2.1 General

[¹⁸F]fluoride produced via the ¹⁸O(p,n)¹⁸F nuclear reaction on a 16.5 MeV cyclotron PETtrace® (GE Healthcare Technologies, USA) was supplied by the Department of Nuclear Medicine, National Cancer Insitute (NCI), Putrajaya and delivered in a liquid form to the Centre of Diagnostic Nuclear Imaging (CDNI), Universiti Putra Malaysia (UPM). The no-carrier-added [¹⁸F]fluoride solution was then transferred to the Scintomics GRP 4 V module (Scintomics GmbH, Germany) to produce [¹⁸F]FET, [¹⁸F]FMISO and [¹⁸F]FLT.

2.2 Reagent and chemicals

Ethanol and acetonitrile with the purity of 99.9% and 99.5% respectively, were purchased from Merck (Darmstadt, Germany) and used to prepare analytical solutions. Reagent kits and sterile cassettes used in the preparation of [¹⁸F]FET, [¹⁸F]FMISO and [¹⁸F]FLT were purchased from ABX Advanced Biochemical Compounds (Radeberg, Germany). All precursors and eluents were stored following ABX Advanced Biochemical Compounds instruction.

2.3 Instrumentation

All experiments were performed on a Shimadzu (Japan) GC-2010Plus AF gas chromatography equipped with FID-2010Plus flame ionisation detector (FID) and integrated AOC-20i autoinjector. The system was controlled by LabSolution (Version 5.82). The chromatographic separation was performed on the Agilent J&W DB-200 column (mid-polarity), 30 m × 0.53 mm × 1 µm, with the sample volume of 1 µL. The ultrapure nitrogen gas, 99.9995% purity (Air Products, Malaysia) was used as the carrier gas and set at 30 mL min⁻¹. The FID airflow (Air Products, Malaysia) was set at 400 mL min⁻¹, and 40 mL min⁻¹ for hydrogen gas, 99.9992% purity (Air Products, Malaysia). The FID detector temperature was set at 250 °C.

Other chromatographic factors such as the injection temperature (220 °C) and column temperature (50 °C initial and hold for a minute, before linearly increase at 10 °C min⁻¹ to 90 °C) optimised in the previous study, were maintained. The carrier gas flow (mL min⁻¹) and the injection split ratio were determined using the multivariate optimisation techniques.

2.4 Optimisation of the experimental variables

The outcome of carrier gas flow and injection split ratio on the analysis time and resolution between adjacent peaks was assessed using a two-level full factorial design. The desirability function was used to enable these responses to be optimised simultaneously (Derringer & Suich, 1980; dos Santos Costa et al., 2019). Each response (y_i) was converted to an individual desirability function (d_i), which varies between 0 and 1 (0 ≤ d_i ≤ 1). The individual desirability (d_i) then was determined using the Eqs. (1) and (2) for the analysis time and resolution between adjacent peaks.

In order to support this finding of an optimum set of input variables (“experimental matrix”), a Factorial ANOVA (Two-Way ANOVA) was used to determine whether there is an interaction between the factors on the responses. If the interaction between the factors exists, which factor significantly contributes to the response, percentage variation of the response that can be predicted from the factors and which simple effects of experimental matrices that contribute statistically significant responses. The data analysis derived from the multivariate optimisation was performed using the software SPSS Statistics 26 (IBM, USA).

2.5 Validation of the proposed procedure

The procedure was validated following the ICH’s analytical procedure validation guideline (International Council for Harmonisation Of Technical Requirements for Pharmaceuticals for Human Use, 2006). The matrix effect, linearity, precision (repeatability and intermediate precision), the limit of detection (LOD) and quantification (LOQ), accuracy (recovery) and robustness parameters were evaluated.

2.6 Application of the procedure

Radiolabeling of ¹⁸F-derivative samples of [¹⁸F]FET, [¹⁸F]FMISO and [¹⁸F]FLT, three samples each, were performed in Scintomics GRP 4 V automated synthesis module (Scintomics, Germany) at CDNI, UPM. Samples were then analysed using the proposed procedure.

3.1 Optimisation of the experimental variables

The experiments were executed using a 0.9% (w/v) NaCl solution spiked with ethanol and acetonitrile concentrations equivalent to 1 mg mL⁻¹ for the simulation of a radiopharmaceutical matrix. The two-level full factorial design was performed in two-level: high (+) and low (−). Table 1 shows the coded and actual values of the selected variables at this level, the central points triplicates and the d_i values for the resolution and analysis time.

Carrier gas flow and injection split ratio were selected as factors as they potentially influence the analysis time and resolution between ethanol and acetonitrile peaks (dos Santos Costa et al., 2019). Analysis time and resolution of neighbouring peaks were chosen as outcomes in order to assess the suitability of the method in terms of rapid analysis time and separation efficiency. The factors were optimised using a two-level full factorial design which includes the central points triplicates for estimating the experimental error.

The statistical analysis of Factorial ANOVA showed a relatively high observed power, 0.98 and a significant interaction between the carrier gas flow and injection split ratio on the resolution of neighbouring peaks (p < 0.001). The Partial Eta squared revealed that about 97% of the variance in the resolution can be predicted from the injection split ratio. In comparison, only about 78% was predicted for the carrier gas flow. The finding supported the statement that injection split ratio only influenced resolution (dos Santos Costa et al., 2019).

As the interaction between these factors on the resolution was significant, further Games-Howell posthoc analysis was performed to determine the combination of experimental matrices that demonstrated significant difference in resolution. The Games-Howell posthoc analysis revealed that changing the injection split ratio from 1:10 to 1:50, produced a significant difference in the resolution. Nevertheless, the resolution only showed a statistically significant difference if the carrier gas flow was maintained at 3 mL min⁻¹.

Further statistical analysis of Factorial ANOVA on the analysis time also revealed a high observed power, 0.99 and a significant interaction between these factors (p < 0.001). The Partial Eta squared reported that 100% of the variance in the analysis time can be predicted from the carrier gas flow. It is unsurprising to find that only about 91% was predicted for the injection split ratio. These results are in line with our expectations as, other than column temperature, the carrier gas flow influences the analysis time and resolution (dos Santos Costa et al., 2019).

The Games-Howell posthoc analysis on the analysis time revealed that increasing the carrier gas flow to 4.0 mL min⁻¹ substantially leads to shorter analysis time. Nevertheless, it yields an imperfect separation of the adjacent peaks as compared to 3.0 mL min⁻¹ of carrier gas flow used. The results were exciting and help to justify that the use of a relatively moderate gas carrier flows at 3 mL min⁻¹ was adequate for an appropriate analysis time that <5 min with an excellence resolution (R > 5). Concerning the injection split ratio, using a lower injection split ratio of 1:10 will take more time for the sample to enter into the column. Additionally, an internal diameter column of 0.53 mm requires a minimum injection split ratio of 1:10. A higher injection split ratio of>1:50 will result in low sensitivity because less sample is injected into the column as well as wasting a carrier gas.

After the validation and optimisation of the two individual desirabilities (d_i) for analysis time and resolution, both functions could be combined into the overall desirability (D), which was computed using Eq. (3)

From the calculation, the optimum conditions for the ethanol and acetonitrile in the sample to be well resolved with a rapid analysis time (within 5 min, Fig. 1.) and highest D (0.51) were: 3.0 mL min⁻¹ of carrier gas flow and 50 injection split ratio. Under these conditions, the resolution values were in between 7.8 and 8.1, with a relative standard deviation (RSD) of<2%. The statistical analysis of Two-Way ANOVA performed earlier significantly strengthened the optimum conditions that were obtained using the overall desirability (D). In summary, the optimal experimental conditions used for validation in the later part was as follow (Table 2).

Close

Fig. 1

GC-FID chromatogram of 0.9% (w/v) NaCl solution spiked with ethanol and acetonitrile (1 mg mL⁻¹) under optimised chromatographic conditions.

Export to PPT

3.2 Validation of the proposed procedure

The proposed method was validated based on matrix effect, linearity, precision, the limit of detection (LOD) and limit of quantification (LOQ), accuracy and robustness following to international regulation (International Council for Harmonisation Of Technical Requirements for Pharmaceuticals for Human Use, 2006).

Following a statistical analysis of Student’s t-test, the matrix slopes and standard calibration curve of acetonitrile showed a significant result of the matrix effect (p < 0.001). It may be due to the interaction between the matrix and acetonitrile in the sample that produces a significant difference in response. For this reason, fortified samples were used to construct the calibration curve for acetonitrile.

In the linearity study, the Levene’s test was performed to support homoscedasticity (p > 0.05) before the ordinary least squares method (OLSM) was used to examine the slope, intercept, residual and correlation coefficient (r). The correlation between the standard solution at the six concentrations levels and the response of area under the curve for both solvent was significantly strong, r > 0.990. Through the logistic regression analysis (R²), about 99% of the variance in the response can be predicted from the concentration.

All calibration points were found within the prediction interval (Fig. 2A.), and R² present value of 0.999. Studentised residual plot (Fig. 2B.) also showed all the computed residuals within the ± 2 limits and appeared normally distributed and centred at zero. The same can be observed in the case of response factor (Fig. 2C.), whose values were included between 5%. In Fig. 2D., back-calculated relative errors (%RE) were represented against the concentration of standard solutions. Only a single point that corresponds to an ethanol concentration of 0.8 mg mL⁻¹ showed closer to the computed residuals limit of 2 and 15% of RE. From the author's point of view, this finding is expected since the very low concentration. Although the response factor for the respective concentration was slightly above 5% limit, it did not affect the linearity of the curve.

Close

Fig. 2

Linearity study for the calibration curve of ethanol by GC-FID in the calibration range, 0.8 to 7.5 mg mL⁻¹.

Export to PPT

In Fig. 3., all calibration points are found within the prediction interval (Fig. 3A.) and R² present value of 1.000 by using the concentration range from 0.1 to 1.0 mg mL⁻¹. Studentised residual plot (Fig. 3B.) also showed all the computed residuals within the ± 2 limits with the back-calculated %RE were lower than 5% (Fig. 3D). In the case of the response factor, it was obvious to notice that acetonitrile concentration at 0.1 mg mL⁻¹ was the problematic concentration level. The response factor was out of range for the respected concentration and considered as outlier. In this sense, a new calibration line was generated accordingly by removing the respective concentration points (Fig. 4.).

Close

Fig. 3

Linearity study for the calibration curve of acetonitrile by GC-FID in the calibration range, 0.1 to 1.0 mg mL⁻¹.

Export to PPT

Close

Fig. 4

Linearity study for the calibration curve of acetonitrile by GC-FID in the calibration range, 0.2 to 1.0 mg mL⁻¹.

Export to PPT

The calibration line for acetonitrile has been optimised by using the concentration range between 0.2 and 1.0 mg mL⁻¹. The logistic regression analysis, R² was maintained but most remarkably improved was the response factor and the back-calculated %RE (Fig. 4C. and 4D.). The back-calculated %RE was lower than 1% (Fig. 4D.).

The estimated LOD and LOQ were subsequently verified by the independent analysis of solutions prepared solutions at these ranges (n = 6). The LOD and LOQ were higher for ethanol when compared to acetonitrile as the concentration range of the calibration curve were not similar (Mihon et al., 2017). The LOD was observed but not generally quantified as an accurate value, while LOQ demonstrated excellent precision (RSD below 5%) and accuracy (recovery from 84.8 to 104.6%) for both solvents.

The assessment of precision for repeatability and intermediate precision was carried out by the RSD % acquired from the recovery analysis (<2%). A statistical analysis of Student’s T-test was carried out and confirmed that there is no statistical difference between the result (p > 0.05).

In the evaluation of accuracy, the recovery values ranging between 96 and 119% and 95 to 99% of the spiked value for ethanol and acetonitrile respectively, were in agreement with the literature (dos Santos Costa et al., 2019). The recovery value was slightly higher for ethanol at the concentration of 6.0 mg mL⁻¹ probably due to an interaction of analytes in the spiked that causing higher response obtained. Table 3 below summarised the linearity study, regression analysis, LOD and LOQ, assessment of precision and recovery.

Another critical element for validating the proposed method is the method’s robustness. Robustness is best described as the ability of the analytical method to remain unchanged, even in the presence of minor variations which have been intentionally implemented in the experimental conditions (Bosque-Sendra et al., 1999; Ferreira et al., 2017). The two-level full factorial was selected to evaluate the robustness of the proposed method as it is possible to assess the significance of all main effects and their interactions without confusing them. Besides, it is the most efficient chemometric tool for assessing robustness when the number of factors is not high (Ferreira et al., 2017).

In the evaluation of the robustness of the proposed method, the experimental matrix was developed based on altering factors related to the optimised conditions, taking into account the retention time of ethanol and acetonitrile as a response. The carrier gas flow was adjusted from 3.0 mL min⁻¹ to 3.2 mL min⁻¹, and the injection split ratio was modified from 50 to 30. The experiments were conducted randomly, and the findings were presented in Table 4. All the standardised effect values attained were below 4.303, the critical value of Student’s t-test for the degree of freedom of 2 at a confidence level of 95%. The proposed method is, therefore, robust for the experimental matrix set out in the analysis of robustness.

The calculated standardised effects for ethanol was found higher than the acetonitrile due to variations in the retention time of ethanol when changes were made to the proposed experimental method. Nevertheless, the RSD % for the precision was still <2%.

3.3 Application of the procedure

Following the production of [¹⁸F]FET, [¹⁸F]FMISO and [¹⁸F]FLT samples, ethanol and acetonitrile in the samples were evaluated using the developed method. Ethanol and acetonitrile were well resolved and determined within 3.5 min, with a total analysis time of<5 min, including sample injection and separation. Table 5 displays the concentration values for the radiopharmaceuticals, and all tested samples were within the acceptance range (ethanol < 5.0 mg mL⁻¹; acetonitrile < 0.41 mg mL⁻¹).

The separation between the two peaks was excellent. The total analysis time of<5 min was acceptable, as the procedures described in the literature required a total separation time of >5 min (Mihon et al., 2017). Even though the carrier gas flow and column temperature influence the analysis time, the size of the column’s internal diameter also plays a significant role from our point of view.

According to the author’s perspective, in any proposed method, the resolution between the adjacent peaks is also imperative as is the total analysis time. Excellent resolution, preferably R>5 is needed especially when the GC column begins to age, which may cause the peak to appear broad and potentially increase the tailing factor, 1 < T < 2. These results provide evidence that with a 0.53 mm of an internal diameter of the column, the resolution of adjacent peaks was excellent and comparable to the column size of 0.25 mm and 0.32 mm (Channing et al., 2001; dos Santos Costa et al., 2019; Kilian et al., 2014; Nader et al., 2011).

The analysis time also was within 3.5 min and shorter than the previous study when similar column sizes were used (Mihon et al., 2017). Nevertheless, the study did not specify the carrier gas flow. The lower injection split ratio of 1:15 used in the study was suspected to contribute to the longer analysis time as a lower injection split ratio takes longer time for the sample to enter the column (Mihon et al., 2017). The shorter analysis time observed in others was unsurprising given the smaller internal diameter of the column, 0.25 mm and 0.32 mm, which was used. The tabulated data in Table 6 below showed a 0.53 mm of the internal diameter of the column might require a longer time for the total separation