A facile lyophilisation-based sample preparation approach for the determination of selected wastewater-borne antiretroviral drugs and metabolites by SFC-MS/MS

2.1 Reagents and chemicals

HPLC-grade chemicals, reagents and standards were used in all experiments. Analytical standards (i.e., >98 % purity) for the target compounds (Fig. 1) were procured either from ClearSynth™ (Mumbai, India) as it is the case with efavirenz, emtricitabine, lamivudine, nevirapine, ritonavir, zidovudine, or Toronto Research Chemicals (Toronto, Canada) for 8,14-dihydroxy efavirenz, deuterated nevirapine, desthiazolylmethyloxycarbonyl ritonavir and zidovudine glucuronide (Mosekiemang, 2021; Mosekiemang et al., 2019; Venter and Onselen, 2023). Acetonitrile (ACN), methanol (MeOH), ammonium hydroxide (NH₄OH) and formic acid (HCOOH) were all procured from Romil Ltd. (Waterbeach, Cambridge, GB).

2.2 Instrumentation and stationary phase evaluation

A Waters Acquity UPC² SFC system (Acquity Ultraperformance Convergence Chromatography™) was used in this study. It comprised of a binary solvent delivery pump, temperature-regulated column compartment kept at 60 °C, sample manager module and an automatic back pressure regulator (ABPR) set at 1700 psi. The mobile phases (CO₂ and MeOH) were set to be delivered at a flow rate of 1.30 mL min⁻¹. A 5 µL sample volume was injected into the column housed in a temperature-controlled (60 °C) column compartment. Several Acquity™ UPC² stationary phases (HSS C18 SB, Torus-1 AA, Torus-2 PIC, 2-EP and BEH) were evaluated for their suitability in the separation of target analytes (Fig. 1). All columns (100 mm × 3.0 mm, 1.7 µm d_p) were made of sub-2 µm fully porous silica particles except for the HSS C18 SB (100 mm × 3.0 mm, 1.8 µm d_p). The best results were obtained out of the BEH column and thereafter used for the rest of the study. The mobile phases’ gradient was set as follows: 98 % CO₂ (0–0.2 min), 98–77.7 % CO₂ (0.2–5 min), 77.7–40 % CO₂ (5–7 min), 40 % CO₂ (7–7.2 min), 40– 98 % CO₂ (7.20–9 min), and 98 % CO₂ until re-equilibration.

2.3 Mass spectrometric operating conditions

A Xevo TQ-S QqQ MS (Waters®) was hyphenated to an UPC² system equipped with an ancillary pump (Waters 515™) designed to deliver an additional modifier (0.1 % HCOOH in MeOH at 0.3 mL min⁻¹) to the mobile phase prior to the ESI in a configuration known as ‘pre-BPR splitter with sheath pump interface’ (Akbal and Hopfgartner, 2020; Gros et al., 2021; Mosekiemang, 2021; Perrenoud et al., 2014). The temperatures for the ion source and desolvation gas were set at 150 °C and 350 °C, respectively. The N₂ cone gas- and desolvation gas flow rates were set at 150 and 900 L h⁻¹, respectively. This included the capillary- and cone voltages which were set at 3.8 kV and 20 V, respectively.

The MS detector was operated in the MRM scanning mode to monitor two transition ion pairs (i.e., precursor- and two fragment ions) and/or two identification points per cycle time (t_cycle). The MRMs were optimized using one ionization mode at a time (i.e., ESI^– followed by ESI⁺ and vice versa). In the final method, all analytes were ionised in the positive mode (ESI⁺). Briefly, individual standard (52 ng/mL) for each analyte was infused post-column directly into the ESI at the following series of cone voltages (CV): 15, 20, 25 and 30 V to determine the best possible CV that produced the most intense precursor ion, and the obtained conditions were transferred to set-up the daughter ion scan mode. Then in this mode, the precursor ions were subjected to a series of collision energies (CE) in the range 15, 20, 25 and 30 eV to determine the optimum CE yielding the most intense fragment ions. Other parameters such as the dwell- (t_dwell) and cycle times (t_cycle) were software-set (MassLynx™ version 4.1) at ∼ 10 data points/peak. These were deemed adequate points/peak not to require implementation of a peak smoothing algorithm during peak integration considering the importance of peak purity and/or peak symmetry in quantitative chromatography. In that way all quantifier (Q) ions and their corresponding qualifier (q) ions were successfully assigned. Peak identity was confirmed if the ratio of peak intensities (q/Q or Q/q) was constant over the calibration range and independent from matrix and vice-versa. Similarly, the retention time shifts (Δt_R) between the Q- and q ions was monitored and expected to be negligible and independent of concentration and matrix effects (Kruve et al., 2015a, 2015b). Optimised MRM experimental conditions are summarized in Table S1 ‘supplementary information (SI)’ section. Even under these conditions, AZTG and EFVM could not be detected possibly due to the acidified ion source chamber owing to the flow injection of 0.1 % HCOOH in MeOH modifier solvent.

2.4 Preparation of calibration standard solutions

Calibration standards in the range of 0.610– 312 ng mL⁻¹ were prepared by an appropriate dilution of a 20-µg mL⁻¹ cocktail standard with neat MeOH. That is, from each concentration level, a 2000 µL aliquot of the cocktail standard was drawn to which an additional 900 µL of ILIS (100 ng mL⁻¹) was added and diluted to volume (3 mL) with MeOH. In that way, a ten (10)-point equidistant calibration level system (0.407–208 ng mL⁻¹), with each calibration point containing a uniform concentration of ILIS (100 ng mL⁻¹ NVP-D₃) was produced. This resulted in a constant volume fraction of a cocktail standard to a diluent throughout the experiment.

2.5 Sample collection and preparation

Wastewater samples were collected as previously described (Mosekiemang, 2021; Mosekiemang et al., 2019). Samples were vacuum filtered using a Whatman® filter paper (150 mm diameter and 1 µm pore size) and subsequently the sample pH was adjusted to ∼ 7 by a dropwise addition of 5 % NH₄OH or 5 % HCOOH for increment or reduction of pH, accordingly. Prior to lyophilization- and SPE processing, method blanks were prepared by addition of a 300 μL ILIS (NVP-D₃) and diluting to 50 mL with pH-adjusted wastewater matrix. The spike and recovery samples (i.e., fortified samples) were prepared in the same way except for an additional 75 µL spike of cocktail standards at appropriate concentrations to give the final spiking concentrations at low- (0.03 ng mL⁻¹), mid- (0.3 ng L^-1) and high (3 ng mL⁻¹) levels prior to processing by SPE or lyophilization. Samples were split into two batches for processing by each method as shown in Fig. 2.

Close

Fig. 2

An illustrative experimental design for the comparison of lyophilization and SPE. Preparation process for method blanks and fortified samples intended for the spike and recovery experiments is also shown.

Export to PPT

2.6 Method validation

The method fitness for purpose was assessed in terms of the traditional validation approaches to ensure accuracy and reliability of the quantitative results. At the start, a response function was established, this is the relationship between the instrument response (peak area/intensity) as a function of concentration. Then, the calibration data was tested for statistical normality and the best regression model for data-fitting was determined. This task entailed performing several statistical tests for linearity such as inspection of the spread of residuals to establish (hetero)homoscedasticity, test for outliers using the Cochran’s test and the analysis of variance lack-of-fit (ANOVA_LOF) (Kruve et al., 2015b, 2015a; Van Loco et al., 2002).

For most analytes, linearity range was up to 208 ng mL⁻¹ while NVP and its metabolite (NVPM) displayed narrow/reduced linear ranges of up to 104 ng mL⁻¹, possibly due to over-saturation of the detector. Other parameters such as the detection- and quantification limits (LODs and LOQs) were determined from the calibration data according to Equations S1–S4, (SI), while recovery and matrix effects were determined according to Equations S5–S5. Errors (%RSD) related to repeatability and reproducibility were determined from the spike and recovery data (Evard et al., 2016; Hewavitharana et al., 2018; Kruve et al., 2015a, 2015b).

3.1 Method performance and column selection

A Water Acquity UHPLC system coupled to a Xevo TQ-S MS system was previously fine-tuned for the detection of the same target analytes (Mosekiemang, 2021; Mosekiemang et al., 2019). In this study, re-optimisation was performed, this time the same MS detector was coupled to a Waters Acquity UPC² system using CO₂ and MeOH as mobile phases to which an additional make-up flow of 1 % HCOOH in MeOH modifier was added in a post column fashion (Guillarme et al., 2018; Perrenoud et al., 2014; Tarafder, 2018). The optimised experimental conditions such as the transition ions (MRMs), Q/q ion ratios, collision energies etc., are shown in Table S1, (SI). AZTG and EFVM, previously detected in the negative ionisation mode (Mosekiemang et al., 2019), were not detected this time around, possibly due to the infusion of acidified organic modifier to the ESI chamber. The non-detection of these compounds under acidified ionisation chamber conditions could be ascribed to the ‘wrong-way-around’ phenomenon as recently observed for the same compounds (Venter and Onselen, 2023). Briefly, the ‘wrong-way-round’ ionisation implies that contrary to expectations, protonated ions can be produced under basic conditions and vice versa where deprotonated ions are formed under acidic conditions. This often leads to the formation of weak precursor ions characterised by low peak intensities that often falls below the threshold detection limits leading to non-detection (i.e., false negative detection).

Column screening was performed to identify the SFC-phase that provided sufficient chromatographic separation for the target analytes, considering that SFC inherently produces poor peak shapes, and that peak symmetry is a crucial attribute in quantitative chromatography (Grand-Guillaume Perrenoud et al., 2012a; Wahab et al., 2017). Five SFC column phases were evaluated using peak asymmetry factors (As) as a criterion for selection (Ashraf-Khorassani and Taylor, 2010; Desfontaine et al., 2016; Grand-Guillaume Perrenoud et al., 2012a; Wahab et al., 2017). Interpretation of results for these experiments considered the extensive findings of previous work on SFC column classification especially by computational models such as linear solvation energy relationships (LSER) and other relevant data related to ARVDs on the same stationary phases (Khater et al., 2013; C West and Lesellier, 2006a; C. West and Lesellier, 2006).

The elution order of ARVDs on all columns was in reverse to what was previously observed on reversed phase (Desfontaine et al., 2016; Mosekiemang et al., 2019) and generally resembled that of normal phase, where nonpolar compounds (EFV, NVP, log K_O/W≥2) showed markedly lower retention compared to their polar counterparts. Early eluting compounds, particularly EFV, NVP and NVPM displayed gaussian peaks compared to late eluting compounds (i.e., FTC, RTVM and 3TC) on all columns. Surprisingly, the RTVM peak displayed a tailing phenomenon which was worse off on the BEH 2EP and Torus 2 PIC but phenomenally improved on the Torus 1AA, BEH and HSS C18 SB. In contrast, RTV (i.e., a structural analogue of RTVM) was sufficiently symmetrical on all columns. At this stage, the reason for these variations in peak asymmetry factors for these structurally similar compounds is unexplainable but could be partially ascribed to an interplay of silanol and π-π interactions (Ashraf-Khorassani and Taylor, 2010; Desfontaine et al., 2016; Grand-Guillaume Perrenoud et al., 2012a). AZT was characterized by noisy peaks on all columns except for the BEH.

Also noteworthy is the fact that it took significantly high proportions of the organic mobile phase to elute the highly retained compounds, especially for RTVM. But, under these conditions (i.e., 60 % MeOH), the gaseous CO₂ is no more supercritical but subcritical, and the chromatography can be more accurately termed as ‘enhanced fluidity’, and better-still with extended scope of applicability to separate even the extremely polar compounds (Bennett et al., 2019; Cui and Olesik, 1991).

The column chemistries assessed in this study comprised of a polar unbonded (BEH), polar bonded (BEH-2EP, 1-AA, 2-PIC) and a nonpolar straight chain non-end capped phase (HSS C18 SB) with some polar character (Akbal and Hopfgartner, 2020; Farrell, 2017; Grand-Guillaume Perrenoud et al., 2014; Nováková et al., 2014). The key thing to note here is that the pH of the CO₂/MeOH mobile phase is predominantly acidic at (sub-)supercritical conditions (i.e., pH∼4) (Desfontaine et al., 2016) and at this pH the compounds are expected to be protonated because of their high pKa’s (∼10–14). For instance, 3TC, FTC, NVP, NVPM, RTV and RTVM are predominately protonated under acidic conditions. Therefore, similar elution orders were observed on the 2-EP, 1-AA, 2-PIC (Fig. 3). For these phases, analyte-stationery phase-interaction proceeds by a way of hydrogen bonding and dipole–dipole interactions while the effects of secondary ionic interactions are greatly suppressed by the electrostatic forces resulting from the protonation of basic bonded ligands (Desfontaine et al., 2016). This could possibly be the reason for the gaussian peak shapes observed on the 2-EP, 1-AA and 2-PIC especially for the early eluting analytes (e.g., EFV, NVP, AZT and NVPM). Amongst these phases, late eluting analytes (e.g., FTC, RTVM and 3TC) were to a greater extent highly retained on the 1-AA, a phenomenon ascribed to the reduced electrostatic repulsion forces due to the high pKa of 1-aminianthracene and possibly the contribution of additional analyte-stationery phase interactions (i.e., π-π interactions) (Grand-Guillaume Perrenoud et al., 2014; Wahab et al., 2017).

Close

Fig. 3

Typical total ion chromatograms (TICs) of target analytes separated on the five (5) stationary phases evaluated. Each TIC indicates the column name and the corresponding ligand chemistry. Target analytes are arranged according to elution order 1–9, corresponding to EFV, NVP, NVP-D3, AZT, NVPM, RTV, FTC, RTVM and 3TC, respectively. Compounds abbreviations were already described in Fig. 1.

Export to PPT

The BEH phase provided sufficiently symmetrical peaks, a phenomenon possibly attributed to the uniform distribution of silanol functional groups and that the ethylene bridge contained the BEH silica material is known to provide good peaks compared to erstwhile silica raw material (C West and Lesellier, 2006b). Also noteworthy is that AZT produced a symmetrical peak only on this column compared to the noisy peaks observed on other columns (Fig. 3).

The HSS C18 SB phase also provided overall symmetrical peaks for most compounds except for slight peak-tailing tendencies observed for the late eluting compounds (FTC and 3TC) and peak-fronting with slight shouldering for RTVM. The retention mechanisms for this phase are predominantly governed by the hydrophobic interactions of the octadecyl functionality (C₁₈ straight chain backbone). Therefore, the observed tailing observed for the late eluting compounds may be an indication of the presence of acidic residual silanols which may have facilitated additional interaction with basic analytes (Ashraf-Khorassani and Taylor, 2010; Hirose et al., 2019; C West and Lesellier, 2006b).

The data discussed above are conclusive regarding the elution order observed for the test analytes with late eluting analytes (FTC, 3TC and RTVM) suffering slight tailing effects due to the onset of secondary ionic interactions. In contrast, early- and mid-eluting peaks were generally symmetrical, but AZT displayed moderately noisy peaks on all columns except for the BEH which was eventually selected for the present study. Comparatively, better peak shapes were obtained in our previous study using RP-LC (Mosekiemang et al., 2019) and also by Venter and van Onselen (Venter and Onselen, 2023), although not unexpected considering that SFC inherently suffers poor peak shapes which may be improved by addition of suitable additives including small amount of water (Grand-Guillaume Perrenoud et al., 2014, 2012a, 2012b; Wahab et al., 2017).

Considering the above, the best column for the present study was considered the BEH column because all compounds were sufficiently separated with good peak symmetries. Partial co-elution was observed for NVP/AZT and RTV/FTC, but these pairs were subsequently separated based on their unique m/z ratios of the transition ions. This may be true because as stated earlier, the MRM detection mode was optimised for the best sensitivities for all analytes, which in a way demonstrated the excellent selectivity offered by the MRM mode to detect and distinguish between closely eluting analytes (Chen et al., 2016; Hermes et al., 2018). As an example, the retention time-aligned peaks for the respective Q- and q transition ions of AZT/NVP peaks are shown in Fig. 4. As can be seen, there is a t_R-difference between the Q- and q ions for the partially coeluting peaks (i.e., AZT, t_R 3.97 min. eluted earlier than NVP, t_R 3.98 min.).

Close

Fig. 4

A typical total ion chromatograph (TIC) obtained using the Acquity UPC² BEH column for the separation of target analytes (A). Peak annotations 1–9 represents elution orders for target analytes as already described in Fig. 3. Peaks 2 and 4 are partially co-eluting but successfully separated by the differences in their t_R. Integrated peaks show that AZT eluted at t_R 3.97 min (B) while NVP eluted at t_R 3.98 min (C). The peak denoted by an asterisk is possibly baseline noise or an unknown entity.

Export to PPT

3.2 Evaluation of lyophilisation and SPE

Previous studies reported poor SPE-recoveries for the polar NRTI ARVDs such as abacavir, indinavir and 3TC (Aminot et al., 2015), which was also confirmed by our previous study for other wastewater-borne NRTI compounds (i.e., AZTG, 3TC, AZT and FTC) (Mosekiemang et al., 2019). This prompted the present work to probe lyophilisation process for analyte extraction and enrichment, as this was considered a solventless alternative procedure to RP-SPE and most importantly to prevent analyte losses associated with poor recoveries for polar analytes (Hu et al., 2014; Narayan et al., 2011; Ramirez et al., 2014; Zhang et al., 2021). To achieve this, an ILIS was added before the analyte extraction process to make up for sample handling variations and both methods were assessed in terms of the ILIS- and non-ILIS-corrected recoveries (Hewavitharana et al., 2018). Recovery data was evaluated at low- (0.03 ng mL⁻¹), mid- (0.3 ng mL⁻¹) and high spiking levels (3 ng mL⁻¹). Relevant data are presented in Table S4–S6, (SI).

The relative recoveries (ILIS-corrected) were adequate and in the range 65–99 % for effluent samples and 68–100 % for influent samples processed by lyophilisation across the spiking levels. Surprisingly, RTVM could not be quantified in all fortified samples due losses ascribed to irreversible adsorption to the system’s tubing or vacuum-induced losses during lyophilization (de Voogt et al., 2000). Regarding SPE, recoveries for the influent and effluent samples were only good for the apolar analytes of the nNRTI and PI pharmaceutical classes (EFV, NVP, NVPM, RTV and RTVM) and poor for the polar NRTI class (FTC and 3TC). This observation is in line with findings of previous studies (Aminot et al., 2015; Mosekiemang et al., 2019). Generally, the obtained results were conclusive that lyophilisation offered improved recoveries for all pharmaceutical classes compared to SPE which suffered losses for the polar analytes particularly for FTC and 3TC. This is a clear indication of the suitability of this form of sample preparation to process analytes with diverse physico-chemical properties.

For completeness, non-ILIS-corrected recoveries were also considered. The results obtained in this way are informative in highlighting the true analyte extraction efficiency of a sample preparation method (i.e., not compensated for analyte losses due to sample handling variations). A few notable differences were observed between non-ILIS- and ILIS-corrected recoveries implying that minimal analyte losses may have been incurred during sample preparation except for the known poor recoveries of the polar analytes due to non-retention on SPE. Problems associated with analyte losses due to sample handling can be easily overcome by incorporation of a suitable ILIS, but this may not always work as expected due to ILIS-analyte structural dissimilarities (Boix et al., 2016; Ibáñez et al., 2021; Simarro-gimeno et al., 2023). For instance, the results clearly show poor ILIS- and non-ILIS-corrected recoveries for 3TC and FTC by SPE in all samples considered. The recoveries in the range 23–42 % and 25–41 % were obtained for FTC and 3TC, respectively in ILIS-corrected effluent samples while almost similar results in the range 22–40 % as well as 18–31 % were obtained for the respective compounds in non-ILIS effluent samples. The same trend was manifested in influent samples where recoveries for the ILIS-corrected samples were measured in the range 23–66 % and 20–52 % in contrast to slightly worse recoveries in the range 13–34 % and 11–22 % for FTC and 3TC, respectively in non-ILIS-corrected influent samples. The poor recoveries particularly for the ILIS-corrected samples could be ascribed to the ILIS-analyte structural mismatches that may have resulted in the partial compensation for sample handling- and ionization-induced analyte losses. Table 1 provides a detailed summary for these data. To this end, lyophilisation can be viewed as an improvement to (large volume)direct injection (LVDI) (Backe and Field, 2012; Mosekiemang et al., 2019; Simarro-gimeno et al., 2023) in the context of improving instrument sensitivity (i.e., high LODs), considering that unlike LVDI it allows for preconcentration of large sample volumes and thus a substantial EF (50 × ) needed to considerably lower achievable method detection limits.

3.3 Method validation and performance

The first parameter of paramount importance in quantitative studies is the determination of accuracy for the proposed method and this is determined in several ways that includes identification of the best regression model that fits the calibration data, identification, and elimination of outliers in the data, inspection of residuals and calculation of the coefficient of determination (R²) for the regression plots. All these data are presented in Table 2. As can be seen, all plots turned up R² values above 0.9996 with significantly low residual errors (<10 %) in all the calibration plots. Follow up Cochran’s tests for outliers for the individual plots were also in agreement with the assumption of homoscedasticity in all the plots of residuals. Homoscedasticity is confirmed when the experimental Cochran’s C value (C_Calc.) is less than the critical Cochran’s value (C_Tab.) (Van Loco et al., 2002). In addition, the ANOVA_LOF test confirmed the ordinary least square regression model as the best model to fit the obtained calibration data, in turn confirming the assumption of linearity over the calibration ranges for the individual calibration plots. A worked-out example for these chemometric tests is illustrated in Table S3. For all analytes linearity was confirmed for up to 208 ng mL^-1calibration level, except for NVP and NVPM which displayed reduced linearities for up to 104 ng mL⁻¹ calibration level due to peak truncation and/or detector saturation beyond 104 ng mL⁻¹.

The method detection limits (MDLs) were calculated from the LOD data considering the EF (50 × ). As such, all MDLs were measured from 30 to 37 ng L^-1 except for 113 ng L^-1obtained for AZT. The method quantification limits (MQLs) were determined the same way using the LOQ data and taking into account the EF (50 × ). The MQLs ranged from 103 to132 ng L^-1, a typical range at which in-sewer compounds are normally detected. The method precision was measured in terms of intra- (repeatability) and inter-day precision (reproducibility). As expected, the method proved to be extremely repeatable since all repeatability tests yielded %RSD values less than 9 %. In contrast, reproducibility values were not that excellent as evidenced by %RSD values of ∼ 15 %. Nonetheless, all the precision data were within acceptable levels.

Matrix effects (MEs). Despite the complementary assembly and/or hyphenation of SFC to ESI, the configuration may be susceptible to matrix effect-like challenges due to analyte precipitation resulting from the gaseous CO₂ decompression at the interface junction where SFC connects to the ion source (i.e., ESI) (Akbal and Hopfgartner, 2020; Perrenoud et al., 2014). In addition, interferents displaying similar chromatographic behaviour as target analytes are also notorious for introducing matrix effects such as the ion source-induced ion enhancement or suppression. Given the importance matrix effects in quantitative chromatography (Stahnke et al., 2012; Svan et al., 2018), Equation S6 (SI) was employed to evaluate the extent of ion suppression or enhancement for the present method. The results obtained herein (Table 3) show that ion suppression was largely encountered in influent samples regardless of the sample preparation method and in late eluting compounds as evidenced by the strong retention observed for FTC (t_R 5.87 min.) and 3TC (t_R 6.55 min.). The ion suppression phenomenon observed here could be attributed to several factors, the main one being that raw wastewater is naturally dirtier than treated wastewater and may contain more interferents. In addition, matrix effects seemed to have been t_R-dependent and varied at the different chromatographic regions as evidenced by ion suppression especially for the late eluting compounds. The ion suppression was worse in lyophilised samples compared to SPE-processed. This may be a true observation considering that the mobile phase gradient went up to 60 % MeOH at the 7–7.2 min. region to facilitate elution of highly retained compounds and to flush out dirt from the column. Therefore, it is logical to suspect severe matrix effects for compounds eluting at this chromatographic region (See Section 2.1 for mobile phase gradient description). Also, noteworthy is that RTVM could not be detected and quantified in all fortified samples processed by lyophilisation possibly due to losses related to volatility during lyophilization (de Voogt et al., 2000). Apart from the ion suppression effects observed for FTC, 3TC and the non-detection of RTVM, lyophilization technique has proved to be a suitable and promising sample preparation method and compares adequately with SPE.

3.4 Quantitative data: SFC-MS/MS versus UHPLC-MS/MS

The quantitative data obtained by lyophilization-SFC-MS/MS method was compared to that of direct injection-UHPLC-MS/MS as described in our previous study (Mosekiemang et al., 2019). As already discussed, AZT was not detected in the present study while RTVM was only detected in SPE-processed samples and could not be detected in lyophilized samples, possibly to due losses associated with lyophilization (de Voogt et al., 2000) or severe ion suppression as already hypothesized elsewhere. Despite these disparities, there was an excellent agreement in the overall quantitative data obtained by both methods, as it is established by the statistical mean differences test of < 3.5 % for both methods and also verified by an insignificant paired t-test (p > 0.05). All data pertaining to such comparison are presented in Table 4.

3.5 Comparative method performance between SFC and UHPLC

On evaluation of the present method and in comparison to the preciously published one based on direct injection-UHPLC-MS/MS for the same suite of analytes, several accomplishments and shortcomings were noted. Regarding achievements, LODs, LOQs, and recoveries were markedly improved especially for FTC and 3TC in lyophilized samples (Backe and Field, 2012; Hermes et al., 2018; Mosekiemang et al., 2019; Richardson et al., 2021; Simarro-gimeno et al., 2023). This was not unexpected because the calculations for MDLs, MQLs and recoveries took into consideration the EF (50 × ) according to Eqs. S1–6. Conversely, the high MDLs and MQLs observed for the direct injection UHPLC method was because of the small sample volume used.

RTVM could not be satisfactorily validated by UHPLC due to poor reproducibility. In contrast, the same compound was accurately validated using SFC-MS/MS but could not be detected in real samples possibly because it was absent in the considered samples. Similarly, it could not be detected in fortified lyophilized samples due to losses associated with lyophilization but was successfully quantified in the same samples processed by SPE.