Impact of Salt and Urine Matrices on Electrokinetic Stacking Coupled to Paper Spray Mass Spectrometry

Abstract

Relative to liquid chromatography coupled to electrospray ionization (ESI), paper spray ionization (PSI) can more rapidly ionize analytes from complex matrices with less sample preparation and less expensive instrumentation. However, matrix effects can lead to ion suppression and poorer sensitivity. To overcome these problems, we combined on-paper faradaic ion concentration polarization (f-ICP), a form of electrokinetic stacking, with PSI to preconcentrate analytes directly on the spray substrate prior to ionization. Electrokinetic stacking also enables desalting because of the separation of analyte molecules from small ions like sodium. 3D-printed cartridges containing chemically modified PTFE papers were utilized with a high voltage isolated power supply to perform on-paper stacking from the same device as paper spray. In this study, ion suppression was compared for paper spray with and without electrokinetic stacking in the presence of five salts and artificial urine at varying concentrations. The signal enhancement achieved for f-ICP/PSI relative to normal PSI was also quantified. Finally, the effect of salt type and concentration on stacking time was assessed. Increasing salt concentrations generally results in longer stacking times and increased currents, indicative of increasing quantities of charge being desalted. In the presence of single salt-containing matrices, f-ICP/PSI frequently decreased ion suppression and gave a signal enhancement of 2 to >200× for small molecule drugs compared to unstacked paper spray (uPSI). In artificial urine, a more complex matrix with various salts and metabolites, f-ICP/PSI did not reliably decrease matrix effects due to creatinine and other matrix components co-stacking with the analytes. Nevertheless, f-ICP/PSI still resulted in signal enhancements of 8 to >50× relative to uPSI due to analyte preconcentration from electrokinetic stacking. Overall, this research demonstrated that f-ICP/PSI results in significant sensitivity improvements relative to uPSI.

GRAPHICAL ABSTRACT

INTRODUCTION

Paper spray ionization (PSI) allows for the analysis of proteins, lipids, drugs, and metabolites in biofluids rapidly and with minimal sample preparation.^1–4 PSI has also been used in food analysis, environmental monitoring, forensic science, and pharmaceutical drug checking due to its relatively simple setup.^5–9 Indeed, at its simplest, PSI utilizes only a triangular strip of paper coupled to a high voltage output via an alligator clip. Despite the improvements in analysis time and cost, matrix effects and a relatively low level of sensitivity compared to traditional LC-MS approaches are recognized limitations of PSI.^{10, 11}

The most commonly proposed solutions to enhance PSI have been modifications to the paper, or using a different material entirely, to decrease the polarity and/or porosity of the spray substrate¹². These changes likely increase analyte recovery, improve ionization efficiency by decreasing Taylor cone size, or optimize analyte-substrate interactions. Examples include spraying from hydrophobic paper¹³ or teflon,¹⁴ wax-patterning microfluidic channels into the substrates,¹⁵ coating the paper with polystyrene beads,¹⁶ using silylated or thiolated papers with chemical modifications¹³, and coating the paper with carbon via sputtering¹⁷ or nanotubes¹⁸. Another approach is selective preconcentration of the analyte, which has been achieved with blade spray,^{19, 20} PSI with aptamer-coated²¹ or molecularly imprinted polymer²² papers, chemical modifications to the paper²³, or by combining solid-phase extraction²⁴ and antibody purification^{25, 26} within the PSI cartridges.

We recently reported on-paper preconcentration using electrokinetic stacking, namely field-amplified sample stacking (FASS) and Faradaic ion concentration polarization (f-ICP), directly on the spray substrate.^27–29 In this paper, we focus on f-ICP because it is more amenable to salty samples.^30–32 In f-ICP, an additional voltage is applied across the length of the paper, inducing electrophoresis and electroosmotic flow, causing charged analytes to stack into a thin region of increased local concentration as they migrate toward the tip of the paper for ionization. This method yields crude separations and semi-selectively extracts analytes from matrix components (more on the specific mechanism in the Theoretical Basis subsection, below). These improvements in f-ICP/PSI come at the expense of slower analysis time and more complexity relative to uPSI.

In this study, we examined the effects of salt and urine components on ion suppression and signal enhancement for f-ICP/PSI relative to uPSI. Urine contains salts, like sodium chloride, as well as metabolites, such as creatinine, all of which can cause significant ion suppression.¹⁰ As such, we tested the effects of multiple salts ranging from 10 to 250 mM as well as artificial urine at multiple dilution levels with both uPSI and f-ICP/PSI.

THEORETICAL BASIS

Faradaic ion concentration polarization operates by an electrokinetic stacking mechanism exemplified in Figure 1 for both an offline device (A) and an online cartridge (B) for coupling to PSI.^{29, 33} For cationic analytes, a positive zeta potential is imparted to the paper using silylation reagents (see methods). Electroosmotic flow (EOF), therefore, moves toward the anode (positive terminal in this electrolytic cell), opposite to electrophoresis (EP) for cations. A background electrolyte (BGE) containing MS compatible salts, such as ammonium acetate, is added to a reservoir at the cathodic end, and the sample solution is pipetted primarily on the top paper (although all papers in the circuit must be wetted – either with BGE or analyte solution – to ensure conductivity throughout). Upon application of a ~200-300 V potential across the paper, EOF and EP begin along with the formation of an ion depletion zone (IDZ) at the anode due to neutralization of acetate ions by protons produced through solvent electrolysis.³¹ As the reaction progresses, the IDZ expands towards the cathode. At the border of the IDZ is a steep electric field gradient owing to the difference in conductivity between the IDZ and the rest of the paper containing unneutralized BGE. Cationic analytes with an EP mobility within a broad range will form a concentrated band, “stacking” at the IDZ boundary where their EP is offset by the opposing EOF. Highly mobile cations, like sodium ions, will not stack because their EP velocity greatly exceeds their EOF.^{30, 31} For the online device, seen in Figure 1(B), the stacked band forms on a lower paper and will pass through the paper tip-junction, where ionization occurs. The entire process can be reversed for anionic analytes²⁹, but only cationic analytes were studied here.

Figure 1.

Depiction of the (A) offline and (B) online f-ICP 3-D printed apparatuses. Apparatus (A) was utilized in all chronoamperometric and chronometric experiments; apparatus (B) was utilized in all online MS experiments. (A) shows the anodic (left) and cathodic (right) redox reactions driving IDZ formation. (B) differs mainly in the arrangement of the paper substrates to create a point from which a Taylor cone can be generated. The shaded region on the paper substrate in (A) represents the IDZ, which expanded from the anode to the band position. In (B), the IDZ forms on the lower substrate (not visible in the diagram) and expands through the junction of the two paper tips, eventually passing onto the top substrate. (B) additionally shows the alignment of the online device, relative to the mass spectrometer, with the sweep cone and ion transfer tube visible in bronze. The anode and cathode are each depicted at the “| + |” (red) and “| − |” (black) positions, respectively.

EOF, EP, and the IDZ propagation can be fine-tuned through selective substrate choice and/or modification, alterations to the substrate geometry, changes to the BGE concentration or type, adjusting the applied voltage, etc. to control net movement of the band as well as its formation position. In paper-based f-ICP systems specifically, stacking time depends on several factors including electric field strength, substrate material, electrode placement and surface area, and solvent composition.^34–36 The salt content of the sample matrix is also a notable variable, the effects of which were explored in this publication.

MATERIALS AND METHODS

CHEMICALS AND MATERIALS

LC/MS water, methanol (HPLC grade), ethanol (reagent alcohol), 2-propanol (HPLC grade), acetic acid (glacial), formic acid (88%, ACS grade), ammonium acetate (LC-MS grade) and bis[3-(trimethoxysilyl)propyl]amine (98%) were purchased from Fisher Scientific (Pittsburgh, PA, USA). Solid methylthioninium chloride trihydrate (methylene blue dye) was purchased from Alfa Aesar (Haverhill, MA, USA). Stock standards of 1 mg/mL in methanol of each methamphetamine, norfentanyl (oxalate, as free base), N-desmethyl-cis-tramadol (HCl, as free base), and norhydrocodone (HCl, as free base), as well as solid potassium chloride, sodium chloride, sodium fluoride, sodium carbonate, sodium sulphate, sodium citrate dihydrate, and urea were all purchased from Sigma Aldrich (St. Louis, MO, USA); solid magnesium sulphate heptahydrate, ammonium chloride, and disodium hydrogen phosphate dihydrate were purchased from Fisher chemicals (Fair Lawn, NJ, USA). Solid uric acid, creatinine, magnesium chloride, calcium chloride, and potassium oxalate monohydrate were purchased from Thermo scientific chemicals (Heysham, Lancashire, UK). Additional information on the chemical analytes, including their formulae, structures, and masses can be found in Supplementary Table S6.

AP40 glass fiber filters were purchased from Millipore Sigma (Burlington, MA), and hydrophilic polytetrafluoroethylene (OH-PTFE) membrane filters were purchased from Cobetter (Hangzhou, China).

Platinum wire (0.25 mm diameter, 99% metal basis) was purchased from Thermo Scientific chemicals. An LCS150 AC/DC converter and a RECOM REM5E DC/DC converter (8 kV isolation) were purchased from Mouser (Mansfield, TX). A DEVMO 8-32V to 45-390V DC-DC ZVS step-up boost converter module was purchased on Amazon. 0.75 mm Fiberlogy polypropylene filaments (natural, grey colors) were purchased from Granger.

SAMPLE PREPARATION

A 5 ppm solution of methylthioninium in 50:50 (v/v) MeOH/H₂O with 1% acetic acid and 0.01% formic acid was prepared as an analyte solution for all offline trials. At this concentration, methylthioninium is visible to the eye after stacking.

A 1 ppm solution of methamphetamine, norhydrocodone, norfentanyl, N-desmethyl-cis-tramadol, and methylthioninium chloride was prepared in a 30 mM ammonium acetate with 1% acetic acid and 0.01% formic acid solution in 50:50 (v/v) MeOH/H₂O for all online f-ICP/PSI trials. The lower 1 ppm methylthioninium is also visible to the eye on the online geometry, although the band is faint. For online uPSI trials, ammonium acetate was left out of the analyte solution formulation.

Identical solutions with concentrations of 10, 25, 50, 75, 100, or 250 mM of sodium chloride, magnesium chloride, potassium chloride, sodium fluoride, or sodium carbonate were used as analyte solutions for the offline trials. All online trials were analyzed in 10 mM of the same salts (one salt per sample) or 10-100 mM concentrations for NaCl only.

Two separate BGE solutions were used: 20 mM ammonium acetate for online trials and 30 mM for offline trials, both with 0.01% formic acid and 1% acetic acid in 50:50 (v/v) MeOH/H₂O (pH ≈ 4). The BGE was deposited into the solvent well in the cartridge. The lower BGE concentration was preferred for online MS experiments due to improved ionization. However, the same trends observed in the offline experiments at higher BGE concentrations were assumed to hold with the lower BGE concentrations and differing geometries of the online experiments (more on this below, in the Offline Results: Chronometric Measurements subsection).

A spray solvent of 50:50 (v/v) MeOH/H₂O with 1% acetic acid and 0.01% formic acid was used for all uPSI experiments. For the online f-ICP experiments, the BGE and sample solution served as the spray solvent.

Multipurpose artificial urine (MP-AU) was prepared via a modified procedure from Sarigul, et al. to ensure each component included in the MP-AU recipe was of average concentration between the two sexes.³⁷ The MP-AU contained urea, uric acid, creatinine, and various salts at typical physiological urine concentrations (see detailed composition in Supplementary Table S1) and was serial diluted to 75%, 50%, and 25% of the original average physiological concentrations. Methamphetamine, norhydrocodone, norfentanyl, N-desmethyl-cis-tramadol, and methylthioninium chloride in MeOH were then spiked 1:1 into the MP-AU dilutions to a final concentration of 1 ppm. The dilution of MP-AU samples with methanol prior to spotting ensured the sample would absorb into the relatively hydrophobic modified PTFE spray substrate.

All online MS samples were run at a final concentration of 1 ppm, regardless of variable matrix concentration. The only concentrations varied for this study were the component salt or MP-AU matrices.

SUBSTRATE PREPARATION

Glass fiber and OH-PTFE papers were prepared as described by Rydberg, et al.²⁹ All papers were sonicated in 50:50 (v/v) methanol/isopropanol for 20 minutes prior to silylation. The silane coupling reagent Bis[3-(trimethoxysilyl)propyl]amine was prepared at 1% (v/v) in 85:15 ethanol/water and was allowed to equilibrate for 15 minutes. Then, the papers were immersed in the reagent solution for two hours, stirring every 20 minutes. The papers were next rinsed in methanol, air dried for one to two hours, and kept under house vacuum overnight. Finally, the modified papers were sonicated in 80:20 methanol/isopropanol for 30 minutes as one final cleaning. This modification imparted a positive zeta potential to the substrate, reversing EOF as inferred from stacking behavior. Both papers were inherently negatively charged without this modification.

OFFLINE CARTRIDGE

Offline chronoamperometric (current versus time) and chronometric (measuring time required for band formation) experiments were performed using a 3D printed cartridge, holding a cathodic substrate (S_C) and an anodic substrate (S_A). CAD files for all cartridges used in this study can be found in the supplementary materials. Both papers were glass fiber filters hand-cut with a razorblade to 5 mm × 17 mm (S_C) and 5 mm × 40 mm (S_A). The sample solution was deposited onto S_A, which was secured longitudinally across the substrate by printed tabs and laterally by the anode, and contacted S_C. S_C was positioned in a 0.75 mL well with a platinum electrode secured vertically 4.6 mm from the end of S_C. This substrate does not directly contact the cathode; the circuit is closed through the aqueous media. A 0.5 mm gap between the papers and cartridge walls was included to prevent undesired pressure gradients due to solvent contacting the walls and physical constriction of the paper capillary channels. The cartridge surface contacting the substrate had 1 mm deep trenches 1 mm wide and 1 mm apart to minimize solvent pooling under the substrates.

f-ICP/PSI CARTRIDGE

All online experiments were run on 3D printed cartridges, adapted from Ryberg.et.al, which consist of two distinct substrate holders, as below in Figure 2.²⁹

Figure 2:

Schematic of the f-ICP/PSI online cartridge paper alignment and construction. (A) shows the lower portion of the device, with the cathodic (S_C) and anodic (S_A) substrates’ positioning and highlights the well where background electrolyte (BGE) is deposited, to the posterior of the device. (B) shows the upper part of the device and the positioning of the deposition substrate (S_D) within. (C) shows how the two halves of the device fit together.

The online cartridge contained three substrates: the anodic and cathodic substrates, S_A and S_C, respectively, as well as the deposition substrate S_D. The additional substrate is required for MS ionization; a sharp point is needed between the anode and cathode for PSI. Substrates were hand-cut via a razor blade to 35 mm × 5 mm (S_D, OH-PTFE), 25 mm ×5 mm (S_A, OH-PTFE), and 15 mm × 4 mm (S_C, glass fiber), with S_A and S_D additionally having 30° tips.

S_A and S_C were secured in the lower substrate holder. S_A was positioned proximally to the mass spectrometer, in the lower anterior part of the cartridge (off-white) to contact S_D at a 25.3° angle and was secured beneath the anode. S_C, positioned in the lower part of the device within the BGE well (which also contained the cathode), came up to contact the S_D at a 90° angle. Similar to the offline device, this substrate does not directly contact the electrode; the circuit here is also closed through the aqueous media. S_D was held within the second, upper substrate holder. This upper part of the cartridge had a mesh-like surface for the paper to sit upon, preventing solvent from pooling underneath S_D. The papers must be well-aligned and centered with respect to one another with only the tips contacting. Bad tip alignment could lead to poor spray, no continuity between the two substrates, or the analyte band bypassing the ionization region.

Additional information on the device wiring and electrode placement can be found in the supplementary materials section, Schematic of Device Wiring. These angles, lengths, contact areas, and device alignment were previously optimized by Rydberg, et al. for ionization and stacking efficiency.^{29, 33}

uPSI CARTRIDGE

For all uPSI experiments, a 3D printed cartridge was used to secure the substrate and allow for high voltage contact via a copper wire. Papers were laser cut to 5 mm × 1.8 mm (glass fiber) with a triangular tip of 30°. The copper wire was energized through the mass spectrometer via an alligator clip connection to its high voltage source.

OFFLINE CHRONOAMPEROMETRIC AND CHRONOMETRIC EXPERIMENTS

For all chronoamperometric studies, papers were secured within the offline device followed by deposition of 500 μL of 30 mM BGE solution into the well and spotting 60 μL of 5 ppm offline methylthioninium solution onto S_A. 240 V was applied to initiate stacking. The circuit was connected to a multimeter to measure current with respect to time, and video was recorded of each trial using a mobile phone camera for 6 minutes. Current was recorded every 15 seconds. The videos were used to note the time when the methylthioninium band initially formed visibly and when it reached its maximum visual intensity (t_{s, initial}, and t_{s, concentrated}, respectively).

f-ICP/PSI PROCEDURE

For all online f-ICP experiments, the papers were positioned within the device as described above and the assembled cartridge was placed on an XYZ stage ~0.5 cm away from the MS inlet. The external power supply was connected to the cartridge, as well as a cable from the MS high voltage supply to provide ionization. For all samples, 500 μL of 20 mM BGE was deposited in the well, and 60 μL of analyte solution was spotted on S_D and S_A. An external 240 V was applied to initiate stacking, with band propagation tracked via the spiked methylthioninium. Visible indication of band formation via the spiked dye is convenient but not required.

In the control sample and single-salt matrices, analyte solution was divided between S_D and S_A, with the bulk on S_D. For the analysis of MP-AU, the sample solution was only deposited on the rear of S_D, near S_C, and the sample was allowed to spread over the remainder of S_D toward S_A. To fully wet the substrate area, 10 μL of BGE solution was additionally spotted onto S_A. For more information on the variable spotting techniques, with method optimization between sample types, see the supplementary materials section, Optimization of f-ICP Parameters.

uPSI PROCEDURE

The cut paper was placed into the 3D-printed cartridge and positioned ~0.5 cm away from and in line with the MS inlet. 15 μL of the sample solution was spotted in the middle of the substrate and allowed to dry for 30 seconds. 75 μL of spray solvent was spotted at the rear of the substrate and a spray voltage was applied for 60 seconds.

INSTRUMENTAL PARAMETERS

All MS analyses were conducted using a Thermo Fisher LTQ-XL linear ion trap (Thermo Scientific, San Jose, CA, USA). A positive polarity was used, with a scan range of 100 to 300 Th with one microscan. The spray voltage was set between 4.0 and 4.6 kV, with saltier or relatively “dirtier” samples requiring higher ionization voltages to maintain a consistent spray current but maximized such that less than 1 μA spray current was achieved without coronal discharge. All uPSI experiments used a consistent 4 kV spray voltage. Voltage was only varied for f-ICP/PSI, for the non-control solutions. The capillary inlet temperature was set to 275°C. A maximum injection time of 100 ms was allowed, with an AGC target of 3×103. The LTQ scan time was about 150 ms plus the ion injection time, so spectral collection frequency varied from about 4 to 7 Hz depending on ion flux. For all f-ICP/PSI experiments, only full MS data was collected. For all uPSI experiments, full MS and MS², via collision-induced dissociation (CID), data were collected simultaneously. MS² data was collected only for confirmation of peak identity. Fragmentation energy was optimized per analyte to ensure that the parent peak was consistently ~30% the height of the base peak, and exact values with resultant exemplar spectra can be found in the supplementary materials section, Supplementary Spectra. Absolute abundances were collected as averaged peak heights of the molecular ions (full MS). For uPSI, the entire 60 s signal block was averaged. For f-ICP/PSI, spectra were averaged during the period of increased abundance as the stacked analyte band passed through the paper tips, which varied between trials and conditions.

MATHEMATICAL ANALYSES

Microsoft Excel and R Studio were used for all statistical analyses. All “±” errors reported herein are 95% confidence intervals. Any error bars are at k = 1.

Ion suppression was calculated according to Equation 1 by comparing a matrix containing sample with a neat, matrix-free control and is expressed as a percentage.^{10, 38}

$$IonSuppression=IS=1-\frac{Abundance(Matrix)}{Abundance(Control)}$$ (1)

Change in ion suppression was calculated according to Equation 2 and is expressed as a percentage point change.

$$ChangeinIonSuppression=\Delta IS=IS\left(f-ICP/PSI\right)-IS\left(uPSI\right)$$ (2)

Therefore, negative changes indicate improvement (less ion suppression) when going to the electrokinetic methodology.

Signal Enhancement was calculated according to Equation 3 and is expressed as a fold-increase.^{27, 29}

$$SignalEnhancement=SE=\frac{Abundance(Analyte,f\text{-I}CP)}{Abundance(Analyte,uPSI)}$$ (3)

Note that f-ICP/PSI could theoretically be more impacted by ion suppression than uPSI while still having high signal enhancement. Thus, conclusions regarding methodological efficiency from either signal enhancement or ion suppression alone cannot be drawn.

All figures show unweighted linear regressions for visualization purposes only; conclusions were drawn on the weighted least squares (WLS) models in R Studio. All conclusions are at the α = 0.05 level. Uncertainties and effective degrees of freedom were obtained via Welch–Satterthwaite propagation during preprocessing.^39–41

For all per-point significances and significances of differences between points in directly measured and averaged values, such as abundances and stacking times, a pairwise Welch–Satterthwaite two-sample t-test was performed.⁴⁰ For significances of signal enhancement, the comparison was against 1×; all other measurements were against 0×. This was additionally performed pairwise between the points composing any linear model to determine at what unit change a result shows a measurably different value (i.e. to determine rather a difference is present between 10 mM vs 25 mM, or only 10 mM vs ≥50 mM, etc.), as well as pairwise across factors.³⁹ Two-sided p-values were computed from the resulting t-statistics.

For per-point and between-point significances of ion suppression, signal enhancement, and change in ion suppression between methods, comparisons were made using log-transformed absolute abundances.^{42, 43} Because ratio data are right-skewed, analyses on the raw scale would violate the normality assumption of Welch–Satterthwaite tests. Log transformations of differences, which is algebraically equivalent to comparing ratios, yields approximately normal sampling distributions, consistent with the assumption of log-normal underlying processes.^42–44 Where raw replicates were unavailable due to the desired measurements themselves being a function of those raw replicates, Wald tests were performed for comparisons.⁴⁰

For comparisons of trends between datasets, weighted least squares linear regressions were first performed. All data points were weighted against 1/Standard Error².^{45, 46} Weighted R² values were used to assess linearity. Two-sided p-values were calculated via Welch-type t-tests to test the hypothesis that the slope differed from zero. Trend strengths were quantified using standardized regression coefficients (β*), with qualitative cutoffs of <0.1 being negligible, 0.1 – 0.3 small, 0.3 – 0.5 moderate, and ≥0.5 large.^{47, 48} Differences between trend estimates were assessed using pairwise z-tests on the regression coefficients to yield p-values for comparison.

Effect sizes were quantified using Hedge’s g, a bias-corrected standardized mean difference, similar to Cohen’s d, which ensures that effect sizes are not systemically inflated due to modest sample sizes.^{49, 50} Differences in effect sizes across factors were calculated, with Welch-type analyses performed on the δg statistic, to assess significance of the differences. Interpretation of the g-statistic is identical to that of Cohen’s d. Test statistics of <0.2 were considered negligible, 0.2 – 0.5 small, 0.5 – 0.8 moderate, and ≥ 0.8 large.^{48, 51}

Under standard statistical frameworks, effect size calculations require true experimental replicates rather than repeated observations derived from those replicates. This is because effect size is defined as the magnitude of an observed effect relative to the standard deviation of the effect itself, not relative to the uncertainty in the measurement.^{50, 52} Furthermore, Hedge’s g adjusts for bias based on a small sample size.⁵⁰ Propagated statistics, including ion suppression, signal enhancement, and change in signal enhancement, do not have a true ‘n’ value, and uncertainty is measured in standard errors, rather than in standard deviations, and do not retain direct estimates of sample variance.^{40, 41, 53} Therefore, an indirect measure of effect size for these metric variables was required. To that end, raw abundances were compared because the change in abundance between methods is equivalent in effect to the signal enhancement observed, by definition.^{10, 38} Similarly, for effect sizes of change in ion suppression, differences between the effect sizes of ion suppression between methods were calculated, rather than calculating an effect size for the change itself.

For all chronoamperometric studies, integrated charge was measured as a unit of area under curve (AUC) for the current vs. time graphs via trapezoidal integration from t₀, the initial application of voltage, to 6 min.⁵⁴

RESULTS AND DISCUSSION

OFFLINE RESULTS

Chronometric Measurements

Trends in stacking times and electric current as the salt matrix concentration and type varied were evaluated using the offline device in Figure 1(A) for convenience. Comparisons were performed between the offline and online devices to ensure trends generated on the offline device were transferrable to the online device. No significant differences were observed between the curves generated between the two devices (see the supplementary materials section, Comparison of Data Generated on Offline and Online Cartridges, for additional information on the comparisons performed).

Stacking time, both initial and concentrated, was determined for five different salts at varying concentrations. The concentration range for MgCl₂ differs because stacking terminated at a lower concentration for that species. Stacking time was positively correlated with salt concentration in all cases, as seen in Figure 3, although the correlation for NaF was weak. The addition of salt at any concentration ≥25 mM increased stacking time (g > 0.8), except for 25 mM NaCl and 25 mM MgCl₂ concentrated stacking times, where no significant change was observed (p = 0.4979 and 0.1008, respectively).

Figure 3.

(A) Initial and (B) concentrated stacking times for methylthioninium dye in differing matrices composed of variable concentrations of variable identity salt, with 30 mM BGE, at 240 V; p-values test the hypothesis that m ≠ 0 (the trend is significant).

Furthermore, only MgCl₂ had a significant effect at the 10 mM (lowest) concentration, where a large increase in initial stacking time (p < 0.0001, |g| = 23.947) and a large decrease in concentrated stacking time (p = 0.0094, |g| = 5.006) was observed. Interestingly, initial band formation with MgCl₂ occurred last among all salt matrices, but concentrated formation occurred first. Bands in these trials appeared visually less intense relative to the control and other salt solutions. This trend was not present among the other salts analyzed. The decrease in t_{s, concentrated} implies that the baseline inclusion of some salts may result in faster concentrated band formation, but is still largely disadvantageous due to ion suppression (more on this in the Online Results subsection, below).

A pairwise comparison was conducted at the 50 mM concentration between each of the salt matrices analyzed against both initial and concentrated stacking times, presented in Table 1. Fifty millimolar was the highest concentration where stacking was still achievable for all matrices. The data in Figure 3 and Table 1 suggest that both cation and anion identity affect stacking time. For example, stacking was not achieved with MgCl₂ concentrations above 50 mM but remained possible at >250 mM NaCl. If chloride alone drove the trend, the cutoff for which stacking could be achieved in a NaCl matrix should be ~100 mM (twice that of MgCl₂), but this was not observed. Similarly, if only the cation influenced the rate, doubling sodium concentration should alter stacking time; however, NaF vs. Na₂CO₃ showed no significant difference (p = 0.3247 t_{s, initial}, 0.8896 t_{s, concentrated}).

Table 1.

Pairwise comparisons of stacking times of methylthioninium in differing salt matrices; negative differences indicate a longer stacking time of the second solution analyzed relative to the first.

Welch-Type Pairwise t-Tests
	Initial Stacking Time (s)			Concentrated Stacking Time (s)			Summary
	Mean Diff.	p Value	Significant	Mean Diff.	p Value	Significant	Significant in
Control-vs. NaCl	−89	0.0002	Yes	−105	0.0042	Yes	Both
Control vs. KCl	−101	0.0001	Yes	−155	0.0001	Yes	Both
Control vs. MgCl2	−303.3	<0.0001	Yes	−45	0.0076	Yes	Both
Control vs. Na2CO3	−42.3	0.0325	Yes	−136	0.0003	Yes	Both
Control vs. NaF	−30.3	0.0037	Yes	−133.3	0.0071	Yes	Both
NaCl vs. KCl	−12	0.1222	No	−50	0.0290	Yes	Concentrated
NaCl vs. MgCl2	−46.3	0.0006	Yes	−108	0.0301	Yes	Both
NaCl vs. Na2CO3	46.7	0.0336	Yes	−31	0.1025	No	Initial
NaCl vs. NaF	58.7	<0.0001	Yes	−28.3	0.2243	No	Initial
KCl vs. MgCl2	−34.3	<0.0001	Yes	−58	0.0021	Yes	Both
KCl vs. Na2CO3	58.7	0.0110	Yes	19	0.1669	No	Initial
KCl vs. NaF	70.7	0.0016	Yes	21.7	0.3030	No	Initial
MgCl2 vs. Na2CO3	93	<0.0001	Yes	77	0.0047	Yes	Both
MgCl2 vs. NaF	105	0.0003	Yes	79.7	0.0270	Yes	Both
Na2CO3 vs. NaF	12	0.3247	No	2.7	0.8896	No	ns

Chronoamperometric Measurements

Chronoamperometric data for the five salts at varying concentrations was collected to determine the integrated charge, which served as a measure of the electrochemical work associated with changes in ion depletion zone formation and matrix desalting. Exemplar data for NaCl-containing sample solutions are shown in Figure 4 (A). Each salt gave similar curve shapes and changes in shape with increasing salt concentration. Figure 4 (B) shows the integrated charge versus concentration for NaCl. The inclusion of any salt at any concentration resulted in a large increase in integrated charge in the system, relative to the control (p < 0.05; |g| > 3). Moreover, all concentrations of all salts run were differentiable within species, except the 50 mM NaCl sample which had larger measurement uncertainty than other concentrations. The general trend of increasing current with increasing salt concentration, and therefore increasing integrated charge, was similar among all salt types analyzed.

Figure 4.

(A) Chronoamperometry of f-ICP stacking of methylthioninium dye in variable concentrations of NaCl with 30 mM BGE at 240 V; (B) Integrated charge after 6 min with respect to an increasing concentration of NaCl included in the matrix.

The experimental integrated charge and the initial and concentrated stacking times for each 50 mM salt matrix, alongside relevant trends for each matrix across concentrations, are shown in Table 2. When the salt identity and concentration are known, initial stacking time can be approximately predicted based on the integrated charge (and therefore the current measurement). The same can be said for concentrated stacking time as well, except for in NaF and Na₂CO₃ matrices, both of which exhibited poor linearity and slopes indifferentiable from zero. There was not a strong correlation between the integrated charge and the stacking time across all salt types, however, indicating that current measurement cannot be used to predict approximate stacking time when sample salt compositions are unknown or variable. However, strong trends were observed between some intrinsic properties specific to salt types (conductivity, hydrodynamic radii, mobility, diffusion coefficients) and stacking times (see the supplementary materials section, Chronometric Trends, for specific information regarding these trends).

Table 2.

Integrated charge in select 50 mM salt matrices and stacking times for f-ICP performed in the same matrix, presented as 95% CI’s, alongside the results of weighted linear regressions of integrated charge, relative to stacking time in variable concentration salt matrices. The overall trend reported is representative of the 50 mM variable salt matrix only. MgCl₂ lacks regression because stacking did not occur above 50 mM with this salt.

Chronological and Chronoamperometric Results
Matrix Contributor		Band Formation	Stacking Time (sec)	Integrated Charge (mC)	Integrated Charge vs. Stacking Time Trend
					Regression Equation	R2	p-value	Slope Sig.
Control		Initial	32 ± 15	0 (Baseline)	-	-	-	-
		Concentrated	122 ± 25		-	-	-	-
NaCl		Initial	121 ± 8	872 ± 74	y = 0.077x + 37.9	0.908	0.0007	Yes
		Concentrated	227 ± 50		y = 0.136x + 105.2	0.981	0.0037	Yes
KCl		Initial	133 ± 32	1093 ± 114	y = 0.446x − 354.0	0.991	0.0066	Yes
		Concentrated	277 ± 32		y = 0.195x + 65.0	0.982	0.0459	Yes
NaF		Initial	62 ± 67	1016 ± 177	y = 0.060x − 6.7	0.803	0.0127	Yes
		Concentrated	255 ± 67		y = 0.08x +160.3	0.319	0.3501	No
MgCl2		Initial	167 ± 31	1136 ± 93	-	-	-	-
		Concentrated	335 ± 31		-	-	-	-
Na2CO3		Initial	74 ± 36	629 ± 289	y = 0.109x − 0.4	0.743	0.0384	Yes
		Concentrated	258 ± 36		y = −0.02x + 264.8	0.042	0.3858	No
Overall		Initial	-	-	y = 0.014x + 92.5	0.001	0.4017	No
		Concentrated	-		y = −0.183x +386.2	0.413	0.0079	Yes

ONLINE RESULTS

Representative uPSI and f-ICP/PSI Data

Representative total ion chronograms (TIC’s) for 10 μL of 1 ppm each methylthioninium and the four drug or drug metabolites, as well as creatinine, a primary matrix contributor, run in artificial urine can be seen in Figure 5 (A) and (B) for uPSI and f-ICP/PSI, respectively. The uPSI chronograms are cropped to 1 minute, but a continuous signal from all analytes can be observed over several minutes, as is typical for paper spray. Paper spray signal will eventually decrease as analyte is depleted and will stop altogether when the spray solvent is consumed. Representative extracted ion chronograms (EIC’s) for each method, also in MP-AU, can be seen in (C) and (D). Methylthioninium, the four drug or drug metabolite analytes, and creatinine all largely co-eluted in the f-ICP/PSI method. The increased analyte signal at 9 – 10.5 minutes in (B) and (D) correspond to when the stacked band reaches the tip-junction and is sprayed into the mass spectrometer. Signal increased well-over ten-fold relative to uPSI (discussed in detail below) due to the increased local analyte concentration present in that band.

Figure 5.

(A), A representative uPSI TIC and (C), EIC’s, for creatinine (brown) and 1 ppm each methylthioninium (blue), methamphetamine (red), norfentanyl (green), norhydrocodone (yellow), and N-desmethyl-cis-tramadol (pink), with the corresponding (E), full MS spectrum, in 100% concentrated MP-AU, alongside (B), (D), and (F), the same, run via the f-ICP/PSI methodology.

Exemplar full MS spectra for uPSI (E) and f-ICP/PSI (F), corresponding to the chronograms can also be seen in Figure 5. Those spectra and the resulting abundances used for later numerical comparisons averaged the regions indicated in (A) and (B). The electrokinetic method shows comparatively higher relative abundances for each of the five analytes and an increased signal-to-noise ratio. The analyte peaks for f-ICP/PSI are no longer found only in the baseline due to the signal enhancing preconcentration effect of f-ICP. For some sample matrices, there is also a decrease in ion suppression for f-ICP/PSI due to separation of analytes from matrix salts.

Salt Solution Analysis

f-ICP/PSI and uPSI were compared for the analysis of the four-drug mixture in variable concentrations of NaCl and between each of the previously analyzed salts at 10 mM. Two criteria were assessed: signal enhancement of f-ICP/PSI relative to uPSI and ion suppression.

Ion suppression for uPSI and f-ICP/PSI is shown in Figure 6 for NaCl-containing samples. As expected, most analytes showed a large amount of ion suppression in the presence of NaCl when analyzed by uPSI (p < 0.04, |g| > 1.7). Only N-desmethyl-cis-tramadol in the 25 mM NaCl matrix and methamphetamine in the 10 mM matrix failed to exhibit significant suppression, likely due to relatively large measurement uncertainty (p = 0.0600, 0.0582, respectively). Each drug or metabolite showed large increases in ion suppression with increasing NaCl concentration in uPSI (β* > 0.87); only methylthioninium failed to show the same response (p = 0.0910). As expected, increasing salt concentrations cause higher ion suppression for small molecule drugs in uPSI.

Figure 6.

Ion Suppression in (A) uPSI and (B) f-ICP with increasing NaCl concentrations in the sample matrix. The regression equations are shown for each analyte (middle), with significances of the slope. (C) Change in ion suppression between methods; more negative (upward) values indicate less ion suppression, an improvement in the electrokinetic method. Δm refers to the change in slope for ion suppression vs NaCl concentration between the two methods, with significances.

Compared to uPSI, f-ICP/PSI showed an improvement in ion suppression in most cases for NaCl matrices. Two of the four drugs/metabolites and methylthionium exhibited significantly lower ion suppressions, as seen in Figure 6 (C). Norfentanyl and n-desmethyl-cis-tramadol were the exceptions; there were no significant differences in ion suppression for these two drug metabolites between uPSI and f-ICP/PSI (p >0.45, >0.05, respectively). Another noteworthy difference was that ion suppression did not increase with concentration as much for the electrokinetic method as it did with uPSI for N-desmethyl-cis-tramadol (p = 0.0021). Moreover, only methamphetamine exhibited a significant slope (p = 0.0084) for ion suppression versus NaCl concentration in the electrokinetic method. Taken together, these findings indicate that f-ICP/PSI tolerates high salt concentrations better than uPSI.

Figure 7 (A) compares the change in ion suppression between uPSI and f-ICP/PSI for 4 additional salts (KCl, NaF, Na₂CO₃, and MgCl₂) at 10 mM for the same set of compounds. Contrasting with what was observed with NaCl, the other salt matrices showed significant worsening of ion suppression when going to the electrokinetic methodology. Lack of significance in most samples can be attributed to large measurement uncertainty owing to small sample size. Exceptions to this are in the dye indicator only, where methylthioninium showed a decrease in ion suppression relative to uPSI in all matrices (p < 0.045; |g| > 2.6) but KCl (p = 0.9325). Ion suppression in f-ICP/PSI improved at higher NaCl concentrations (Figure 6 (C)); it may be that improvements are only observed at higher salt concentrations overall, but the inhibition of stacking at these higher concentrations for some salts make such comparisons null.

Figure 7.

(A) Change in ion suppression observed in the presence of 10 mM salts and concentrated artificial urine when undergoing f-ICP enrichment, relative to uPSI. (B) Signal Enhancement between methods for the same. Note that for change in ion suppression, more negative (upward) values indicate less suppression, so an improvement for the electrokinetic method.

Ion suppression in uPSI was lower with NaCl ($\left|\stackrel{\_}{\mathrm{I}\mathrm{S}}\right|=22\%$) than in any other salt matrix for all analytes ($\left|\stackrel{\_}{\mathrm{\delta }\mathrm{g}}\right|=5.70$; p < 0.02). Analytes generally exhibited the worst ion suppression in Na₂CO₃ ($\left|\stackrel{\_}{\mathrm{I}\mathrm{S}}\right|=77\%$; $\left|\stackrel{\_}{\mathrm{\delta }\mathrm{g}}\right|=15.84$; p < 0.044), except norfentanyl and methylthioninium where the difference was not always significant (p > 0.07). In KCl ($\left|\stackrel{\_}{\mathrm{I}\mathrm{S}}\right|=40\%$), analytes generally performed similarly to those in MgCl₂ ($\left|\stackrel{\_}{\mathrm{I}\mathrm{S}}\right|=44\%$), where ion suppression was less than in NaF ($\left|\stackrel{\_}{\mathrm{I}\mathrm{S}}\right|=61\%$). Ion suppression in uPSI for small molecule drugs, among the five salts analyzed, ranked from least to most is: NaCl ≪ KCl ≲ MgCl₂ ≪ NaF ≪ Na₂CO₃. A similar trend was observed in the electrokinetic method when excluding the dye indicator; the dye behaved aberrantly in NaF where it alone exhibited a large enhancement (p = 0.0004; |g| = 5.672). The low ion suppression for NaCl is beneficial for toxicological analyses since it is the main electrolyte in biofluids.

The signal enhancements for f-ICP/PSI relative to uPSI for a variety of salt conditions are shown in Figure 7 (B). Signal improved 1.8 – 223.7×, analyte and matrix dependent. The increased intensity in f-ICP/PSI relative to uPSI was significant in all cases (p < 0.005, |g| > 0.9). While signal enhancement was independent of NaCl concentration for three of the four drugs or metabolites and the dye indicator, norhydrocodone had a significant positive slope for signal enhancement vs. NaCl concentration (p = 0.0487; β* = 0.89). For some analytes, therefore, greater signal enhancement relative to uPSI is observed in higher concentration matrix conditions. The intercepts for all four drug analytes were also significantly greater than zero (p < 0.03), suggesting that the presence of a small amount of salt increases signal enhancement relative to salt-free samples for small molecule drugs (0 mM salts were excluded from all regressions).

For the 10 mM inter-matrix comparison, the greatest average improvements were seen in KCl ($\stackrel{\_}{\mathrm{S}\mathrm{E}}=57\times$), followed by NaCl ($\stackrel{\_}{\mathrm{S}\mathrm{E}}=44\times$), Na₂CO₃ ($\stackrel{\_}{\mathrm{S}\mathrm{E}}=40\times$), NaF ($\stackrel{\_}{\mathrm{S}\mathrm{E}}=29\times$), and MgCl₂ ($\stackrel{\_}{\mathrm{S}\mathrm{E}}=13\times$). As a reference, the salt-free control had an average signal enhancement of 36×.

Multipurpose Artificial Urine (MP-AU) Analysis

The performance of f-ICP/PSI and uPSI was compared in artificial urine at four different matrix dilutions and can be seen in Figure 8. Both uPSI (p < 0.004; |g| > 4.2) and f-ICP/PSI (p < 0.041; |g| > 5.7) exhibited large ion suppression at all analyzed urine concentrations. For uPSI, there were significant positive slopes for ion suppression with increasing MP-AU concentration for methamphetamine, norhydrocodone, and methylthioninium (p = 0.0444, 0.0408, 0.0326). For f-ICP, on the other hand, no analyte showed significant changes in ion suppression with increasing matrix concentration (p > 0.3). For both uPSI and f-ICP/PSI, no analyte exhibited a significantly different trend than the others.

Figure 8.

Ion Suppression in (A) uPSI and (B) f-ICP/PSI between MP-AU dilutions. The regression equations are shown for each analyte (middle), with significances of the slope. (C) Change in ion suppression between methods; more negative (upward) values indicate less ion suppression, so an improvement, in the electrokinetic method. Δm refers to the change in slope for ion suppression vs MP-AU concentration between the two methods, with significances.

Unlike with the NaCl matrices discussed in the previous section, f-ICP/PSI did not show a notable improvement in ion suppression compared to uPSI in artificial urine. Indeed, there was a significant increase in ion suppression observed in 8 of the 20 cases, compared to uPSI (p < 0.0115, |g| = 4.135). We hypothesize that this worsening of ion suppression in the f-ICP method is due to the presence of creatinine in the urine. Creatinine, as seen in Figure 5, was a primary matrix contributor, present at concentrations of up to 7.8 mM (at 100% MP-AU concentration). It frequently co-stacked with the drug analytes, as demonstrated by the stacking chronograms, and was over 1000-fold more concentrated than the drugs. The change in ion suppression between the two methods was independent of MP-AU concentration for all drug or drug metabolites tested. A difference was observed only in the dye indicator for 25% compared to 75% MP-AU (p = 0.0493, $\left|\mathrm{\delta }\text{g}\right|=5.199$).

Despite the ion suppression caused by the co-stacking of creatinine and other components with drug analytes, f-ICP/PSI still showed large increases in signal intensities compared to uPSI in all analytes (p < 0.0065, |g| > 3.8); for 100% concentrated urine, f-ICP/PSI signal intensity was 7.8 – 53.1× higher than uPSI, compared to urine-free controls where abundances were 7.8 – 140.5× higher. These results can be visualized in Figure 7 (B). No significant slopes were observed in the regression of signal enhancement vs. urine concentration (p > 0.22), although all drug analytes did exhibit a significant intercept (p < 0.05). This finding lends credence to the earlier hypothesis that f-ICP/PSI, in the presence of very low concentrations of salts, may yield some additional enhancement to analyte signal relative to a neat solution (0% MP-AU was excluded from all regressions).

CONCLUSION

By combining electrokinetic manipulations of analytes directly on the spray substrate with PSI-MS, desalting and preconcentration of drug analytes directly on the substrate improved analytical performance. Some of paper spray’s largest limitations, namely small sample volumes, sometimes inadequate LOD’s, and matrix effects, can largely be overcome, with 1.8× – >200× enhanced signals and an improvement in ion suppression in single-salt matrices of up to ~50%. The method did show an increase in ion suppression in urine due to costacking of analytes such as creatinine with ion suppressing agents. However, large enhancements in analyte signal were still achieved owing to the preconcentration.

While these improvements are significant, the increase in analysis time and complexity relative to unstacked paper spray are understood limitations. Nevertheless, the method remains simpler than traditional methods; the required voltage is much lower and the MS interface is much simpler than traditional CE-MS. There are no high-pressure pumps, chromatography column, or plumbing connections as in HPLC-MS, and the solvent consumption is on the order of 10-fold higher compared to f-ICP/PS. Future work will seek to improve f-ICP/PS throughput using multi-channel stacking devices where multiple samples are stacked in parallel. Multichannel stacking is simple relative to multichannel HPLC because the stacking voltage is merely applied to multiple parallel paper strips positioned in an expanded cartridge. In multichannel HPLC, on the other hand, multiple columns and independent fluid paths with their own mixers are required. This innovation, combined with mass production of the f-ICP/PSI device, could enable relatively simple and high-throughput drug screening from complex matrices.

Supplementary Material

The following files are available free of charge.

CAD Files of:

• uPSI Cartridge

• f-ICP/PSI Cartridge Lower Part

• f-ICP/PSI Cartridge Upper Part

• Cartridge Stage Bracket/Mount

PDF File with:

• Information on Device Printing, Construction, and Wiring

• Wiring Schematic for PSU

• Mechanism of Substrate Modification

• Data for Optimization of Substrate Choice

• Additional Information on Preparation of MP-AU

• Further Procedural Information Regarding Spotting of Analyte and BGE Solutions

• Additional Equations for Statistical Analyses

• Calculated Values of Intrinsic Properties of Salts

• Comparison of Data Generated on Offline and Online Devices

• Chronoamperometry in Additional Single-Salt Matrices

• uPSI Ion Suppression Data in Additional Single-Salt Matrices

• Additional Chronometric, Chronoamperometric, and Spectrometric Trends Observed

• MS² Spectra for all Analytes and Primary Matrix Interferants

ABBREVIATIONS

AUC

Area Under Curve

BGE

Background Electrolyte

EOF

Electroosmotic Flow

EP

Electrophoresis

FASS

Field-Amplified Sample Stacking

f-ICP

Faradaic Ion Concentration Polarization

IDZ

Ion Depletion Zone

IS

Ion Suppression

LOD

Limit of Detection

MP-AU

Multipurpose Artificial Urine

OH-PTFE

Hydroxylated Polytetrafluoroethylene

PSI

Paper Spray Ionization

PSU

Power Supply Unit

S_A

Anodic Substrate

S_C

Cathodic Substrate

S_D

Deposition Substrate

SE

Signal Enhancement

uPSI

Unstacked Paper Spray Ionization