New protocol for αAstree electronic tongue enabling full performance qualification according to ICH Q2

Journal of Pharmaceutical and Biomedical Analysis 83 (2013) 157–163

Contents lists available at SciVerse ScienceDirect

Journal of Pharmaceutical and Biomedical Analysis journal homepage: www.elsevier.com/locate/jpba

New protocol for ␣Astree electronic tongue enabling full performance qualification according to ICH Q2 Miriam Pein ∗ , Carolin Eckert, Maren Preis, Jörg Breitkreutz Institute of Pharmaceutics and Biopharmaceutics, Heinrich-Heine University Duesseldorf, Universitaetsstr. 1, 40225 Duesseldorf, Germany

a r t i c l e

i n f o

Article history: Received 5 February 2013 Received in revised form 19 April 2013 Accepted 7 May 2013 Available online 16 May 2013 Keywords: Taste sensing system ␣Astree Electronic tongue Performance qualification ICH guideline Q2 Sensor performance control

a b s t r a c t Performance qualification (PQ) of taste sensing systems is mandatory for their use in pharmaceutical industry. According to ICH Q2 (R1) and a recent adaptation for taste sensing systems, non-specificity, loglinear relationships between the concentration of analytes and the sensor signal as well as a repeatability with relative standard deviation (RSD) values <4% were defined as basic requirements to pass a PQ. In the present work, the ␣Astree taste sensing system led to a successful PQ procedure by the use of recent sensor batches for pharmaceutical applications (sensor set #2) and a modified measurement protocol. Log-linear relationships between concentration and responses of each sensor were investigated for different bitter tasting active pharmaceutical ingredients (APIs). Using the new protocol, RSD values <2.1% were obtained in the repeatability study. Applying the visual evaluation approach, detection and quantitation limit could be determined for caffeine citrate with every sensor (LOD 0.05–0.5 mM, LOQ: 0.1–0.5 mM). In addition, the sensor set marketed for food applications (sensor set #5) was proven to show beneficial effects regarding the log-linear relationship between the concentration of quinine hydrochloride and the sensor signal. By the use of our proposed protocol, it is possible to implement the ␣Astree taste sensing system as a tool to assure quality control in the pharmaceutical industry. © 2013 Elsevier B.V. All rights reserved.

1. Introduction As taste is a decisive quality characteristic in different lines of business, today there are many instrumental approaches for its objective evaluation [1,2]. Well established to control the quality of food and beverages [3,4], two different types of electrochemical taste sensing systems (Insent SA402B/TS-5000Z, Atsugi-Shi, Japan and ␣Astree, Alpha MOS, Toulouse, France) are commercially available. The results of these analytical tools are described to be acceptably correlated to the results of humans taste panels [3,5]. The evaluation of human taste is based on responses of receptors that are specific for defined taste sensations. The structure of these receptors is not uniform. Receptors of sour and salty taste are ion channels, whereas receptors for sweet, umami and bitter taste sensation are G-protein coupled receptors [6]. In contrast, the basic principle of the commercially available electronic taste sensing systems is based on potentiometry [5]. However, they also enable the

∗ Corresponding author. Tel.: +49 211 8114225; fax: +49 211 8114251. E-mail addresses: [email protected] (M. Pein), [email protected] (C. Eckert), [email protected] (M. Preis), [email protected] (J. Breitkreutz). 0731-7085/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jpba.2013.05.005

determination of different API concentrations and give non-specific information on taste. Electronic tongues have gained interest in the pharmaceutical drug development, especially since Pediatric Investigation Plans were introduced in 2007 [1,7–14]. To avoid later rejection by children, it would be beneficial to implement those systems in the early stage of drug development. Thus, the need increased to evaluate the (dis)advantages of electronic taste sensing systems and to assess, whether they could be used as possible tools to avoid taste panels conducted with children. In the development of drugs, analytical instruments have to be qualified and methods demonstrated to be valid. In a novel approach, ICH guideline Q2 (R1) [15] was adapted to taste sensing systems by Woertz et al. [16]. Within this approach, a successful performance qualification (PQ) was undertaken using the system Insent SA402B and quinine hydrochloride as model substance. The authors concluded that a log-linear relationship between sensor signal and the concentration of an analyte, an acceptable repeatability (relative standard deviation (RSD) values <4%) and non-specificity were basic requirements to pass a PQ. They also reported that they were not able to fulfill these requirements using the ␣Astree in an equivalent setting. Promising results obtained from our own studies on the ␣Astree encouraged us to improve the protocols to accomplish the PQ, to

158

M. Pein et al. / Journal of Pharmaceutical and Biomedical Analysis 83 (2013) 157–163

repeat previous measurements and to enable further usage in pharmaceutical industry.

2. Experimental 2.1. Chemicals and drug substances Sodium chloride (Ph. Eur., VWR International, Belgium), hydrochloric acid (1 M, Merck, Germany) and L(+)-monosodium glutamate hydrate (p.a., Sigma–Aldrich Chemie, Germany) were used as purchased. Aqua purificata was prepared via distillation of water which was previously demineralized. Caffeine citrate (DAC) was purchased from Fagron (Barsbüttel, Germany), amodiaquine hydrochloride (USP) from Ipca Laboratories (Mumbai, India), and sodium benzoate (Ph. Eur.) was donated by Ethicare (Haltern, Germany). Caffeine (Ph. Eur.) was purchased from Caesar & Loretz (Hilden, Germany) and used to prepare caffeine hydrochloride (using hydrochloric acid, 1 M, Merck, Germany), caffeine maleate (using maleic acid, p.a., Merck KGaA, Germany) and caffeine benzoate (using benzoic acid, p.a., Riedelde-Haen, Seelze, Germany). o-Phosphoric acid (85%, Gruessing, Germany), acetonitrile (Ph. Eur., Sigma–Aldrich Chemie, Germany). Hydrochloric acid (0.01 mM), L(+)-monosodium glutamate hydrate (0.01 mM) and sodium chloride (0.01 mM) for the operational qualification (OQ) of the ␣Astree were prepared using the appropriate standard solutions purchased from Alpha MOS (Toulouse, France). 2.2. Electronic tongue system and measurement setup Measurements were performed using the electronic taste sensing system ␣Astree (Alpha MOS, Toulouse, France) equipped with the sensor set for pharmaceutical applications (sensor set #2), or for the taste sensation of food (sensor set #5), an Ag/AgCl reference electrode, a stirrer and an auto sampler offering 48 beaker positions. Included cross-selective sensors were labeled ZZ, AB, GA, BB, CA, DA and JE (sensor set #2), or SRS, GPS, STS, UMS, SPS, SWS and BRS (sensor set #5), respectively. Sensors were stored dry before the starting procedure. To prove the OQ, the steps ‘conditioning’, ‘calibration’ and ‘diagnostic’ were carried out prior to experiment sequences. In between every new sequence, discriminative power of the sensors was proven by conducting at least one ‘diagnostic’ step. If sensors lacked discriminative power, also the ‘conditioning’ and ‘calibration’ step were again carried out. The experiment sequence was built according to AABBCC order (A, B, and C represent different sample concentrations). To perform the measurement, sensors were dipped into a sample beaker and each sample was analyzed over a period of 120 s subsequently eight times. After this procedure the sensors were dipped into a washing beaker, three times ten seconds, before the next sample was analyzed.

2.4. Performance qualification The PQ was carried out following the adapted guideline for electronic taste sensing systems [16]. As basic requirements to pass the PQ non-specificity, log-linearity between sensor signals and the concentration of different active pharmaceutical ingredients (APIs) and RSD values <4% for the repeatability were considered. 2.4.1. Precision Precision comprises repeatability, also called intra-day precision, intermediate precision (inter-day precision) and reproducibility (inter-laboratory precision) [15]. In this study, sensor set #2 was used to evaluate repeatability and intermediate precision by analyzing 15 different concentrations of caffeine citrate in a range of 0.001–100 mM. For repeatability investigations, intensities of the last three out of eight measurements (Section 2.3) were taken to calculate mean and standard deviation. RSD values of 4% were defined as upper limit [16]. To evaluate the intermediate precision, samples were measured three times in the aforementioned setup, while one week lay in between the first two measurements and a month lay between the second and the third run. 2.4.2. Linearity In order to investigate the relationship between concentration and sensor responses of sensor set #2, calibration curves for caffeine citrate, sodium benzoate and amodiaquine hydrochloride were established in a concentration range of 0.0005–100 mM. Loglinearity, slope of the regression line, coefficient of determination (R2 ) and residual standard deviations (sy ) were determined as described by Woertz et al. [16]. In addition, the standard deviation of the procedure (sx0 = sy /slope) and the relative standard deviation of the procedure (Vx0 ) were evaluated. The standard deviation of the procedure (sx0 ) defines the capability of a method, whereas Vx0 allows comparability between different methods. It was assessed by dividing the standard deviation of the procedure through the arithmetic mean of the linear concentration range. 2.4.3. Non-specificity Specificity is defined as the ability of an analytical method to assess one substance in presence of other substances without falsification [15]. A specific method is therefore able to identify one particular substance (or group of substances) in the presence of other compounds unequivocally. Taste sensing systems are mostly used to verify taste masking approaches. Hence, a specific method is not useful and sensors have to be non-specific. To prove that the sensors are non-specific, three different bitter substances with different structural and pharmacological properties (sodium benzoate, amodiaquine hydrochloride, caffeine citrate) in concentrations of 1 mM as well as a mixture containing these three APIs (1 mM each) were evaluated. Solutions containing 1 mM of different salts of caffeine (caffeine hydrochloride, caffeine benzoate, caffeine maleate) were compared to each other as well as to the deprotonated caffeine to investigate the influence of the counter ion.

2.3. Evaluation of the results To take sensor equilibrium into account, only the last 20 s of each measurement were taken to calculate the mean intensity. To avoid any unstable data the first five measurements were discarded for further calculations [14]. Sensor responses were used as obtained from the Alpha Soft software. As this software considers an offset value, which depends on the calibration step of the OQ procedure, sensor responses were defined as relative intensities, not absolute mV values. Data was recorded univariate using Excel 2007 (Microsoft, Redmond, US) and multivariate using SIMCA-P+ v12.0.1 (Umetrics, Umea, Sweden).

2.4.4. Accuracy As no recognized chemical reference substance of the model substance is available, accuracy had to be proven differently to Woertz et al. [16]. Within this study, results obtained by the electronic taste sensing system were compared to those obtained by a known chromatographic method. According to this method, which is described in the ‘Deutsche Arzneimittelcodex’ [17], mobile phase contains diluted o-phosphoric acid 85% (5.78 g/l) and acetonitrile (85:15, v/v), column (Nucleosil 100-5, C18, Macherey-Nagel, Dueren, Germany), flow rate (1.0 ml/min) and wavelength (272 nm) for detection were applied. Investigations were carried out under

M. Pein et al. / Journal of Pharmaceutical and Biomedical Analysis 83 (2013) 157–163

room temperature. The chromatographic system LaChrom Elite (Hitachi, Tampa, USA) being used, consists of a pump (L-2300), an autosampler (L-2200), an oven (L-2300) and an UV detector (L2400). 2.4.5. Limit of detection, limit of quantitation Based on the methods described in the ICH guideline Q2 (R1) and according to Woertz, visual evaluation was described as the most suitable method to determine LOD and LOQ concerning taste sensing systems [16]. As this has just been proven for the Insent taste sensing system, in this study all proposed approaches have been applied and evaluated using the ␣Astree. 2.4.6. Robustness An extensive robustness study has been investigated by Jespensgaard [18]. Consequently, no further measurements regarding the factors light, moisture, temperature and position of the sample beaker have been carried out within this study. However, influence of sensor age and change in the sensor set has not been determined so far. Based on a comparison of the slopes of the log-linear ranges of old and new sensors, a statistical approach was therefore carried out. 2.4.7. Determination of quinine hydrochloride using sensor sets #2 and #5 Sensor arrays #2 and #5 were used to determine the concentration dependency of the sensor responses on quinine hydrochloride. Quinine hydrochloride was therefore determined in a concentration range of 0.005–50 mM using the new measurement protocol (Section 2.2). RSD values and linearity parameters (slope of the regression-line, R2 , sy , sx0 , and Vx0 ) were calculated for both sensor sets and compared.

159

Table 2 Sensor set #2: RSD values [%] for the intermediate precision; underlined: precision requirement (RSD < 4%) is not fulfilled. Intermediate precision RSD values [%]. n = 9 Conc. [mM]

ZZ

AB

GA

BB

CA

DA

JE

0.001 0.01 0.1 1 10 100

16.62 17.84 3.85 3.02 3.66 2.95

10.48 15.19 5.69 5.72 6.81 8.79

14.21 16.09 12.40 11.23 6.86 5.42

19.81 25.72 29.12 33.68 48.52 230.23

7.94 9.85 6.39 4.29 3.39 25.85

6.49 6.96 4.77 6.19 6.37 10.94

1.17 4.71 0.72 1.78 2.64 2.39

3.1.2. Intermediate precision sensor set #2 Because only sensor ZZ and JE met the specified requirements for the intermediate precision over a broad concentration range (Table 2), it was investigated whether the use of an external standard would be beneficial. Caffeine citrate was therefore measured in two different concentrations. A solution containing a concentration of 5 mM was taken as analyte and a 1 mM solution was defined as external standard. Within one sequence, the analyte was measured in eight, respective four different sample beakers in the aforementioned setup, before the external standard was determined. But neither the drift within one sequence nor the intermediate precision could be improved by subtracting the sensor values of the standard from those of the analyte. Unlike the recommendation being made for the Insent, an external standard should not be used in ␣Astree measurements. It has therefore to be taken into consideration, that only relative but no absolute comparison of datasets is possible. However, failed intermediate precision due to drift is in accordance to the results obtained with the Insent taste sensing system. 3.2. Linearity

3. Results and discussion As final step of a successful qualification, a PQ should only be conducted after a successful OQ. In terms of the ␣Astree, sensors were therefore ‘conditioned’, before the ‘diagnostic’ power was proven with the ‘calibrated’ sensors. The PQ was principally carried out according to [16] with slight modifications regarding the accuracy and robustness using sensor set #2. It was additionally investigated, whether sensor set #5 showed a benefit regarding linearity with comparable repeatability results. 3.1. Precision 3.1.1. Repeatability sensor set #2 Using the measurement sequence AABBCC and investigating caffeine citrate in a concentration range of 0.001–100 mM, RSD values could be decreased to less than half of the values obtained with the old measurement setup ABCABC (Table 1). Table 1 Sensor set #2: RSD values [%] evaluated for the repeatability of the modified measurement sequences (n = 3); bold: RSD value is less than half of the RSD value for the same concentration using the other sequence.

Ranges of log-linearity, slope of the regression-line, coefficient of determination (R2 ), residual standard deviation (sy ), standard deviation of the procedure (sx0 = sy /slope) and the relative standard deviation of the procedure (Vx0 ) are summarized in Fig. 1. For the three investigated APIs, every sensor showed a log-linear relationship between a defined concentration range and the corresponding sensor signal. Coefficients of determination (R2 ) exhibited values >0.98. Some sensors showed higher residual standard deviations (sy ) compared to others (e.g. sensor ZZ and CA for caffeine citrate, Fig. 1). But as these sensors showed sensitive log-linear dependencies, evidenced by higher slopes, the calculated standard deviation of the procedure (sx0 ) defining the capability of the method was still desirable small. The obtained linear ranges showed acceptable values for Vx0 except from sensor GA and CA analyzing amodiaquine hydrochloride. The high Vx0 values of these particular two sensors are due to the small absolute values of the linear range and the small slopes. As caffeine citrate (Fig. 1, graph) showed the most promising results concerning log-linearity, coefficient of determination and Vx0 values, it was chosen as model substance for further investigations. 3.3. Non-specificity

AABBCC

Repeatability RSD values [%]. n = 3

Conc. [mM]

ZZ

AB

GA

BB

CA

DA

JE

0.001 0.01 0.1 1 10 100

1.69 0.47 1.04 0.52 0.49 0.50

0.32 0.17 0.34 0.29 0.15 0.13

0.35 0.08 0.66 0.76 0.93 0.23

0.16 0.25 1.45 1.60 2.06 1.48

1.14 0.43 0.32 0.26 1.68 0.10

0.10 0.28 0.62 2.44 0.47 0.19

0.42 0.19 0.13 0.03 0.03 0.06

Univariate evaluation of the obtained data showed that apart from sensor CA each sensor was able to distinguish between the different APIs and no sensor was specific for only one API (Fig. 2(a)). In addition, different salts of caffeine (individually and as mixture), the caffeine bases and a mixture containing caffeine benzoate and sodium chloride were measured to determine the influence of the counter ion. As a result, sensor responses of sensor AB are

160

M. Pein et al. / Journal of Pharmaceutical and Biomedical Analysis 83 (2013) 157–163

Fig. 1. Linearity evaluation of three bitter tasting APIs using sensor set #2; graph exemplarily shown for caffeine citrate; each concentration of the concentration series was measured in triplicate; arithmetic mean ± s.

exemplarily shown (Fig. 2(b)). While caffeine salts (abbreviated as CC, CM, CM, CB) result in remarkably decreased sensor responses compared to the corresponding bases, the counter ion showed a minor but still significant influence. When comparing the sensor responses for caffeine benzoate (CB), sodium chloride (NaCl) and a mixture containing these two components (CB + NaCl), sodium chloride showed the major impact.

Data obtained from all seven sensors was evaluated multivariate and the resulting principal component analysis (PCA, data UV scaled) was generated including the information of the two different investigations (Fig. 3). 60% of the variance is explained by the first component (PC 1), 30% by the second component. Data discrimination was made based on both axes. Regarding these information from all sensors and taking the Euclidean distances into

Fig. 2. (a) Responses of each sensor regarding three different bitter APIs (caffeine citrate, sodium benzoate, amodiaquine hydrochloride) in solutions of 1 mM and a mixture containing 1 mM of each API; (b) responses of sensor AB to different salts of caffeine (caffeine citrate (CC), caffeine maleate (CM), caffeine hydrochloride (CH), caffeine benzoate (CB)), measured individually and as a mixture, to sodium chloride (NaCl) and sodium benzoate (SB), individually and in a mixture (1 mM each), arithmetic mean ± s, n = 3.

M. Pein et al. / Journal of Pharmaceutical and Biomedical Analysis 83 (2013) 157–163

161

Fig. 3. Principal component analysis (PCA) containing the information of all seven sensors of the ␣Astree; mean (n = 3); data UV scaled.

account, the mixture containing the three different bitter APIs (indicated in Fig. 3 as ‘mixture CC + SB + amoH’) was influenced almost equally by each API. In contrast, the mixture containing caffeine benzoate and sodium chloride (NaCl + caffeine benzoate) is mainly influenced by the protonated caffeine. This result is divergent to the univariate conclusion of sensor AB and can be explained by use of all 7 sensors for the multivariate evaluation. 3.4. Accuracy Precision, linearity and specificity could be proven (Sections 3.1–3.3) for the instrument when using the new protocol. As no recognized primary pharmacopoeial standard of caffeine citrate is available to prove these results, the data obtained with the ␣Astree were compared to results obtained with a known HPLC method [17]. Two different concentrations of caffeine citrate (0.1 mM and 1 mM) were measured 6 times with the HPLC. The obtained values (0.1 ± 0.0005 mM; 1.0 ± 0.01 mM) were taken as accurate and defined as 100%. Afterwards, these two concentrations were measured in 6 subsequent cycles using the ␣Astree, each sample in triplicate. To eliminate the drift, the obtained arithmetic means were used to calculate the confidence intervals (CI) shown in Fig. 4.

Although the values obtained with the ␣Astree showed increased confidence intervals compared to the values obtained with the HPLC, only sensor DA failed to monitor accurate values for a concentration of 0.1 mM. This can easily be explained by this concentration being out of the log-linear range of this sensor (Section 3.2). The other sensors showed acceptable results for the accuracy as taste sensing systems are not considered to be high-precision devices but to confirm taste masking effects. Therefore, accuracy could be shown for the investigated concentrations and is assumed to be valid for higher and lower concentrations in the linear range of the sensors. 3.5. Limit of detection, limit of quantitation Visual determination was proven to be the most feasible method to determine the limit of detection (LOD) and limit of quantitation (LOQ). The lowest concentration showing the first sensor signal below the sensor signal of the blank was chosen as LOD. As LOQ, the lowest concentration of the log-linear dependency was defined. Results obtained with this approach were comparable to those of the signal-to-noise ratio approach (Table 3). Determination is exemplarily shown for sensor JE in Fig. 5. Since aqua purificata, which was used as blank, showed low conductivity and therefore high standard deviations, LOD and LOQ calculated based on the standard deviation of the blank led to increased values (Table 3). 3.6. Robustness

Fig. 4. Arithmetic mean values obtained from HPLC analysis were assumed to be accurate and 100%; accuracy values (␣Astree) were obtained by calculating the arithmetic mean and 95% confidence interval out of n = 6.

Storage of the sensor set utilized for the initial measurements over a period of three weeks under dry conditions led for three sensors to matchable results compared to the results obtained in Section 3.2. However, storage was accompanied with a total loss of the informative value of sensor JE, which was in use for 4 months before the storage. In addition to this observation, decreased slopes of the calibration curves of caffeine citrate were monitored for sensors GA and BB and a change in slope behavior of DA. The significance of the change in signal responses was proven by an analysis of variance (ANOVA) based on a 95% level of significance for these particular sensors (Table 4). Analyzing a concentration series of caffeine citrate and comparing the resulting sensor slopes with those of implemented ones,

162

M. Pein et al. / Journal of Pharmaceutical and Biomedical Analysis 83 (2013) 157–163

Fig. 5. Visual evaluation and signal-to-noise-ratio approach to determine the LOD; exemplarily shown for sensor JE; aqua purificata was used as blank; n = 3, arithmetic mean ± s. Table 3 LOD and LOQ determined by the four different methods described by ICH guideline Q2 (R1). Method (ICH Q2)

LOD [mM]

Visual evaluation Signal-to-noise-ratio Based on s of the blank Based on the calibration curve

LOQ [mM]

ZZ

AB

GA

BB

CA

DA

JE

ZZ

AB

GA

BB

CA

DA

JE

0.05 0.05 1.07 0.74

0.05 0.0001 1.40 0.30

0.1 0.05 1.84 0.21

0.1 0.1 1.10 0.13

0.05 0.01 0.79 0.52

0.05 0.05 1.20 0.08

0.5 0.1 0.75 0.22

0.1 0.05 3.25 2.24

0.5 0.01 4.26 0.91

0.5 0.05 5.58 0.63

0.1 0.1 3.33 0.38

0.1 0.05 2.40 1.58

0.1 0.05 3.63 0.25

0.5 0.5 2.27 0.66

Table 4 Statistical comparison of sensor slope before and after storage; t values are two sided and based on a 95% level of significance. Range: 0.01–30 mM

Slope before storage Slope after storage t(exp) t(2;0.05;6)

ZZ

AB

GA

BB

CA

DA

JE

−166.1 −176.6 0.53

−150.8 −156.5 0.28

−104.9 −52.1 8.08

−112.6 −22.9 85.0 2.45

−111.9 −137.3 1.43

−220 −208 5.20

−85.74 −5.3 97.30

Table 5 RSD values [%] of sensor signal intensities for different concentrations of quinine hydrochloride. RSD [%], n = 3 Conc. [mM]

ZZ

AB

GA

BB

CA

DA

JE

SRS

GPS

STS

UMS

SPS

SWS

BRS

0.01 0.1 1 10

0.14 0.20 0.11 0.03

0.18 0.13 0.31 0.32

0.01 0.11 0.01 0.26

0.22 0.04 0.09 0.67

1.52 1.16 1.16 1.28

0.37 0.30 0.29 0.12

2.60 0.64 0.81 0.13

0.08 0.37 0.12 0.19

0.06 0.15 0.33 2.38

1.31 0.92 0.99 0.12

0.20 0.22 0.26 0.28

0.49 0.37 0.43 1.21

0.07 0.18 0.28 0.47

1.27 0.17 0.31 0.19

Fig. 6. Determination of quinine hydrochloride using sensor sets #2 and #5; marked green: log-linear sensor responses; n = 3; arithmetic mean ± s. (For interpretation of the references to color in figure legend, the reader is referred to the web version of the article.)

M. Pein et al. / Journal of Pharmaceutical and Biomedical Analysis 83 (2013) 157–163

seemed to be a feasible extra tool to verify the performance of sensors. And in fact, this tool confirmed easily the state of questionable sensors. 3.7. Determination of quinine hydrochloride using sensor set #2 and #5 Each sensor of both sensor sets fulfills the repeatability requirements with RSD values <2.6% for the investigated quinine hydrochloride concentrations (Table 5). Advantageous and in contrast to the sensors of sensor set #2, two sensors from sensor set #5 showed a sensitive log-linear concentration dependency when detecting different concentrations of quinine hydrochloride (Fig. 6). Sensor set #5 is recommended for food application by the supplier, but it shows no adverse behavior compared to sensor set #2 regarding the conducted experiments. It rather implies an improved evaluation of bitterness, which will be discussed elsewhere more detailed. 4. Conclusions Due to new sensor batches and a newly implemented and improved measurement protocol, the ␣Astree taste sensing system equipped with the sensor set #2 for pharmaceutical applications could be proven to give reliable and valid results. The PQ procedure according to ICH Q2 as adapted for taste sensing systems by Woertz et al. [16] could with slight changes successfully be applied. Sensors showed a high repeatability based on RSD values <2.6%, which lie considerably below the required limit of 4%. To evaluate the accuracy, data obtained with the ␣Astree was compared to the results of an HPLC method. Every sensor provided non-specificity and an acceptable degree of repeatability within its log-linear range. As accuracy was assumed to be valid for this range, it was defined as working area. In the context of the robustness study, a valuable tool was introduced providing an objective and accurate assessment of the performance of questionable sensors. Additionally, findings suggest that the field of application of sensor set #5 might be extended to pharmaceutical questions. Furthermore, sensitive loglinear concentration dependency for quinine hydrochloride implies an improved evaluation of bitter substances using sensor set #5 besides sensor set #2. Using the implemented measurement protocols, the ␣Astree offers an analytical instrument providing reliable and valid data that could be used in the development of taste masked oral dosage forms.

163

Acknowledgment We thank Alpha MOS for providing the taste sensing systems ␣Astree. References [1] P. Ciosek, W. Wróblewski, Sensor arrays for liquid sensing – electronic tongue systems, Analyst 132 (2007) 963–978. [2] L. Escuder-Gilabert, M. Peris, Review, Highlights in recent applications of electronic tongues in food analysis, Anal. Chim. Acta 665 (2010) 15–25. [3] C. Eckert, C. Lutz, J. Breitkreutz, K. Woertz, Quality control of oral herbal products by an electronic tongue – case study on sage lozenges, Sens. Actuators B: Chem. 156 (2011) 204–212. [4] Y. Uchiyama, M. Yamashita, M. Kato, T. Suzuki, M. Omori, R. Chen, Evaluation of the taste of tea with different degrees of fermentation using a taste sensing system, Sensor Mater. 23 (2011) 501–506. [5] K. Woertz, C. Tissen, P. Kleinebudde, J. Breitkreutz, A comparative study on two electronic tongues for pharmaceutical formulation development, J. Pharm. Biomed. Anal. 55 (2011) 272–281. [6] B. Lindemann, Taste reception, Physiol. Rev. 76 (1996) 719–766. [7] G.A. Campbell, J.A. Charles, K. Roberts-Skilton, M. Tsundupalli, C.K. Oh, A. Weinecke, R. Wagner, D. Franz, Evaluating the taste masking effectiveness of various flavors in a stable formulated pediatric suspension and solution using the Astree electronic tongue, Powder Technol. 224 (2012) 109–123. [8] K. Woertz, C. Tissen, P. Kleinebudde, J. Breitkreutz, Rational development of taste masked oral liquids guided by an electronic tongue, Int. J. Pharm. 400 (2010) 114–123. [9] J.Y. Zheng, M.P. Keeney, Taste masking analysis in pharmaceutical formulation development using an electronic tongue, Int. J. Pharm. 310 (2006) 118– 124. [10] T. Ishizaka, Y. Miyanaga, J. Mukai, K. Asaka, Y. Nakai, E. Tsuji, T. Uchida, Bitterness evaluation of medicines for pediatric use by a taste sensor, Chem. Pharm. Bull. 52 (2004) 943–948. [11] A. Tanigake, Y. Miyanaga, T. Nakamura, E. Tsuji, K. Matsuyama, M. Kunitomo, T. Uchida, The bitterness intensity of clarithromycin evaluated by a taste sensor, Chem. Pharm. Bull. 51 (2003) 1241–1245. [12] I. Stoltenberg, G. Winzenburg, J. Breitkreutz, Solid oral dosage forms for children – formulations, excipients and acceptance issues, J. Appl. Ther. Res. 7 (2010) 141–146. [13] K. Woertz, C. Tissen, P. Kleinebudde, J. Breitkreutz, Development of a taste-masked generic ibuprofen suspension: top-down approach guided by electronic tongue measurements, J. Pharm. Sci. 100 (2011) 4460–4470. [14] K. Woertz, C. Tissen, P. Kleinebudde, J. Breitkreutz, Taste sensing systems (electronic tongues) for pharmaceutical applications, Int. J. Pharm. 417 (2011) 256–271. [15] International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use ICH, Q2 (R1): Validation of Analytical Procedures: Text and Methodology. [16] K. Woertz, C. Tissen, P. Kleinebudde, J. Breitkreutz, Performance qualification of an electronic tongue based on ICH guideline Q2, J. Pharm. Biomed. Anal. 51 (2010) 497–506. [17] Deutscher Arzneimittelcodex, Coffeincitrat, C-250, 2003. [18] L.N. Jespensgaard, Evaluation of an electronic tongue in the pharmaceutical industry – performance evaluation and sensory validation of taste masking, in: Faculty of Life Science, University of Copenhagen, Copenhagen, 2012.