methanolic solution of silver ions to induce ionization when triglycerides in vegetable oils had to be characterized by silver ion pSFC coupled to mass spectrometry with atmospheric pressure chemical ionization and coordination ion spray;27 also, methanol flow was used to interface SFC to ELSD and ESI-MS.25 However, to the best of our knowledge, postcolumn addition of an organic solvent aimed at obtaining a uniform response independent of the mobile-phase composition in SFC has not been described in the literature yet. Mobile-phase compensation as described by Go´recki at al.21 requires a mobile phase of a reverse composition to be added to the column effluent. In HPLC, this could be easily accomplished by using an additional binary pump. In SFC, however, another (26) Baker, T. R.; Pinkston J. D. J. Am. Soc. Mass Spectrom. 1998, 9, 498-509. (27) Sandra, P.; Medvedovici, A.; Zhao, Y.; David, F. J. Chromatogr., A 2002, 974, 231-241.

Figure 4. Averaged calibration curves for the nine analytes obtained at different MeOH concentrations in the mobile phase in isoconfertic mode and in gradient elution: linear scale (a), logarithmic scale (b).

Figure 5. Averaged calibration curves for the nine analytes obtained using CAD and UV detection at 254 nm.

complete SFC system would have to be used to provide both pressurized CO2 and the organic modifier in correct proportions. While theoretically possible, such an approach seemed impractical. It was decided, therefore, to try the compensation with pure organic modifier (MeOH) only. This could be accomplished Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

2477

Figure 6. Configuration of the pSFC-UV-CAD system with flow compensation.

relatively easily by mixing the SFC effluent with MeOH delivered by an additional, flow-programmable single-channel HPLC pump. Since no changes in the composition of the mobile phase reaching the detector were involved (always pure organic modifier saturated with CO2), this approach was called mobile-phase flow compensation. The additional MeOH stream could be added to the effluent either before or after the SFC BPR. Sandra et al.27 reported that pre-BPR mixing of the additional stream might lead to accelerated failure of the expensive restrictor nozzle; hence, the latter possibility might seem preferable. However, this approach has significant disadvantages, the most important of which is the possibility of analyte precipitation in the transfer line between the BPR and the mixing T at low organic modifier concentrations. In addition, it was observed that pre-BPR mixing led to more uniform aerosol formation at the outlet of the transfer line. Finally, the upstream location of the secondary pump operated in flow-control mode allowed better control of the system pressure by the BPR itself. Consequently, pre-BPR mixing was used in all experiments. No deterioration of the BPR performance was noticed over the 3 month duration of the study, most likely owing to a better design of the nozzle in present SFC instrumentation compared to the one used in previous work.27 Figure 6 shows the schematic diagram of the experimental setup used in the study. The compensating MeOH stream was delivered by an HPLC pump controlled by the PC2 (see the Experimental Section). The MeOH stream was mixed with the SFC column effluent in a zero dead volume T installed directly before the BPR. In this configuration, the flow of CO2 could not be compensated; therefore, it changed whenever the composition of the mobile phase changed. Two approaches to the compensation of the CAD response were examined. In the first approach, the flow rate of the additional MeOH was adjusted to keep the amount of MeOH reaching the detector constant at all times. In the second approach, the amount of the additional MeOH was adjusted to keep the CO2/MeOH

Figure 7. CAD response factor vs the percentage of MeOH in the mobile phase for cortisone (a) and sulfamethoxazole (b) with flow compensation.

ratio in the detector nearly constant. Preliminary results demonstrated that the second approach produced a more uniform response. This observation can be explained when the mechanism of the variable response proposed above is taken into consideration. With the total flow of MeOH kept constant in the detector, the ratio of CO2 to MeOH varied considerably at different compositions of the mobile phase. As a result, the nebulization efficiency, and consequently the size (thus also the number) of the aerosol droplets, varied as well, leading to more pronounced differences in the detector response. In the second approach, the CO2/MeOH ratio remained nearly constant while the total flow varied, which led to more uniform aerosol generation. The CO2 and MeOH flow rates used in this approach are summarized in Table 1. It should be emphasized that both approaches produced more uniform responses compared to conventional SFC. Figure 7 shows the dependence of the response factors of the detector on the mobile-phase composition for cortisone and sulfamethoxazole using mobile-phase flow compensation. Com-

Table 1. Summary of the CO2 and Methanol Flow Rates Used in the Flow Compensation Experimentsa

CO2 flow rate (mL/min) MeOH (SFC) flow rate (mL/min) MeOH (comp) flow rate (mL/min) total mobile phase flow rate reaching the detector (mL/min) a

10% MeOH in mobile phase

20% MeOH in mobile phase

30% MeOH in mobile phase

35% MeOH in mobile phase

40% MeOH in mobile phase

1.80 0.20 1.20 3.20

1.60 0.40 0.80 2.80

1.40 0.60 0.40 2.40

1.30 0.70 0.20 2.20

1.20 0.80 0.00 2.00

MeOH (SFC) ) MeOH delivered by the SFC pump. MeOH (comp) ) flow compensation MeOH delivered by the HPLC pump.

2478 Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

Figure 8. Averaged calibration curves for the nine analytes obtained at different MeOH concentrations in the mobile phase in isoconfertic mode and in gradient elution with flow compensation: linear scale (a), logarithmic scale (b).

pared to conventional analysis (Figure 3), the differences in response factors at different mobile-phase compositions were significantly reduced (especially at higher concentrations, for which the data were more reliable). The reduced dependence of the response on the mobile-phase composition is also evident when Figure 4 is compared with Figure 8, which shows averaged calibration curves (linear and logarithmic) obtained at different compositions of the mobile phase with flow compensation. The reproducibility of the results obtained with the CAD using flow compensation was estimated by running each analysis in triplicate. The estimated average RSDs of the results ranged from 9% at 1000 mg/L to 15% at 5 mg/L; a minimum in the RSD values was again observed around 100 mg/L. The slightly worse precision of the results obtained with flow compensation was most likely related to somewhat higher noise observed in this case.

The sensitivity of the CAD under flow compensation conditions was evaluated by determination of the LODs of the analytes in gradient elution. The LODs for the individual compounds ranged from 0.6 to 2.5 mg/L, with an average value of 1.6 mg/L (from 3.0 to 7.5 ng on column, average 8.0 ng). Compared to CAD application with no flow compensation, the average LOD was higher by a factor of ∼1.8, again due to slightly higher noise. However, the advantages of obtaining a nearly uniform response in gradient elution outweigh the slight decrease in sensitivity observed. Figure 9 shows chromatograms of the mixture of nine analytes (100 mg/L each) obtained in gradient analysis with flow compensation using the CAD (Figure 9a) and UV detection at 254 nm (Figure 9b). A comparison of the two chromatograms shows clearly that the response of the CAD was significantly more Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

2479

Figure 9. Chromatograms of the mixture of the nine analytes (100 mg/L) obtained using gradient elution: (a) CAD with flow compensation, (b) UV at 254 nm. Peak identities: 1, theophylline; 2, testosterone; 3, cortisone; 4, naproxen; 5, sulfadimidine; 6, sulfamerazine; 7, sulfamethoxazole; 8, sulfaquinoxaline; 9, sulfamethizole; *, unidentified impurity in the methanol used to prepare the standards.

uniform than the response of the UV detector. In addition, the CAD response remained nearly uniform throughout the gradient for the different analytes, naproxen being the only exception. Peak widths were somewhat greater with the CAD (by ∼17% on average), which is understandable considering the relatively large internal volume of this detector. The increase in peak width was in fact comparable to that observed in our previous study with HPLC.21 The “hump” on the CAD chromatogram near the end of the run was the result of imperfect compensation; a significant baseline drift was observed in the noncompensated chromatogram of the mixture obtained using gradient elution (not shown), which could not be completely eliminated using the technique of mobilephase flow compensation. Figure 10 shows a comparison of the chromatograms of a mixture of seven steroids (100 mg/L each) obtained with the CAD (a) and the UV detector at 210 nm (b) and 254 nm (c). It is clear from this comparison that the performance of the CAD in this application was again superior. The responses of all seven analytes were nearly uniform, which would make quantitation easy, even with a single calibrant. On the other hand, the UV detector did not produce a uniform response at either of the two wavelengths examined. At 254 nm, estrone, estradiol, and estriol produced very weak responses; at 210 nm, 17-methyltestosterone, testosterone, and cortisone produced lower responses. Thus, to obtain optimal results, quantitation of the analytes with the UV detector would 2480

Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

have to be carried out at different wavelengths using standards of each individual analyte rather than a single calibrant. For the same reason, quantitation of unidentified impurities is practically impossible using the UV detector. An impurity check of a drug substance is still one of the most important and challenging analytical tasks in the pharmaceutical industry. The guidelines of the International Conference on Harmonization (ICH), including the current Good Manufacturing Practice (cGMP), have to be taken into consideration when new methods are being developed. One of the parameters required for the validation of quantitative impurity methods used for the registration process is the determination of impurities at the 0.05% level with respect to the main compound.28-30 The easiest route to accomplish this goal is to use a detector with a dynamic range covering at least 4 orders of magnitude and with universal, uniform response to all compounds. Few (if any) detectors are able to satisfy these requirements in fluid-based separation methods. For example, while the UV detector has a sufficient dynamic range, its response is strongly dependent on the molecular structure of a compound, with no response produced for analytes containing (28) Anton, K.; Siffrin, C. Analusis 1999, 27 (8), 691-701 and references therein. (29) Impurities in new drug substances; ICH-Q3A(R2); International Conference on Harmonization: Geneva Switzerland, October 2006. (30) Impurities in new drug products; ICH-Q3B(R2); International Conference on Harmonization: Geneva, Switzerland, June 2006.

Figure 10. Chromatogram of the mixture of seven steroids (100 mg/L) obtained in gradient elution. Detection with the CAD using flow compensation (a) and UV at 210 nm (b) and 254 nm (c). Peak identities: 17-methyltestosterone (1), testosterone (2), progesterone (3), cortisone (4), estrone (5), estradiol (6), estriol (7).

Figure 11. Chromatogram of simulated impurities at the 0.05% level with respect to the main compound (sulfamethoxazole at 10000 mg/L). Concentration of the simulated impurities: 5 mg/L each. Peak identities: 1, sulfadimidine; 2, sulfamerazine; 3, sulfamethoxazole; 4, sulfaquinoxazline.

no chromophores. In addition, the noise level observed in SFC is usually higher than in HPLC due to the complex nature of pressurized binary mobile phases and compressibility of nearcritical fluids. For these reasons, quantitation of impurities at the 0.05% level is not always possible with this detector. The charged

aerosol detector seems to be a suitable tool for impurity profiling owing to its universal, uniform response and wide dynamic range. Figure 11 shows a chromatogram obtained for a 5 µL injection of a standard solution with simulated impurities at the 0.05% level. All analytes could be easily quantified at this level. Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

2481

CONCLUSIONS In spite of the versatility of pSFC, its use in the pharmaceutical industry is still confined to niche applications. Recently, a generic method suitable for initial screening in drug development has been developed,5 but the lack of sufficiently sensitive universal detectors suitable for gradient elution currently precludes the use of pSFC in impurity profiling, which is one of the most important stages of the drug discovery process. The research presented in this paper demonstrates the great potential of the charged aerosol detector in pSFC. The CAD responds to all nonvolatile compounds soluble in the mobile phase, and its response is fairly uniform. Although this study presents the results only for one organic modifier, methanol, similar trends were observed in preliminary studies with other modifiers, including ethanol, 2-propanol, and acetonitrile. Further improvement in response uniformity can be obtained by using the flow compensation technique. Consequently, the CAD has the potential to become the detector of choice in the development of pSFC-based analytical methods for drug discovery. The research presented in this paper was performed using an

2482

Analytical Chemistry, Vol. 79, No. 6, March 15, 2007

unmodified HPLC version of the CAD. Further improvements of the performance of this detector might be obtained by modifying it to better suit the specific requirements of pSFC. For example, it seems likely that the noise could be reduced with a modified sample introduction system and temperature control of the nebulizer. The latter feature is already standard on the current generation of the CAD. Unfortunately, this version of the detector was not available to us for testing. ACKNOWLEDGMENT We thank ESA Analytical (Aylsbury, Buckinghamshire, England) for the loan of the charged aerosol detector and Mettler Toledo (AutoChem Inc., Newark, DE) for providing the SFC instrumentation used in this work.

Received for review October 3, 2006. Accepted January 4, 2007. AC061854Q