Improving the universal response of evaporative light scattering detection by mobile phase compensation

Transcription

1 Journal of Chromatography A, 1161 (2007) Improving the universal response of evaporative light scattering detection by mobile phase compensation André de Villiers a,1, Tadeusz Górecki a,2,frédéric Lynen a, Roman Szucs b, Pat Sandra a, a Pfizer Analytical Research Centre, Ghent University, Krijgslaan 281 S4-bis, B-9000 Ghent, Belgium b Analytical R&D, Pfizer Limited, Ramsgate Road, Sandwich, Kent, CT13 9NJ, United Kingdom Received 12 January 2007; received in revised form 22 May 2007; accepted 25 May 2007 Available online 31 May 2007 Abstract Mobile phase compensation, first reported for the charged aerosol detector (CAD), was used as a suitable method to overcome problems related to the mobile phase-dependent response of the evaporative light scattering detector (ELSD). Mobile phase compensation was effectively performed both in the flow injection- and in gradient modes. Without compensation, the response factors of the ELSD for six sulfonamide drugs differed by a factor of two when varying the mobile phase composition between 10 and 90% acetonitrile. This change could be effectively eliminated using the technique of mobile phase compensation, where a secondary pump with a reversed gradient was used to provide the detector with a constant composition of the mobile phase. For identical experimental conditions, the ELSD showed a nearly constant, albeit somewhat reduced, response with compensation. This indicates that under such conditions, the ELSD behaved as a concentration-sensitive detector. The analysis of sulfonamides drugs at 0.05% level using gradient UPLC-ELSD separation with mobile phase compensation is demonstrated Elsevier B.V. All rights reserved. Keywords: Liquid chromatography; Evaporative light scattering detection; Solvent compensation; Sulfonamides 1. Introduction The ever-increasing demand for high throughput analyses in the pharmaceutical industry has led to mounting interest in fast, quantitative and generally applicable HPLC methods. This aspect has gained added significance with the advent of the combinatorial approach to drug development [1]. The demand for fast and reliable quantitative screening methods of complex pharmaceutical samples often containing structurally diverse compounds has placed the emphasis on the development of more universally applicable detection strategies. Since analyte-specific calibration protocols are rarely feasible in the high-throughput domain of the pharmaceutical sciences, the focus is increasingly shifting toward single-calibrant quantification methods in HPLC. Corresponding author. Tel.: ; fax: Present address: University of Stellenbosch, Department of Chemistry, Private Bag X1, Matieland 7602, South Africa. 2 On sabbatical leave from the Department of Chemistry, University of Waterloo, Waterloo, ON N2L 3G1, Canada. The most common HPLC detectors, UV and MS, suffer from non-uniform responses due to differences in absorptivities and ionization efficiencies as a function of chemical structure, respectively. Although MS detection in particular remains a powerful complementary tool for structural elucidation in screening methods, alternative detection methods are required for more accurate purity and quantitative measurements. Suitable detection methods proposed for this purpose include evaporative light scattering detection (ELSD) [2 5], chemiluminescent nitrogen detection (CLND) [6 8] and more recently flame ionization detection (FID) [9] and charged aerosol detection (CAD) [10,11]. Whilst FID holds promise for future applications in LC using pure aqueous mobile phases, this method is still in the development phase and its routine usage cannot be envisioned in the foreseeable future. CLND has been shown to be a very effective quantitative detector for pharmaceutical products containing nitrogen [12,13]. Drawbacks of this detection method include diminished response for compounds containing adjacent nitrogen atoms [14], incompatibility with nitrogen-containing mobile phases, and the high cost and complexity of the detector /$ see front matter 2007 Elsevier B.V. All rights reserved. doi: /j.chroma

2 184 A. de Villiers et al. / J. Chromatogr. A 1161 (2007) For non-volatile analytes, both ELSD and CAD exhibit massdependent responses, claimed to be independent of the analyte s chemical structure. Considering the instrumental simplicity and relatively low cost, each of these detectors presents an attractive alternative for use in a high-throughput screening environment. While routine application of CAD for pharmaceutical analysis is currently in the evaluation phase [11], numerous reports on the use of ELSD for this purpose have been published. However, general implementation of ELSD for quantitative purposes is still hampered by several factors, the most important of which is the non-linear and mobile phase-dependent response of this detector. During the more than 20 years since the introduction of ELSD, a number of reports have proclaimed a more uniform response for the ELSD, especially compared to workhorse LC detectors such as UV and MS [3,15 17]. For sufficiently representative standard compounds (e.g. a selection of standards from a combinatorial library), errors in quantitation by ELSD in the region of 10 20% are obtained using a general calibration curve, compared to up to 60% for UV detection at different wavelengths [2,4,5]. Theoretically, for sufficiently non-volatile analytes, the response of the ELSD should be approximately analyte-independent [18]. Caution should be exercised, however, since this criterion is not always met in practice and reduced response for volatile impurities can lead to an over-estimation of analyte purities by ELSD [13]. Thus, while significantly more uniform response is the norm in comparison to UV detection, quantitative precision using general calibration curves decreases rapidly for compounds that are not structurally related, or in cases when volatile compounds are present in the samples [5]. The non-linear response of the ELSD is a consequence of the dependence of the efficiency of prevalent light-scattering processes, namely Rayleigh scattering, Mie scattering and reflection refraction, on the average particle size [18]. The particle size is in turn determined by numerous experimental variables (nebulizer design, mobile phase properties, nebulizing gas flow rate, temperature, etc.) as well as the solute concentration. Maximum light scattering efficiency is obtained for particles in the region of 0.1 m [19]. It has been shown that low analyte concentrations lead to particles smaller than the optimal value, resulting in reduced response. Similarly, at high concentrations, particles larger than the optimal value are produced, so that less light is scattered per unit mass. These phenomena are responsible for the sigmoidal calibration curve profile typically observed for the ELSD. In practice, log log curves are constructed to provide the linear curve used for quantitative purposes [4,5]. It has been shown that a consequence of this logarithmic response is that uncalibrated impurities are underestimated, leading to inflated purity estimation of main compounds [13]. The response of ELSD also increases with increasing concentration of the organic modifier in the mobile phase. This phenomenon leads to variations in response factors as a function of retention time in gradient analysis, as observed for ELSD [20,21] and CAD [11]. Mobile phase properties such as surface tension, density and viscosity are known to determine the size and number of the droplets produced by the nebulizer, and in this manner the size of the particles reaching the detection cell [18,19]. Differences in response during a water/acetonitrile gradient have been related to the different properties of these liquids [20]; other authors have ascribed the response increase to increased transport efficiency of the nebulizer [21]. Another consideration is the effect of peak width on the ELSD response. Van der Meeren et al. demonstrated that the calibration curves varied as a function of peak width under otherwise identical conditions [22]. Broader chromatographic peaks resulted in smaller particles and less photomultiplier saturation. All the above mentioned effects lead to variations in ELSD response across a gradient, and as such represent a significant obstacle to single calibrant procedures. Mathews et al. [21] proposed a three-dimensional calibration procedure to effectively compensate for the mobile-phase dependence of ELSD response. In this approach, a single non-retained compound was injected at regular intervals during gradient analysis at different concentrations. A three-dimensional calibration surface was constructed based on the data obtained and used to quantify the analytes. In this manner variations in the response during gradient analysis (resulting both from mobile phase and peak width changes) were accounted for in the quantitation process. However, this method required numerous injections, and the resultant calibration surface was method specific. More recently, Górecki et al. [23] proposed an elegant approach based on mobile phase compensation to correct for the same phenomenon using CAD. In this method, a secondary stream of the mobile phase of exactly reverse composition provided by a second pump was added to the column effluent to ensure a constant mobile phase composition at the detector inlet. This resulted in constant response independent of the mobile phase composition in the column. Compared to the method of Mathews et al., this approach has the advantage of requiring fewer injections for calibration. In addition, calibration runs do not have to be repeated when the gradient is altered, since the compensation gradient can be changed in a complementary manner. A consequence of this methodology is that a single suitably representative standard can be used to quantify all analytes, including unknown components, in gradient analysis (provided that all sample components are sufficiently non-volatile). Since CAD and ELSD are both aerosol-based detectors, it was suggested that gradient compensation should work equally well for ELSD [23]. Removal of the principal source of response variations (mobile phase composition) should improve significantly the utility of the ELSD within the high-throughput pharmaceutical environment, principally by increasing the precision of single-calibrant procedures. In the current contribution, we report the application of the mobile phase compensation approach to ELSD detection. 2. Experimental 2.1. Materials LC MS grade water, acetonitrile (ACN) and formic acid used for the preparation of the mobile phases and samples were from Biosolve (HA Valkenswaard, The Netherlands). Sulfamethoxa-

3 A. de Villiers et al. / J. Chromatogr. A 1161 (2007) zole, sulfadimethoxine, sulfamerazine and sulfamethizole were obtained from Riedel-de Haën (Seelze, Germany), while sulfamethazine and sulfaguanidine were from Sigma-Aldrich Chemie, GmbH (Steenheim, Germany). Stock solutions of the individual sulfonamides were prepared at a concentration of 5 mg/ml in acetonitrile and stored in the refrigerator for no longer than 5 days. Sample solutions were prepared fresh daily by dilution of the stock solutions using water/acetonitrile (90/10) Instrumentation An Acquity UPLC system equipped with a binary solvent manager, sample manager, sample organizer, column heater, PDA detector and ELS detector (Waters, Milford, MA, US), was used in all experiments. Mobile phase compensation in the flow injection mode was performed using an Agilent 1050 pump (Agilent, Waldbronn, Germany). For gradient compensation, an Agilent 1200 LC system equipped with a binary pump (highpressure mixing) was used. The mixing column was replaced by a union to minimize the delay volume. The Agilent pumps were connected to the UPLC flow stream after the PDA detector using mm I.D. PEEK tubing. Both detectors and the UPLC were controlled using Empower software (Waters); the Agilent pumps were controlled by Chemstation software (Agilent). The mobile phases for all experiments consisted of water containing 0.1% formic acid (phase A) and acetonitrile (phase B). Injections were performed in the full-loop mode (1.9 L or 10 L) using phase A as a weak needle wash solvent. UV signals at 210 and 254 nm were recorded on the PDA using an acquisition rate of 20 points per second (pps, Hz). The following optimized ELSD settings were used throughout the study, unless otherwise specified: drift tube temperature 51 C, nebulizer gas pressure 25 psi and unheated nebulizer. Nitrogen 4.0 (Messner, Mechelen, Belgium) was used as the nebulizing gas. 3. Methods 3.1. Flow injection analysis Calibration samples of the individual sulfonamides were prepared at 10, 50, 100, 500 and 750 mg/l as described above. 0.1 mm I.D. PEEK tubing was used to connect injector outlet directly to the PDA inlet, and the ELSD to the PDA outlet. The flow rate in these experiments was 0.5 ml/min, and each sample was injected in triplicate (1.9 L) with blank injections (90/10 water/acetonitrile) between each sample. Experiments were performed for five different compositions of the mobile phase: 10, 30, 50, 70 and 90% phase B. Both UV and ELSD signals were recorded. Mobile phase compensation experiments in the flow injection mode are described in Section Characterization of system gradients Experiments were performed to measure the gradient profiles of the UPLC- and 1200 systems using a mobile phase composition of 30/70 acetonitrile/water (phase A) and 30/69.5/0.5 acetonitrile/water/acetone (phase B). The Agilent 1200 pump was connected to the PDA inlet with red PEEK tubing as in the compensation experiments. A step gradient was programmed as follows: 0 1 min 0%B, min 0 10%B, min 10%B, min 10 20%B, min 20%B, min 20 30%B, min 30%B, min 30 40%B, min 40%B, min 40 50%B, at a constant flow rate of 0.5 ml/min. The UV signal recorded at 254 nm was used to monitor the increase in acetone content of the mobile phase UPLC gradient analysis UPLC separations were performed on an Acquity BEH C18 column, 2.1 mm I.D. 100 mm L, packed with 1.7 m particles (Waters). Calibration curves were constructed using mixtures of the 6 sulfonamides at 10, 50, 100, 500 and 750 mg/l as the average values of triplicate injections (1.9 L). The following gradient was used: min 10% B, min 10 90% B, min 90% B, before returning to the initial conditions at 5.9 min. The flow rate was 0.5 ml/min and the column was equilibrated for 4.1 min at the end of each run. The limit of detection (LOD) for each of the sulfonamides was calculated according to the EPA-recommended procedure [24]: a standard solution containing all analytes at 10 mg/l (S/N = ) was injected seven times, and the standard deviations of the peak heights were calculated for each analyte. LOD-values were obtained by multiplying the standard deviation by 3. The procedure for UPLC analysis with mobile phase compensation is described in Section UPLC gradient analysis of sulfonamides at 0.05% A sample was prepared containing 5000 mg/l sulfadimethoxine and 2.5 mg/l of the remaining sulfonamides. For these experiments the gradient slopes were reduced as follows to allow detection of sulfamethoxazole in the presence of excess sulfdimethoxine: UPLC: min 10% B, min 10 90% B, min 90% B, min, 90 10%B; 1200: min 10% A, min 10 90% A, min 90% A, min 90 10% A. A higher drift tube temperature (60 C) was used, the nebulizer was heated (25% power level) and a 10 L loop was used for these experiments. 4. Results and discussion ELSD responds to all compounds with sufficiently low vapor pressures. The response factors towards different analytes are reasonably uniform at a given composition of the mobile phase, but they change as the organic content of the mobile phase changes. In a previous study involving the charged aerosol detector [23], we proposed to use a new method, so-called mobile phase compensation, to eliminate this problem. Considering that the mechanism of variable response is similar in CAD

4 186 A. de Villiers et al. / J. Chromatogr. A 1161 (2007) and ELSD, the mobile phase compensation method should be applicable to the latter detector as well. The analytes chosen for this research, as well as the methodology used, were similar to those used in the CAD study [23] to facilitate a comparison of the effectiveness of the mobile phase compensation approach for ELSD. A more detailed explanation of the rationale for using the six sulfonamides (sulfamethoxazole, sulfadimethoxine, sulfamerazine, sulfamethizole, sulfamethazine and sulfaguanidine) as the analytes was given in the previous contribution. In brief, these drugs were selected because they all have very low vapor pressures and they elute over a broad range of mobile phase compositions from C18 columns. Flow injection was used to determine the response factors of the analytes at mobile phase compositions ranging from 10 to 90% ACN in water with 0.1% HCOOH. Each compound was injected three times at each concentration level and each mobile phase composition. Fig. 1 presents the calibration curves obtained for all six analytes using a mobile phase consisting of 50% ACN in water with 0.1% HCOOH. As expected, the calibration curves were non-linear (Fig. 1A), but could be easily linearized by plotting them in a logarithmic coordinate system (Fig. 1B). The lines in Fig. 1 are the lines of best fit for power regression in the form of y = ax b. The fits were very good, with the R 2 values above in all cases. The response factors for the analytes were similar, with the difference between the highest and the lowest response factor on the order of 16%. Similar results were obtained at other compositions of the mobile phase studied. The R 2 values for the power regression were greater than in all cases, and the differences between the lowest and the highest response factors did not exceed 20%. Fig. 2 presents averaged calibration curves for all six analytes in both linear and logarithmic coordinates obtained at the different compositions of the mobile phase in flow injection analysis. The error bars in the figure correspond to ±1 standard deviation. Similarly to the calibration curves for the individual compounds, the dependences were non-linear, but could be easily linearized by converting them to logarithmic coordinates. The strong dependence of the detector response on the mobile phase composition is obvious from this figure. The peak areas at 90% ACN in water were nearly twice as large as at 10% ACN. This difference, while significant, was not nearly as large as that observed for the CAD, for which the peak areas at the two compositions differed by a factor of nearly 5 [23]. On the other hand, a similar trend was observed, with the differences being more pronounced when going from low to medium organic modifier concentrations, and less pronounced when going from medium Fig. 1. Calibration curves obtained for all six analytes using a mobile phase consisting of 50% ACN in water with 0.1% HCOOH in flow injection analysis. (A) linear co-ordinates and (B) logarithmic co-ordinates. Error bars show one standard deviation. Fig. 2. Averaged calibration curves obtained for all six sulfonamides at different compositions of the mobile phase in flow injection analysis. (A) Linear coordinates and (B) logarithmic co-ordinates, including power regression lines. Error bars show one standard deviation.

5 A. de Villiers et al. / J. Chromatogr. A 1161 (2007) to high concentrations. This similarity is not surprising considering that both detectors require that the sample is nebulized prior to analyte detection, and the composition of the mobile phase affects the size of the droplets formed [18,19] and the transport efficiency of the nebulizer [21], both of which affect the response of these detectors. It should be pointed out that in contrast to our previous study with the CAD [23], no blank correction was required with the ELSD, even at the lowest analyte concentrations. Under comparable conditions, the CAD always produced a significant signal for blank injections, while the ELSD did not. Considering that the solvents and analytes used in both studies were exactly the same, this seems to indicate that the blank signal observed in the CAD was a transient caused by a change in the composition of the mobile phase (momentarily replaced by a plug of the injection solvent) rather than a real signal caused by unidentified impurities. The dependence of the ELSD response on the mobile phase composition is a significant drawback, especially in applications in which the universal and uniform response of the detector is crucial (e.g. in impurity profiling in pharmaceutical analysis). To eliminate this drawback, Mathews et al. [21] proposed a combined experimental computational three-dimensional calibration method. Mobile phase compensation introduced in our previous work [23] accomplished the same goal by always supplying the detector with a mobile phase of a constant composition. This was realized by combining the effluent from the column with an additional stream of pure mobile phase of exactly reverse composition and identical flow rate before the detector. To test the feasibility of the mobile phase compensation method with ELSD, a PEEK tee was connected between the DAD and the ELSD, and a stream of the mobile phase of reversed composition was supplied with an external HPLC pump to the side arm of the tee. Column effluent was fed to the second side arm. The combined stream was then fed to the ELSD through the lower arm of the tee. The results obtained are illustrated in Fig. 3, showing averaged calibration curves obtained for all six analytes at different compositions of the mobile phase with mobile phase compensation. The Y-axis scale in Fig. 3 was kept the same as in Figs. 1 and 2 to facilitate direct comparisons. Fig. 3 clearly illustrates the effectiveness of the mobile phase compensation method in combination with ELSD in flow injection analysis. The averaged calibration curves were nearly identical at all compositions of the mobile phase studied. The differences did not exceed 5% and were statistically insignificant. Interestingly, the highest response was observed at the lowest concentration of ACN, which might indicate that a very slight overcompensation occurred. This is not surprising considering that the two pumps used in the experiment came from different manufacturers and that nominal rather than absolute flow rates were used. The power regression lines fitted the experimental data points very well, with the R 2 values greater than in all cases. It is worth noticing that the response of the detector in the experiment with mobile phase compensation was similar to the response obtained at 10% ACN in the mobile phase without compensation (Fig. 2). This result was somewhat unexpected Fig. 3. Averaged calibration curves obtained for all six analytes at different compositions of the mobile phase using mobile phase compensation in flow injection analysis. (A) Linear co-ordinates and (B) logarithmic co-ordinates. Error bars show one standard deviation. considering that the mobile phase reaching the detector contained 50% ACN when mobile phase compensation was used. The reduction in the response was possibly caused by the lower concentration of the analyte in the mobile phase reaching the detector due to the dilution of the column effluent by the additional mobile phase stream, in which case the ELSD acts like a concentration- rather than mass-sensitive detector. Another possible explanation is that the increased flow rate of the mobile phase led to a decrease in the size of the droplets generated by the nebulizer [17 19] and the corresponding decrease in the size of the analyte particles formed. It would then be possible to improve the response of the ELSD by optimization of the detector parameters (e.g. the nebulizing gas flow rate), but this was beyond the scope of this study. In any case, in the absence of optimization of the ELSD nebulizer parameters, the detector acted like a concentration-sensitive detector. Interestingly, no dependence of the detector response on the mobile phase flow rate was observed with the charged aerosol detector in our previous work [23], most likely due to the different detection principle, not relying on light scattering by aerosol particles (which is dependent on the particle size). Thus, in practice, the CAD behaved as a mass-sensitive detector, which is advantageous in the case of the mobile phase compensation method. The flow injection experiments confirmed the usefulness of the mobile phase compensation method in combination with ELSD under isocratic conditions. However, since most real

6 188 A. de Villiers et al. / J. Chromatogr. A 1161 (2007) world HPLC separations are carried out using gradient elution, the performance of the method had to be tested under these conditions as well. In our previous study [23], we used a second identical HPLC pump configured with the same components as in the primary channel (column, mixer, connecting tubing, etc.) to accomplish synchronized changes of the mobile phase composition in both channels. This approach could not be used in the study presented herein because of a lack of access to a second UPLC system (or pump). The compensating stream was generated instead using an Agilent 1200 system. Since the design of the Agilent system was different than that of the Waters system, optimization was required first to make sure that the gradients generated by the two instruments had similar shapes and arrived at the mixing tee precisely at the same time. Initial experiments were performed to characterize the gradients formed by each system by making use of an acetone-containing mobile phase as outlined in Section 2. The results indicated that the gradients generated by the 1200 system were characterized by a significantly longer time constant and could not be made as sharp as those generated by the UPLC system. To eliminate this problem, the mixer on the 1200 system was replaced with a low-volume union. With this modification, similar gradient profiles could be obtained with both instruments. Gradient timing was adjusted by correcting for differences in delay volume as measured in the step gradient experiments. The following optimized programs were established based on these experiments: UPLC: min 10% B, min 10 90% B, min 90% B, min, 90 10% B. 1200: min 10% A, min 10 90% A, min 90% A, min 90 10% A. The optimized program for the 1200 system was very close to the program used with the UPLC system, even though no column was installed in the former. This was likely a result of the increased delay volume caused by the PEEK tubing used to connect the two instruments. Fig. 4 presents chromatograms of the analyte mixture obtained with UV (A) and with ELSD without (B) and with (C) mobile phase compensation, together with the profiles of the mobile phase composition in the column and in the compensating stream. The chromatogram in Fig. 4B (without compensation) shows an increase in the response of the ELSD detector with increasing content of the organic modifier in the mobile phase. While the increase in the response was evident, it was not as dramatic as that observed with the CAD under similar conditions (see Fig. 4 in ref. [23]). This observation was in agreement with the results obtained using flow injection under isocratic conditions, which indicated an increase in the response by a factor of 2 when going from 10 to 90% ACN in the mobile Fig. 4. Chromatograms with superimposed organic modifier composition obtained for the separation of the 6 sulfonamides at 100 mg/l. (A) UV signal in normal gradient run; (B) ELSD signal without mobile phase compensation in a gradient run and (C) ELSD signal with mobile phase compensation. Peak identification: 1, sulfaguanidine; 2, sulfamerazine; 3, sulfamethazine; 4, sulfamethizole; 5, sulfamethoxazole; 6, sulfadimethoxine.

7 A. de Villiers et al. / J. Chromatogr. A 1161 (2007) phase (the actual increase in peak area was slightly lower due to the fact that the last peak, sulfadimethoxine, eluted at 50% acetonitrile). Fig. 4C shows a chromatogram of the same sample obtained using the mobile phase compensation method. The response factors were significantly more uniform in this case, as evidenced by the nearly equal peak sizes. The size of the peaks in this run was roughly equivalent to the size of the early eluting peaks in the runs without compensation, which was consistent with the observations made in the flow injection study, indicating that the ELSD behaved as a concentration-sensitive detector under the experimental conditions employed. The height of the sulfaguanidine peak (peak 1 in Fig. 4) was slightly lower than the other peaks, although the area was similar for all peaks. This was due to the fact that sulfaguanidine eluted during the isocratic period of the gradient. The improved uniformity of the response factors was further confirmed by the calibration curves presented in Fig. 5. Without mobile phase compensation (Fig. 5A and C), the response factors of the different analytes differed by as much as a factor of 2; the differences were reduced to less than 27% with compensation (Fig. 5B and D). This spread was somewhat greater than that observed for the individual analytes at any given mobile phase composition in the flow injection experiments (less than 20%), most likely due to non-perfect compensation of the mobile phase composition (recall that systems from two different manufacturers were used to generate the synchronous gradients). Nevertheless, this improvement is significant, especially when one considers that the same calibration curves would be applicable for any gradient profile as long as the initial and final composition of the mobile phase in the gradient stayed the same (which means that the composition of the combined mobile phase reaching the detector remained constant). The reproducibility of the results was overall very good. The average relative standard deviation at the 10 mg/l level was 9.2% (n = 7), with the individual %RSD values for all compounds lower than 12% for all analytes. The RSD values at higher concentration levels were much better, with average relative standard deviation of 2.7% measured at 750 mg/l, and RSD-values lower than 4.5% for all compounds. The limits of detection were estimated for the mobile phase compensation method using the EPA-recommended procedure [24] (see Section 2) using peak heights rather than areas, as the heights determine the signal-to-noise ratio. The values obtained were 1.3, 1.6, 2.0, 2.7, 2.3 and 2.9 mg/l for sulfaguanidyne, sulfamerazin, sulfamethazine, sulfamethizol, sulfamethoxazole and sulfadmiethoxin, respectively, with an average value of 2.2 mg/l. This corresponds to 4.1 ng (1.9 L injection) oncolumn. In other studies, Gamache et al. [11] reported LODs of ng for ELSD; Young et al. mention ng as typical for this detector [16]; Megoulas reported detection of 40 ng on-column [25], while Cobb et al. [26] reported LODs of ng for 1 mm I.D. column, and ng for 0.3 mm I.D. columns for underivatized amino acids. While direct comparisons of these LOD values are difficult because of the variety of drift tube configurations (type A or B), analytes, columns, Fig. 5. Calibration curves obtained for the six analytes by HPLC in gradient analysis: (A) without compensation, linear co-ordinates; (B) with mobile phase compensation, linear co-ordinates; (C) without compensation, logarithmic coordinates and (D) with mobile phase compensation, logarithmic co-ordinates. Error bars show one standard deviation.

8 190 A. de Villiers et al. / J. Chromatogr. A 1161 (2007) separation conditions and LOD definitions used, the sensitivity achieved in this study was certainly very good. The LODs determined herein were higher by a factor of 4.4 compared to the values determined for the CAD in the previous study [23], which was expected considering the shape of the calibration curves typical for ELSD (lower sensitivity at low concentrations). The LODs determined with the ELSD under mobile phase compensation conditions could probably be further improved by optimizing the operating conditions of the detector, but this was beyond the scope of this study. Detection of impurities in pharmaceutical preparations at 0.05% level is of utmost importance for the pharmaceutical industry. The task is complicated by the fact that many of those impurities might be completely unknown or uncharacterized. It is for this reason that the industry is interested in a universal detector, responding uniformly to all possible analytes. Thus far, no detector meets this requirement; consequently, a common strategy involves the use of various detector combinations. ELSD is one of the detectors that are used most often in such combinations, together with mass spectrometric detectors, nitrogen chemiluminescence detectors, UV detectors, etc. The research presented in this paper has the potential to significantly improve the usefulness of ELSD in impurity profiling by eliminating the root cause of the variable response of this detector, i.e. the dependence of the response on the mobile phase composition. Fig. 6 illustrates the separation of a standard solution containing sulfadimethoxine at 5000 mg/l, with the remaining analytes at the 2.5 mg/l level, corresponding to 0.05% with respect to the main component. The additional peaks detected after sulfadimethoxine are impurities associated with the main compound. Under these conditions, all analytes could be detected at this level. It should be kept in mind that the ELSD is not currently considered a suitable detector for the analysis of impurities down to the 0.05% level. While ELSD certainly cannot be used as the sole detector in impurity profiling because of its lack of response to volatile compounds and average sensitivity, the results obtained indicate that it can complement other detectors very well, and its usefulness can be further enhanced by mobile phase compensation. The data presented herein are of particular relevance to the pharmaceutical industry, given its intense interest in UPLC as a possible means of improving sample throughput and/or the quality of analytical data. To the best of our knowledge, this is the Fig. 6. Chromatogram obtained for the 0.05% level analysis. Ten microliter injection of 5000 mg/l sulfadimethoxine standard solution containing 2.5 mg/l (each) of the remaining five sulfonamides; peak identification see Fig. 4. For further experimental details, refer to Section 2. first report describing the application of ELSD in combination with UPLC. 5. Conclusions The mobile phase compensation method was originally developed for the charged aerosol detector, whose response depends very strongly on the organic content of the mobile phase. In the paper introducing the method [23], we predicted that it would be equally applicable to the evaporative light scattering detector, since both CAD and ELSD operate on a similar principle involving evaporation of the mobile phase and detection of the analyte particles formed. The results presented in this paper confirm the predictions. Mobile phase compensation was shown to significantly improve the uniformity of the ELSD response. However, differences were also observed: while the CAD behaved as a mass-sensitive detector whose response did not change when the effluent from the column was diluted with the compensation stream, in practice the ELSD behaved as a concentration-sensitive detector, whose response decreased under similar conditions. While it is likely that the sensitivity of the ELSD under mobile phase compensation conditions could be improved through optimization of the detector parameters, further work is required to clarify this aspect. Acknowledgements The authors would like to thank Waters (Milford, MA, USA) for the kind loan of the ELSD detector. References [1] G.E. Boldt, T.J. Dickerson, K.D. Janda, Drug Discov. Today 11 (2006) 143. [2] C.E. Kibbey, Mol. Divers. 1 (1996) 247. [3] H.B. Hsu, E. Orton, S.-Y. Tang, R.A. Carlton, J. Chromatogr. B 725 (1999) 103. [4] L. Fang, M. Wan, M. Pennacchio, J. Pan, J. Comb. Chem. 2 (2000) 254. [5] L. Fang, J. Pan, B. Yan, Biotechnol. Bioeng. 71 (2001) 162. [6] N. Shah, M. Gao, K. Tsutsui, A. Lu, J. Davis, R. Scheuerman, W.L. Fitch, R.L. Wilgus, J. Comb. Chem. 2 (2000) 453. [7] D.A. Yurek, D.L. Branch, M.-S. Kuo, J. Comb. Chem. 4 (2002) 138. [8] B. Yan, L. Fang, M. Irving, S. Zhang, A.M. Boldi, F. Woolard, C.R. Johnson, T. Kshirsagar, G.M. Figliozzi, C.A. Krueger, N.J. Collins, J. Comb. Chem. 5 (2003) 547. [9] E.W.J. Hooijschuur, C.E. Kientz, U.A. Brinkman, J. High Resol. Chromatogr. 23 (2000) 309. [10] R.W. Dixon, D.S. Peterson, Anal. Chem. 74 (2002) [11] P.H. Gamache, R.S. McCarthy, S.M. Freeto, D.J. Asa, M.J. Woodcock, K. Laws, R.O. Cole, LC GC Europe 18 (2005) 345. [12] M.C. Allgeier, M.A. Nussbaum, D.S. Risley, LC GC North America 21 (2003) 377. [13] S. Lane, B. Boughtflower, I. Mutton, C. Paterson, D. Farrant, D.N. Taylor, Z. Blaxill, C. Carmody, P. Borman, Anal. Chem. 77 (2005) [14] D. Corens, M. Carpentier, M. Schroven, L. Meerpole, J. Chromatogr. A 1056 (2004) 67. [15] A. Stołyhwo, H. Colin, G. Guiochon, J. Chromatogr. 265 (1983) 1. [16] C.S. Young, J.W. Dolan, LC GC Europe 16 (2003) 132. [17] N.C. Megoulas, M.A. Koupparis, Crit. Rev. Anal. Chem. 35 (2005) 301. [18] J.M. Charlesworth, Anal. Chem. 50 (1978) [19] T.H. Mourey, L.E. Oppenheimer, Anal. Chem. 56 (1984) [20] P.A. Asmus, J.B. Landis, J. Chromatogr. 316 (1984) 461.

9 A. de Villiers et al. / J. Chromatogr. A 1161 (2007) [21] B.T. Mathews, P.D. Higginson, R. Lyons, J.C. Mitchell, N.W. Sach, M.J. Snowden, M.R. Taylor, A.G. Wright, Chromatographia 60 (2004) 625. [22] P. Van der Meeren, J. Vanderdeelen, L. Baert, Anal. Chem. 64 (1992) [23] T. Górecki, F. Lynen, R. Szucs, P. Sandra, Anal. Chem. 78 (2006) [24] US EPA s Methods for Organic Chemical Analysis in Industrial Wastewater, EPA-600/ [25] C.N. Megoulas, M.A. Koupparis, J. Pharm. Biomed. Anal. 36 (2004) 73. [26] Z. Cobb, P.N. Shaw, L.L. Lloyd, N. Wrench, D.A. Barrett, J. Microcol. Sep. 13 (2001) 169.