3.1 Matrix Effect and Interference
One of the crucial parts of this study was validating the theoretical design according to the multiple discussed in our previous study22. Since this study dealt with a complex matrix like serum, additional studies had to be performed to assess the matrix effect and its impact on ion suppression and the process efficiency. The matrix study was conducted according to the process published by Matuszewski et al.23, wherein, a set of neat solutions (5 levels) were prepared (using Mobile Phase B, i.e., acetonitrile) and an identical set of solutions was prepared using DDC Mass Spect Gold Serum. The concentration range for this particular study covered 20-1200 ng/mL range. Each level was plated in sextuplets (n = 6 per matrix per analyte) prior to LC-MS/MS analysis. The mean matrix effect range were such that: BUP = 62–101%, CIT = 85–118%, DES = 82–106%, IMI = 71–110%, MLN = 73–109%, OLN = 46–119%, SRT = 71–89%, and VIL = 79–107%. While the matrix effect seemed to be significant based on the upper and lower boundaries of the ranges, the data obtained is comparable to a previous study by Marchet et al.24. Additionally, the concentration range for this study expanded beyond the dynamic range of the standard curve using a linear regression with \(\frac{1}{x}\) weighting, plotting the analyte-to-IS ratios. Hence, the solutions were diluted 10-folds (lower level) and 6-folds (upper level) to generate a standard curve with a linear dynamic range. This resulted in an improvement in the matrix effect as well as recovery efficiency (data not reported). The final mean matrix effect is reported to be (82–105) ± 20% across all analytes, when measured by peak area. Any value above 100% was interpreted as ion enhancement, whereas anything below indicated ion suppression. Since the analytes do not have certified reference materials that could be held as a standard, it was not possible to perform a value assignment assessment for each analyte due to the matrix effect. However, to rule out any interference from the native serum, we ran triplicates of blank serum on each plate and observed no peaks at the detected retention times for all the analytes. This was important to establish that the LOQ values, where a peak was observed for each analyte at their determined Retention Times (RT). Figure 2 illustrates the detected RT for the eight drugs being investigated overlaid on their relative intensity profile. To our greatest surprise, we observed near to no signal for the Eureka antidepressant kit when we ran their calibrators side by side according to the manufacturer protocol (Supplemental Figure S2). This certainly raises concern about the quality of the currently available commercial kits and their application in the clinical diagnostic realm.
3.2 Solvent Analysis: Mobile Phase, DWS, Sample Preparation
A primitive study for a successful chromatographic analysis is to determine an organic solvent that would consistently yield a good signal and be able to elute the desired analytes/fragments with the highest efficiency. Even though methanol was identified as the predominant organic solvent for the antidepressants in discussion, it was still necessary to determine whether methanol would be an optimized and operational mobile phase B. Additionally, it is important to establish the lowest percentage of organic solvent at the final step of sample preparation that would yield an ideal peak shape, i.e., a Gaussian profile with minimal peak broadening22. As a result, we conducted a two-parameter study using methanol and acetonitrile. Both organic solvents were tested as Mobile Phase B, as well as to create neat solutions of 100 ng/mL in \(organic:inorganic=95:5\) to \(5:95\) ratios. This resulted in two sets of 19-samples studied in duplicates for acetonitrile as Mobile Phase B and methanol as Mobile Phase B. The entire study was repeated 3-times prior to concluding that acetonitrile serves as the better Mobile Phase B due to the high signal recovery (almost 2-folds higher in average signal intensity for all eight antidepressants). For the lowest organic solvent percentage, it was determined that a ratio of \(organic:inorganic=60:40\) yielded a Gaussian peak with the best MRM intensity, along with a high signal and low noise, comprehensive across all antidepressants for both methanol and acetonitrile.
While the final ratio of the organic and inorganic solvents was determined, an additional study had to be conducted to assess the performance of the desired internal standards (IS) in the organic solvents26, as well as the said solvent’s efficacy as a protein precipitation solution. To do this, internal standards were all dissolved in: HPLC methanol (with 0.1% formic acid) and HPLC methanol (no acid), and HPLC acetonitrile (with 0.1% formic acid) and HPLC acetonitrile (no acid). This resulted in 4 Daily Working Solutions (DWS) at a concentration of 50 ng/mL across the board for all the antidepressants. These 4 DWS were tested using an identical extraction process: addition to serum, followed by 5-minutes incubation, vortex mixing and centrifugation for separation. For each concentration (calibrators level L1-L5) it was observed that acetonitrile with 0.1% formic acid yielded the highest signal intensity for all analytes (BUP appeared to have a better signal intensity with methanol and 0.1% formic acid, however, the quantified difference was not statistically significant) (data not reported). This study was repeated three times, with n = 3 for each calibrator with each DWS to reach a conclusive result. Unlike the previously reported studies10–14, we shortened the sample preparation protocol by utilizing the effect of pH on the separation buffer. As reported by Lin et al.25, a better separation occurs for when an acidic buffer is used. Utilizing this, we have innovated the sample preparation technique such that the DWS was acetonitrile measuring at pH = 2.3 at room temperature with the labeled internal standard incorporated in it. For this two-fluid system (as shown on Fig. 1), we assumed a thin membrane formation at the interface of the serum and DWS. That said, the bulk pH of a system like this varies vastly from the surface pH28, implying the need for an external motion to minimize the effect of membrane separation height, h, as shown by Ohshima and Kondo28 derivation for repulsion, P, at the DWS and serum boundary:
\({P}_{max}=4nkT\text{sin}{h}^{2}\left({e}^{\frac{{\phi }_{DON}}{2kT}}\right)\) (Eq. 1)
where, \({\phi }_{DON}\) is the Donnan potential. Even though for an almost infinitesimal system like ours with surface pH < 3, the bulk pH should not differ too much from the boundary pH, we still incorporated the principles of micromixing29, for our two-fluid system’s homogeneity as explained by the Kolmogoroff microscale equation \({\lambda }_{K}\) (Eq. 2):
\({\lambda }_{K}={\left(\frac{{\nu }^{3}}{ϵ}\right)}^{\frac{1}{4}}\) (Eq. 2)
where, \(\nu\) is the kinematic viscosity of the fluid in the system, and \(ϵ\) is the power input per unit mass to the bioreactor. In our case, the mixing occurs below the scale of \({\lambda }_{K}\) where molecular diffusion takes place (stage 2 of Fig. 1). For the third stage, i.e., efficient precipitate separation, we utilized the effect of temperature on organic solvents, where a near 0°C temperature leads to certain protein insolubility30. Thus, we have demonstrated the combined effect of centrifugation and temperature at Stage 3 of our separation process, leading to the precipitation of the unwanted proteins and salts, leaving us with a clear and easy to remove supernatant, that can be directly injected to the LC-MS/MS for analysis, thereby eliminating the need for sample drying.
3.3 QC, Linearity, Precision and Accuracy
The concentration of each antidepressant for the calibrators was determined by studying the therapeutic ranges (Table 1b). It was important to establish the point of column saturation since there were 8 different antidepressants. Initially, a prototype was designed such that the concentration varied for each analyte, in order to cover the large therapeutic range of BUP. However, upon investigation, it was concluded that the C-18 column being used would saturate above a 230 ng/mL concentration for all analytes uniformly. This was studied using two different C-18 columns: Perkin Elmer Brownlee SPP C18, and Agilent Poroshell C18, both yielding similar outcomes. Therefore, the final prototype kit was designed to avoid any column saturation. Figure 3 shows the concentration of each calibrator for all eight antidepressants, analyzed in sextuplets, and averaged, and the corresponding theoretical value for that calibrator level. This resulted in measured concentrations at L1 = 1 ng/mL, L2 = 4 ng/mL, L3 = 15 ng/mL, L4 = 60 ng/mL, and L5 = 230 ng/mL, that were further used to access the assay linearity, precision, and accuracy.
Linearity, precision, and accuracy data were all obtained from the same experiment, where the linearity of the prototype kit was assessed by looking at the \({R}^{2}\)-values for each standard curve produced. Precision was reported as %CV measured ± ≤ 20% for intra and inter-assay runs, across 6-days with n = 3 for each calibrator level per antidepressant. Table 3 (a-h) show each antidepressant, with their measured concentrations, %CV, %Accuracy and \({R}^{2}\)-values. Samples were found to be within 20% for all analytes (except e) for intraday and interday measurements. The sample average sample accuracy measured at 100 ± 20. The average \({R}^{2}\)- values were all above 0.9, indicating assay linearity. Achieving the desired precision for this study was challenging, since there were multiple internal standards that would lead to a high background noise and impact the precision. The IS concentration for all the drugs had to be optimized (tested concentrations: 1.5 ng/mL, 15 ng/mL, 25 ng/mL, 50 ng/mL) such that there would be no interference leading to a high %CV. It was determined that 25 ng/mL served the best signal without significantly interfering with the assay precision and still producing IS curves that were consistent. It is worth noting that the LOQ had a notable high data spread for all the tested drugs, since the measured concentration was relatively small and only slightly above the background the noise produced by blank serum.
Additionally, we faced a separation challenge that impacted the assay precision. This was mitigated by investigating and modifying our LC method. We tested three different LC gradients for a partial loop fill. The three separation methodologies tested had the following configurations:
-
3-minutes run time per sample: flow rate from 0.7 mL/min 100% A for one minute, 0.9 mL/min and 95% B, with 30 seconds column equilibration.
-
5-minutes run time per sample: steady flow rate of 0.7 mL/min, 95% A for two minute and 50-seconds, 95% B for one minute, with 2-minutes column equilibration.
-
7-minutes run time per sample: steady flow rate of 0.5 mL/min, 95% A for 50-seconds to 5-minutes, switching to 95% B, and back to 95% A from 5.51 minutes to 7 minutes (equilibration).
Out of the tested separation gradients, the seven-minutes method yielded the best analyte separation in tandem with the optimized MS method. This was crucial for the early eluting analyte OLN that would otherwise pass away without undergoing any separation.
3.4 Carryover and Stability
To establish the translatability of this assay at a clinical setting, it was important to assess the carryover effect as well as the long-term stability of the prototype. A minimal carryover is desirable since it would enable testing of more patient samples, since the number of blank injections between samples can be minimized. Both Carryover and Stability studies were conducted according to our previous work22. Despite employing Needle Wash Solvent Chemistry, and advanced autosampler washes, there seemed to be statistically significant sample carryover for all the antidepressants, i.e., %CV of L1 followed by L5 injected exceeded 25% across the board (n = 24), when compared to the baseline L1 (injected after blank). Therefore, we recommend performing two blank injections prior to each unknown sample quantification. This data spread is shown in Fig. 4a, post outlier removal.
In addition to assessing sample carryover, we performed a 14-days accelerated stability study (224 days extrapolated), with n = 2 per level per day, storing the samples and − 20°C and stressing them at + 20°C. We observed about 80% degradation in samples for all analytes at the 14th -day mark, which translates to about 7.5 months at storage temperature. When looking at concentration by area normalized to 1 to indicate degradation, CIT, DES, IMI, and MLN exhibited a steadier decline unlike the steep decline of the other drugs. However, when the L5 data for all days and all analytes were used to construct a trendline to predict degradation, only CIT, MLN and VIL exhibited analyte stability of about 1.1 months (two days, interpreted). BUP and OLN showed the highest degradation within two-days, while others overlapped in their degradation trend. Since the number of samples tested were not sufficient, statistical claim about the stability of all the analytes in the prototype kit in inconclusive.
3.5 Ease of Automation
Mere establishment of an effective protocol for antidepressants monitoring is futile unless it can be translated for a large number of sample testing, enabled by automation. To do this, we performed a head-to-head analysis for the ease of automation of our prototype kit against the commercial Eureka kit. The parameters investigated included: speed/sample preparation time, automation compatibility, and overall clinical relevance. While both protocols have similar workflows, the commercial kit is less conducive to automation as it relies on a tube-based method rather than a plate-based method and requires 100 µl of sample input. This is a significantly large volume compared to our prototype kit, which requires only 20 µl input. Additionally, tube-based methods decrease the number of patient samples that can be prepared and subsequently analyzed per unit of time. Reported run times for completion of sample preparation automation protocol are reported below in Table 5. Additionally, the feasibility of each of the automation techniques were reviewed with greater ease being seen in the operation of the proposed prototype kit. This means that it has the potential to be a viable option for clinical application, especially those involving low-volume sample collection techniques. Just like our previous study22, we assessed the efficiency of the automated sample preparation using %CV against manual plating. The results obtained were comparable, with the added advantage of automation to minimize error propagation.
Table 5
Time comparison for the proposed study against a commercial kit.
Sample Preparation Method
|
Calibrators in triplicate only (mm:ss)
|
Full plate with triplicate calibrators and 48 patient samples (mm:ss)
|
Additional Time (Off Deck)
|
Total Time (Full Plate) (mm:ss)
|
Proposed Protocol
|
09:21
|
17:15
|
+ 05:00 Plate Shaker
+ 10:00 Centrifuge
|
32:15
|
Eureka
|
05:26
|
39:02
|
+ 00:10 Vortex
+ 10:00 Centrifuge
+ 00:05 Vortex
|
49:17
|