2.1 Chemicals and materials

Analytical-grade standards (Sigma-Aldrich, Saint Louis, Missouri, United States of America, USA) of CAFF and PAR (Fig. 1A and B) in powder form were used for the preparation of the stock solutions. Panadol Extra® tablets (GlaxoSmithKline, London, England, United Kingdom) containing 65 mg of CAFF and 500 mg of PAR in a conventional-release formulation, were used in the method validation experiments.

View Full Size

Analysis of caffeine and paracetamol using diol column. (A) Structure of caffeine molecule and its corresponding molecular form. (B) Structure of paracetamol molecule and its corresponding molecular form. (C) Diol column with caffeine and paracetamol in the mobile phase

Citation: Acta Chromatographica 2025; 10.1556/1326.2025.01316

• Download Figure
• Download figure as PowerPoint slide

The HPLC-grade reagents acetonitrile (ACN) (Fisher Scientific, Pittsburgh, Pennsylvania, United Kingdom), ammonium acetate, and glacial acetic acid (Lachner, Neratovice, Czech Republic) were used for the preparation of the mobile phase. The HPLC grade water was obtained using Barnstead LabTower EDI (Thermo Fisher Scientific, Waltham, Massachusetts, USA). The prepared mobile phases were filtered through a 0.45 μm nylon membrane filter (Merck, Darmstadt, Germany) and deaerated in an ultrasound bath (Bandelin electronic, Berlin, Germany).

2.2 Chromatographic conditions

The Agilent 1200 series chromatographic setup (Santa Clara, California, USA) included a quaternary pump G1311A, an autosampler ALS G1329A, a thermostat TCC G1316A, a detector DAD G1315D, and the data collection system Agilent ChemStation. Separations were conducted using HILIC mode with an Inertsil Diol stationary phase (250 × 4.6 mm, 5 μm particle size) from GL Sciences, Tokyo, Japan.

The mobile phase comprised a mixture of ACN and an ammonium acetate buffer. The content of the organic solvent, the buffer concentration, and the pH value of the aqueous phase were systematically varied according to the plan obtained using DoE (Table S1). The optimal mobile phase was determined to be a mixture of ACN and ammonium acetate (0.017 M, pH 5.8) at a volume ratio of 94.5 : 5.5.

2.3 Standard solutions

CAFF and PAR stock solutions (1 mg mL⁻¹) were prepared in ACN. Working solutions were prepared from stock solutions by diluting with an appropriate mobile phase, according to the experimental plan. All samples were stored at 4 °C for up to seven days to prevent degradation. A placebo mixture consisting of pregelatinized starch, corn starch, povidone, potassium sorbate, talc, stearic acid, carmellose sodium cross-linked, water, hypromellose (6CPS), and triacetin was prepared to match the individual concentrations of excipients found in the Panadol Extra® tablets.

Linearity was assessed by preparing five solutions with range concentrations of 5.0–100.0 μg mL⁻¹ for CAFF and 50.0–250.0 μg mL⁻¹ for PAR (Fig. S1). Accuracy was evaluated through the preparation of three series of three solutions each containing a placebo, CAFF, and PAR. The concentrations for CAFF were 16.0, 20.0, and 24.0 μg mL⁻¹, and for PAR 123.2, 154.0, and 184.8 μg mL⁻¹ (corresponding to 80, 100, and 120% relative to the declared value). Precision was assessed using Panadol Extra® tablets containing 65 mg of CAFF and 500 mg of PAR, preparing six solutions with concentrations of 20 μg mL⁻¹ for CAFF and 154 μg mL⁻¹ for PAR. Selectivity was examined by preparing solutions containing only a placebo and solutions containing a mixture of CAFF, and PAR at concentrations of 20 μg mL⁻¹ and 154 μg mL⁻¹, respectively, and placebo. The limit of detection (LOD) and limit of quantification (LOQ) were determined separately for CAFF and PAR by preparing solutions with decreasing concentrations: 1.0, 0.5, 0.25, 0.10, 0.05, 0.025, and 0.01 μg mL⁻¹.

2.4 Software

Design Expert 13.0.0. (Stat-Ease Inc., Minneapolis, Minnesota, USA) was used for the creation of an experimental plan and data analysis. Marvin Sketch 15.4.13. (ChemAxon Kft, Budapest, Hungary) was used for obtaining log D and pKa values. MODDE 13 (Sartorius AG, Göttingen, Germany) was used for the creation of DS.

The AGREE software is based on the principles of GAC, with each section of the divided circle, comprising 12 parts, aligning with one of the 12 principles [15]. ComplexGAPI covers processes occurring before the actual analysis but does not provide a numerical interpretation of the results [16]. These software were used in combination for the evaluation of the greenness of the method since they were found to provide reliable and precise results [27]. BAGI software, primarily based on evaluating the efficiency of the method in terms of the number of analyzed samples per hour, the number of compounds tested per analysis, and similar factors were used for evaluation of the practicality of the method. This is a relatively new step in these types of analyses, but given the parameters being evaluated, it is crucial for assessing environmental acceptability [17]. An asteroid pictogram with 10 fields covers 10 attributes evaluated by BAGI where hues of blue color indicate the performance of the method – the darker the color, the more practical the method (in terms of acceptability and applicability). The principles of WAC, that combine the GAC and the functionality of the method, were applied through the RGB model. In this model R (red) represents analytical efficiency, G (green) the compliance with GAC principles, and B (blue) the economic and practical efficiency. Analytical efficiency is crucial as it reflects how well a method meets validation criteria. This ensures the reliability of analytical results, making them a trustworthy basis for decision-making. Compliance with the GAC principles includes toxicity of reagents, consumption of energy, and generation of waste. The practical and economic efficiency of the method depends on cost and time efficiency, as well as on requirements and operational simplicity [14, 28].

3.1 Preliminary results

Parameters log D (descriptor of the lipophilicity of an analyte), and pKa (descriptor of the acidity of an analyte), obtained using MarvinSketch software, served for evaluation of the physicochemical characteristics of analytes, which was useful for defining the initial composition of the mobile phase. CAFF, with a pKa value of –0.92 and log D value of –0.55, is in molecular form at pH values above 3.0 while PAR, with a pKa value of 9.46 and log D value of 0.91, is in molecular form at pH values below 7.0. Hence, using pH values between 3.0 and 7.0 will ensure that both CAFF and PAR are 100% in their molecular forms. ACN was chosen as the organic phase, because greener alternatives (methanol, ethanol, and water) being strong protic solvents, did not provide satisfactory retention [29]. Ammonium acetate, ammonium formate, sodium citrate, and potassium phosphate buffers were tested as buffer systems. Sodium citrate and potassium phosphate buffers did not provide satisfactory results, as explained in detail by Knežević Ratković et al [30]. Ammonium acetate and ammonium formate gave no major differences which is why acetate salt was chosen for further research. Four columns were tested as stationary phases: amide, amino, diol, and silica, to identify the one that provided the optimal performance. The amide stationary phase did not provide satisfactory resolution (Rs) between peaks of CAFF and PAR. Additionally, the chromatographic peaks were visually unacceptable in terms of symmetry (asymmetry factor, As) and peak width (W). Moreover, amino and silica stationary phases did not provide adequate retention (retention factor, k). All the listed parameters are key and minimum requirements must be met when developing a method for quantitative analysis [8, 9]. The diol stationary phase ensured optimal resolution and peak symmetry, as well as optimal retention. Therefore, it was chosen for further analysis.

The surface of the diol column is neutral but hydrophilic due to the presence of 2,3-dihydroxypropyl ligands which enable hydrogen bonding (Fig. 1C) [30]. Amide and hydroxyl groups in PAR molecule (N-(4-hydroxyphenyl)acetamide) act as hydrogen bond donors whereas the carbonyl and hydroxyl groups act as hydrogen bond acceptors within the molecule (Fig. 1B). This enables PAR to form intra- and intermolecular hydrogen bonds. CAFF (1,3,7-trimethyl-3,7-dihydro-1H-purine-2,6-dione) is a polar molecule because of the electronegativity difference between the carbon-oxygen and carbon-nitrogen single polar covalent bonds (Fig. 1A). Both molecules can form dipole-dipole intermolecular interactions [31, 32]. Based on the log D values, it can be assumed that PAR, being less hydrophilic, will elute first, while CAFF, as a more hydrophilic molecule, will retain longer in the HILIC system. However, predicting retention based solely on log D values is not sufficiently accurate and shows increasing deviations when analytes can form specific interactions, i.e. dipole-dipole interactions [33]. Moreover, the retention of analytes in the column depends not only on the physicochemical characteristics of the analytes themselves but also on the properties of the stationary phase and the composition of the mobile phase. These factors, combined with the nearly eight times higher concentration of PAR in the sample, likely contributed to CAFF eluting first, followed by PAR [33].

3.2 Quality Risk Assessment using the Ishikawa diagram

QRA was used to identify CMPs whose modification would enable the attainment of an ATP, and it was conducted using an Ishikawa diagram (Fig. S2) [12, 34]. ATP was defined as a satisfactory retention factor (k) of the first eluted analyte (CAFF), satisfactory separation between CAFF and PAR (resolution factor, Rs), asymmetry factor of the critical peak (As of PAR), and optimal analysis duration time – retention time of the last eluted analyte (t_R of PAR). The selected factors are crucial for developing reliable methods. Adequate retention, satisfactory separation and asymmetry factor are mandatory, while minimizing retention time reduces waste production [35, 36]. Following these ATP requirements, corresponding CQAs were defined as follows: retention factor of CAFF ≥ 1, resolution between peaks of CAFF and PAR > 1.2, peak asymmetry of PAR between 0.8 and 1.5 (target 1.0), and optimal retention time of PAR (<10 min) [29, 35]. The identified CMPs included: ACN content, buffer concentration, pH of the aqueous phase, and column temperature. The injection volume, mobile phase flow rate, and detection wavelength, identified as non-critical, were maintained at a constant level during further research: 10 mm³, 1 cm³ min⁻¹, and 225 nm, respectively.

3.3 Optimization of the method and definition of the Design Space

The significance of linear, quadratic, and interaction terms was calculated using the ANOVA test (Table 1). The experimental results exhibit a strong correlation with the model-predicted values, as evidenced by the high coefficient of determination (R²) (≥0.94) and adjusted R² values (≥0.90), being well within acceptable limits of R² ≥ 0.8. This confirms that the quadratic model effectively represents the relationship between experimentally obtained and model-calculated parameters. The absence of a significant lack of fit values along with the slight differences between adjusted and predicted R² (≤0.11 for all CQAs) and high values of adequate precision (≥16) implies that this model can be used to manage the DS [10, 34].

After statistical (eq. 2 to 5) and graphical analysis (Fig. 2), the impacts of CMPs on CQAs were defined and the following can be concluded:

View Full Size

3D response surface graphs for (A) resolution between peaks of caffeine and paracetamol: Rs = f(acetonitrile content, column temperature), (B) peak asymmetry of paracetamol: As = f(acetonitrile content, pH value of aqueous phase), (C) peak asymmetry of paracetamol: As = f(acetonitrile content, column temperature), (D) peak asymmetry of paracetamol: As = f(pH value of aqueous phase, column temperature), (E) retention time of paracetamol: t_R = f(acetonitrile content, column temperature)

Citation: Acta Chromatographica 2025; 10.1556/1326.2025.01316

• Download Figure
• Download figure as PowerPoint slide

The retention factor of CAFF was solely influenced by the ACN content and column temperature, exhibiting an increased value in this factor with ACN content increasing and column temperature decreasing. The resolution between peaks of CAFF and PAR was affected by ACN content, column temperature, and their interaction, increasing with ACN content increasing and column temperature decreasing [37]. The peak asymmetry of PAR was influenced by: ACN content (increasing with ACN content decreasing), buffer concentration (increasing with buffer concentration decreasing), column temperature (increasing with column temperature increasing), as well as interactions between: ACN content and pH value of aqueous phase, ACN content and column temperature, and pH value of aqueous phase and column temperature. The diol column used in this study was not end-capped which resulted in the possibility of forming ionic interactions between free silanol groups and buffer ions. This interaction, in combination with pH value effects, leads to varying water layer thickness and, consequently, impacts the peak asymmetry of PAR [30, 38]. The retention time of PAR was impacted by ACN content, column temperature, and their interaction, prolonging with ACN content increasing and column temperature decreasing. In HILIC, the retention time and retention factor of analytes increase with ACN content increasing because, at high ACN contents, water is strongly adsorbed on the surface of the stationary phase. A water layer thick enough to create a liquid-liquid partition system enhances the retention of polar analytes [31, 38, 39]. Elevated temperatures enhance solute solubility and diffusivity, reduce solvent viscosity, and improve partitioning kinetics, leading to improved peak shapes, increased column efficiency, and shorter analyte retention. Higher temperatures usually decrease retention in HILIC by reducing cohesive energy differences and polar interactions between the mobile and stationary phases [33, 40].

2-D charts showing a relationship between CQAs and CMPs aided in identifying optimal values of CMPs, ensuring that all four CQAs met specified requirements (Fig. 3A). Employing MODDE 13 software, the sweet spot region – region where all criteria are met, was defined by overlapping the 2-D charts of four CQAs (Fig. S3). The expansive green area suggests flexibility in adjusting various combinations of CMPs while maintaining desired responses within defined limits. It is important to note that not all responses within the green area are necessarily robust. Therefore, the chart serves as a region that depicts the fulfillment of CQAs, necessitating the precise definition of the DS as a robust region [8]. The DS was established using Monte Carlo simulation and mathematical models obtained through MODDE 13 software with an acceptance level of 1% (Fig. 3B) and defined as follows:

• 93.08%–96.58% for ACN content,

• 0.0121−0.0233 M for buffer concentration,

• 5.05–6.80 for pH of aqueous phase, and

• 15.3 °C–27.8 °C for column temperature.

View Full Size

Optimization of the method. (A) Sweet spot chart and (B) Design space. The red star sign is the robust setpoint position [94.50% ACN and 5.5% buffer 0.017 M with a pH value of 5.5 and column temperature of 20 °C]. The percent sign and corresponding scale present the probability of failure

Citation: Acta Chromatographica 2025; 10.1556/1326.2025.01316

• Download Figure
• Download figure as PowerPoint slide

The robust setpoint was defined as: 94.50% ACN and 5.5% buffer 0.017 M with a pH value of 5.5 and column temperature of 20 °C, with the corresponding chromatogram presented in Fig. 4.

View Full Size

Chromatogram obtained under the robust setpoint conditions [94.50% ACN and 5.5% buffer 0.017 M with a pH value of 5.5 and column temperature of 20 °C]

Citation: Acta Chromatographica 2025; 10.1556/1326.2025.01316

• Download Figure
• Download figure as PowerPoint slide

3.4 Robustness and System Suitability Testing

The robustness of the method was assessed using a fractional factorial design^4-1 (FFD^4-1), which enables the examination of the impact of four CMPs on four defined CQAs [10]. CMPs examined and CQAs monitored were the same as in the optimization phase. The results showed no significant fluctuations in CQAs values with minor intentional changes in CMPs. Regarding the assessment of the robustness of the method, all examined CQAs had satisfactory values (Table 2), confirming that the method is robust.

The influence of CMPs on CQAs was evaluated by applying Dong's algorithm and the t-test (Table S2). The absolute values of CQAs lower than the corresponding value for E_critical also confirm the robustness of the method within the DS. SST parameters, defined based on FFD^4-1 results, show that conducting experiments, even under the most risky conditions, will yield satisfactory results (Table S3) [8].

3.5 Validation of the method

Validation of the method was performed following the ICH Q2 (R1) guideline [41]. The selectivity of the method was validated by comparing the chromatogram of the tested compounds to that of the placebo mixture (Fig. S4). The absence of interference at the chromatographic peaks of CAFF and PAR confirms the selectivity of the method. The linearity of the method was assessed by establishing linear correlations between the peak area and CAFF or PAR (r = 1 and 0.9999, respectively) concentrations within ranges of 5.0–100.0 and 50.0–250.0 μg cm⁻³, respectively (Table 3, Fig. S1). The accuracy of the method was confirmed by relative standard deviation (RSD) values lower than 2% and recovery values within the acceptable range of ± 5% (Table 3). The precision of the method was verified by satisfactory RSD values (<1%) obtained during the analysis of Panadol Extra® tablets (Table 3) [42]. LOD and LOQ values were determined for both CAFF and PAR, thus completing the validation of the method (Table 3).

3.6 Assessment of greenness, blueness and whiteness

The greenness of the method was first evaluated using the AGREE software. The only drawback is the use of toxic ACN, resulting in the lowest score (0.31) in field 11, marked with an orange color (Fig. 5A). Green color is the most acceptable result, however, going in transitions from green to yellow, orange to red, the result is more environmentally unacceptable. In Fig. 5A, it can be seen that majority of parameters are green, except previously mentioned field 11, but also fields 7 (corresponding to the volume of analytical waste), 9 (corresponding to the use of energy) and 10 (depicting the origin of the reagents) which are yellow. Therefore, the generation of less than 5 g of waste, using liquid chromatograpy and the use of reagents that are not obtained completely from renewable resources resulted in scores of 0.49, 0.50 and 0.50 in fields 7, 9 and 10, respectively. The values of the remaining fields varied from 0.60 to 1.00. The weights of critical fields (7, 10, 11, and 12) were set to the maximum value of 4 to ensure these effects were assessed as accurately and impartially as possible. Hence, with a complete green analysis, the method is environmentally acceptable – the final estimate for the green value and ecological acceptability of the method is 0.66 (the greener the method, the higher the score) and is placed on the light green center field, which visually confirms the greenness of this method. The greenness of the method was additionally evaluated using ComplexGAPI software. Considering the processes that precede the analysis, it is observed that the method remains satisfactorily green. The only critical step is the use of ACN, highlighted in red on the second pictogram, and no waste treatment, highlighted in red on the fourth pictogram (Fig. 5B). Yellow color on the first pictogram is attributed to storage conditions and on the third pictogram to health and safety hazard of reagents used for sample preparation. On the fourth pictogram, yellow color depicts occupational hazard and waste while on the fifth pictogram it refers to health and safety hazard, as well as, to amount of waste generated. White color on the second pictogram refers to absence of extraction. The circle visible at the center of the symbol indicates that the method is suitable for qualitative and quantitative analysis. Yellow color of this field is attributed to the procedures of sample preparation. Taking into account all the parameters evaluated, this method is considered environmentally acceptable since green is the dominant color on Fig. 5B [35].

View Full Size

Assessment of the environmental friendliness and effectiveness of the method. Evaluation of greenness using (A) AGREE and (B) ComplexGAPI software. Evaluation of blueness using (C) BAGI, and whiteness using (D) RGB model

Citation: Acta Chromatographica 2025; 10.1556/1326.2025.01316

• Download Figure
• Download figure as PowerPoint slide

In the second step, the blueness (reflecting the practicality) of the method was assessed using the BAGI software. The light blue color in the second and fourth pictograms is attributed to the number of analytes and simultaneous sample preparation, respectively. The blue color on the third and ninth pictograms depicts the analytical technique and automation degree, while the number in the center of the pictogram reveals the overall score of the method. A score of 85.0 indicates satisfactory method efficiency since the method considered practical needs to have at least a score of 60.0 (Fig. 5C) [17].

Finally, in the third step, the whiteness of method was evaluated by analyzing it through RGB model. The score of 83.3 indicates that the method is remarkable both in terms of greenness and practicality considering that the closer the score is to 100, the whiter the method (Fig. 5D). Additionally, the color of the score field is determined by the intensity of red, green, and blue. High, uniform saturation of these colors results in white. Thus, whiteness can quantitatively reflect the balance of three main pillars of sustainable methods and light grey color of this method confirms its sustainability [35].