Automated, High-Throughput Analysis of Tire-Derived p-Phenylenediamine Quinones (PPDQs) in Water by Online Membrane Sampling Coupled to MS/MS

Abstract

The tire-derived contaminant N-(1,3-dimethylbutyl)-N′-phenyl-p-phenylenediamine quinone (6-PPDQ) was recently identified as a potent toxin to coho salmon (Oncorhynchus kisutch). Studies investigating 6-PPDQ have employed solid-phase extraction (SPE) or liquid–liquid extraction (LLE) with liquid chromatography–mass spectrometry (LC-MS), providing excellent sensitivity and selectivity. However, cleanup and pre-enrichment steps (SPE/LLE) followed by chromatographic separation can be time- and cost-intensive, limiting sample throughput. The ubiquitous distribution of 6-PPDQ necessitates numerous measurements to identify hotspots for targeted mitigation. We recently developed condensed phase membrane introduction mass spectrometry (CP-MIMS) for rapid 6-PPDQ analysis (2.5 min/sample), with a simple workflow and low limit of detection (8 ng/L). Here, we describe improved quantitation using isotopically labeled internal standards and inclusion of a suite of PPDQ analogues. A low-cost autosampler and data processing software were developed from a three-dimensional (3D) printer and Matlab to fully realize the high-throughput capabilities of CP-MIMS. Cross-validation with a commercial LC-MS method for 10 surface waters provides excellent agreement (slope: 1.01; R² = 0.992). We employ this analytical approach to probe fundamental questions regarding sample stability and sorption of 6-PPDQ under lab-controlled conditions. Further, the results for 192 surface water samples provide the first spatiotemporal characterization of PPDQs on Vancouver Island and the lower mainland of British Columbia.

Short abstract

Tire-derived contaminants are widely distributed and threaten aquatic species. Membrane introduction mass spectrometry provides a sensitive, high-throughput approach to assess their fate and distribution for targeted mitigation.

Introduction

Urban runoff represents an ongoing threat to aquatic life. As rainwater washes over anthropogenic surfaces, both inorganic and organic contaminants are released and carried to receiving waterways. In 2020, Tian et al. published a landmark study identifying N-(1,3-dimethylbutyl)-N′-phenyl-p-phenylenediamine quinone (6-PPDQ) as a potent toxin to coho salmon (Oncorhynchus kisutch) in the ng/L range.^1,2 The parent compound, N-(1,3-dimethylbutyl)-N′-phenyl-p-phenylenediamine (6-PPD), is added to tire rubber at 0.4–2.0% by mass as an antiozonant. 6-PPD reacts with ground-level ozone to produce 6-PPDQ and other oxidized byproducts.³ Since the discovery of 6-PPDQ, acute toxicity at environmentally relevant (ng to μg/L) concentrations has been established for several species including white-spotted char (Salvelinus leucomaenis), rainbow trout (Oncorhynchus mykiss), brook trout (Salvelinus fontinalus), and early-life stage Chinook (Oncorhynchus tshawytscha).⁴⁻⁶ Further, several groups have identified sublethal effects of 6-PPDQ (intestinal, neural)^7,8 across multiple species including Caenorhabditis elegans, mice,⁹ and fathead minnow (Pimephales promelas).¹⁰ While the mechanism of 6-PPDQ toxicity is still being uncovered, Mahoney et al. recently reported in vitro evidence that 6-PPDQ toxicity may be mediated by disruption of mitochondrial respiration¹¹ and Blair et al. have observed disruption of the blood–brain barrier in coho salmon treated with roadway runoff.¹²

Several targeted studies have identified 6-PPDQ in various matrices beyond storm- and surface water, including fish tissue,¹³ snow,³ urine,¹⁴ air particulates,^15,16 and sediments¹⁵ suggesting that this contaminant is widely distributed in the environment. Additionally, 6-PPD is only one compound from a family of para-phenylenediamine (PPD) preservatives. Several of the corresponding quinone analogues (Figure S1) have been identified in the environment, with unknown implications for human and environmental health.¹⁵

Most investigations to date have employed solid-phase extraction (SPE) or liquid–liquid extraction (LLE) coupled to liquid chromatography–tandem mass spectrometry (LC-MS/MS).^2,3,6,15 This approach can provide excellent sensitivity and selectivity, and was especially valuable in the initial discovery of 6-PPDQ as the acute toxin responsible for “urban runoff mortality syndrome.”¹ However, the cleanup and pre-enrichment steps (SPE/LLE) can be cost- and labor-intensive and coupled with chromatographic separation time, limit sample throughput. Given the diffuse and ubiquitous nature of 6-PPDQ sources, large sample sets need to be analyzed to characterize regional inputs and identify hotspots for targeted mitigation.¹⁷⁻¹⁹ Therefore, direct mass spectrometry approaches that obviate sample cleanup and pre-concentration steps provide a faster, “fit-for-purpose” approach for analyzing 6-PPDQ and analogues in storm- and surface water.

CP-MIMS interfaces liquid or slurry samples directly with a mass spectrometer (MS) via a semipermeable membrane.^20,21 Analytes that permeate the membrane are carried to the MS ion source by a continuous flow of organic solvent acceptor phase through the lumen of a capillary hollow-fiber membrane immersed in the sample. Generally, a poly-dimethylsiloxane membrane is employed, which allows permeation of neutral, hydrophobic analytes (e.g., 6-PPDQ log K_ow ∼4)¹⁵ while rejecting bulk matrix components and particulate matter. The acceptor phase composition, ion source, and MS parameters can be optimized for a specific analyte class. This approach has been implemented for polycyclic aromatic hydrocarbons,²²⁻²⁴ phthalates,²³ UV filters,²⁵ and naphthenic acids.^22,26,27 CP-MIMS provides continuous, real-time data allowing dynamic processes such as organic synthesis²⁸ or sorption/desorption phenomena²⁷ to be studied directly. We have previously reported a preliminary CP-MIMS method using tandem MS for analysis of 6-PPDQ exhibiting environmentally relevant detection limits (8 ng/L) and a short analytical duty cycle (2.5 min/sample).²⁹

Here, we describe significant improvements and validation for 6-PPDQ analysis with CP-MIMS. A low-cost 3D printer was repurposed and reprogrammed for use as an autosampler (Figure 1) and accompanying software for automated data processing was developed in Matlab. Isotopically labeled internal standards were used to improve quantitative performance, and the technique is expanded to include five additional PPD-quinone analogues (Figure S1). Cross-validation with a commercial LC-MS method exhibited excellent quantitative agreement in a series of real-world samples. We apply the improved CP-MIMS method to probe fundamental questions about the aqueous stability of 6-PPDQ as well as sorption behavior on natural sediments. Further, 192 water samples collected as part of a citizen science campaign were rapidly screened with the presented approach to provide the first spatiotemporal survey of 6-PPDQ and analogues on Vancouver Island and the lower mainland of British Columbia, Canada.

Figure 1.

Instrumental schematic illustrating the CP-MIMS probe, 3D printer-based autosampler, and MS. The chronogram shows automatic peak detection for standards (red), samples (green), and calibration verification standards (blue). Normalized signal intensity for the sample-phase internal standard is shown in gray.

Materials and Methods

Standards and Samples

Analytical-grade 6-PPDQ and ¹³C₆-6-PPDQ (both ≥95%; 100 μg/mL in acetonitrile) were purchased from ACP Chemicals (Montreal, QC). Neat standards of DTDPQ, DPPDQ, IPPDQ, CPPDQ, 77PDQ (10 mg each), and 100 μg/mL 6-PPDQ-d₅ (in acetonitrile) were purchased from HPC Chemicals (Atlanta, GA) at ≥92% purity. Full analyte names and structures are available in Figure S1. Combined substocks of the suite of analytes were prepared gravimetrically in HPLC-grade methanol (Fisher Scientific, Ottawa, ON) at ca. 15–30 μg/kg. Calibration standards were prepared between 8 and 1600 ng/L by spiking this combined standard into deionized water (Facility scale reverse Osmosis/Ion Exchange Water Purification System, Applied Membranes, Inc., Vista, CA).

Water samples (n = 192 total; 24 road runoff samples and 168 stream water samples) were collected at sites on Vancouver Island and in Vancouver, British Columbia, Canada. Samples were collected during October–November 2022 as part of a citizen science campaign led by the British Columbia Conservation Foundation and Fisheries and Oceans Canada’s Whale Contaminants Program. Where possible, samples were collected before, during, and after a rain event (≥5 mm) following a dry period (≥48 h). For “during” samples, volunteers were instructed to collect during estimated peak stream flow. Prior to sampling, bottles were thoroughly rinsed with tap water, then subsequently rinsed 3× with deionized water and finally with a small volume of HPLC-grade methanol. Environmental grab samples were collected by completely filling (i.e., headspace < 2% overall volume) amber glass bottles with PTFE-backed septa (40–1000 mL; Fisher Scientific) after rinsing three times with the sample. All samples were kept at 4 °C until they were picked up for transport in a cooler with ice packs (<1.5 days transit time). Given the limited stability of 6-PPDQ in water observed by others,^5,30 surface water samples were prioritized and analyzed as soon as possible, generally within 5 days for CP-MIMS experiments unless otherwise mentioned. Road runoff samples were held for up to 100 days prior to analysis.

Samples were measured “as is,” with no filtration or sample cleanup. Where high concentrations were anticipated (e.g., roadway runoff), samples were diluted 2- to 10-fold in deionized water to bring them into the working calibration range (ca. 8–250 ng/L) for general sample analysis. As needed, higher calibration standards were also included for storage time and sorption experiments. The linear dynamic range for CP-MIMS analysis of PPDQs extends to the μg/L range (up to 32 μg/L for 6-PPDQ).²⁷ Immediately prior to measurement, all samples and standards were spiked with ca. 100 ng/L 6-PPDQ-d₅ internal standard. Sample analysis was carried out in clean (as described above) 40 mL TOC vials, using both clear and amber glass vials as available.

Storage Time Experiments

For initial evaluation of aqueous stability (Figure 2A), 6-PPDQ was added at ca. 10 μg/L to deionized water and both a filtered (0.45 μm Nylon syringe filter, 33 mm, Fisher Scientific) and unfiltered aliquot of roadway runoff. These solutions were subsampled and diluted 40- to 100-fold in deionized water for manual CP-MIMS analysis using a 2- to 3-point standard addition methodology described previously.²⁹ This was repeated at several time points over the 2-week (340 h) test period. Samples were kept in clear, capped glass 40 mL vials with PTFE-backed septa at room temperature (ca. 22 °C; headspace 10–20% overall volume). Stability in real-world stream samples (Figure 2B) was evaluated opportunistically. Where sufficient sample was available (≥500 mL) and significant 6-PPDQ was detected during initial measurement (≥50 ng/L), fresh aliquots of the sample (35–40 mL) were withdrawn and measured over time to monitor analyte stability. The bulk solution was stored at 4 °C in amber glass bottles sealed with PTFE-backed septa between analyses. These were all measured as described below for general sample analysis using the autosampler and sample-phase internal standard (Figure S2C). Headspace in these containers increased over time as the sample was withdrawn, starting at <2% overall volume and reaching up to ∼50% by the final time point. For the uncapped experiments (Figure 2C,D), a 1 L aqueous solution containing all of the PPDQ analogues was prepared at ca. 1–5 μg/L and transferred into clear, 40 mL vials. The vials were kept uncapped for the indicated time up to 120 h at ambient temperature (22 °C) in a fume hood. Relative concentrations were then measured directly (no dilution) against a capped control (<5% headspace volume; sealed with PTFE-backed rubber septa) after the addition of internal standard. To mitigate losses during sample analysis, in this experiment, vials on the autosampler were kept capped until immediately prior to analysis. This workflow is illustrated in Figure S3. Relative quantification was carried out using the same sample-phase internal standard calibration model employed for real-world samples (Figure S2C).

Figure 2.

(A) Shelf life of 6-PPDQ in DI water (blue diamonds), unfiltered road runoff (green squares), and filtered road runoff (red circles) over a 2-week period. Minimal if any loss is observed throughout this window beyond the precision limits of the method (±15%). (B) Shelf life of 6-PPDQ in two real-world samples measured over a ca. 3.5-month period. Concentrations appear generally stable within the precision limits of the method for the first 400–500 h (ca. 3 weeks) followed by a slow decay during months-long storage. (C) Loss of 6-PPDQ (blue squares), CPPDQ (purple circles), and 77PDQ (gray diamonds) from uncapped solutions of 6-PPDQ over 120 h. Fitting the process to a first-order decay (D) yields t_1/2 ranging from 14 to 108 h for different PPDQ analogues. For analysis of samples shown in (C) and (D), the vials were kept capped until immediately before analysis to minimize atmospheric losses during the analytical run. The full dataset for (C) and (D) is available in Table S2; the t_1/2 values for the compounds not shown here were 108 h (DTPDQ), 69 h (DPPDQ), and 65 h (IPPDQ).

Sorption Experiments

A 1 L aqueous solution of 6-PPDQ at ca. 0.5–1 μg/L with 60 mM concentration phosphate buffer (pH = 7.16) was prepared. This solution was transferred into 40 mL vials, and low-organic-content sediment (clean loam soil; f_OC = 1.85%, Sigma-Aldrich, Oakville, ON), high-organic-content sediment (clean sandy loam; f_OC = 5.96%; Sigma-Aldrich), Ottawa sand (Chromatographic Specialties, Inc., Brockville, ON), and activated carbon (Euroglas, Delft, Holland) were added in the 75–1500 mg range to individual, clear 40 mL vials (equivalent to 3–60 g/L loading). These solutions were capped and allowed to equilibrate for 2 days under gentle shaking (200 rpm, multi-platform shaker table, Fisher Scientific) at room temperature. Manual CP-MIMS analysis was then performed directly on the heterogeneous sediment slurry using the acceptor phase internal standard calibration method (Figure S2B). The fraction of 6-PPDQ remaining in solution was calculated against a control vial that underwent the same equilibration time and analysis method, but with no sorbent added. To calculate the organic carbon binding constant log(K_OC), pairwise calculations between the control vial and each different loading of sediment (n = 5) were calculated as: K_OC = [6-PPDQ]_sediment/([6-PPDQ]_water·f_OC), where K_OC is the binding constant (L/kg organic carbon), [6-PPDQ]_sediment is the concentration of 6-PPDQ on the sediment in ng/kg, [6-PPDQ]_water is the aqueous concentration (ng/L), and f_OC is the mass fraction of organic carbon in the sediment. The results from the five different loadings were then averaged for each sediment and transformed into logarithm space. The uncertainty associated with the resulting log(K_OC) values was then increased to reflect the potential for nonequilibrium conditions (only 2-day equilibration time) and relatively low levels of sorption for the low-f_OC sediment.

Condensed Phase Membrane Introduction Mass Spectrometry

The CP-MIMS apparatus has been described in detail previously.²⁹ Briefly, a 7.6 cm length of 55 μm thick PDMS hollow-fiber membrane (Permselect, MedArray, Inc., Ann Arbor, MI) was mounted onto 31-gauge stainless steel capillaries (Microgroup, Medway, MA). A stainless steel wire support was epoxy-potted into 1/4″ stainless steel tubing alongside the capillaries to improve probe ruggedness. The acceptor phase was composed of 15/85/0.03 heptane/methanol/formic acid (v/v/v; Fisher Scientific) and was flowed through the membrane lumen at 10 μL/min using a syringe pump (Chemyx Fusion 100, Stafford, TX) and 10 mL gas-tight syringe (Hamilton 1000 series, Fisher Scientific). Due to the limited stability of formic acid in methanol,³¹ fresh acceptor phase was prepared for each sampling batch (2–3 days of analysis) unless otherwise mentioned. To monitor ionization suppression and instrument drift, 20 μg/kg ¹³C₆-6-PPDQ internal standard was also included in the acceptor phase. Membrane sampling was carried out under ambient laboratory conditions (ca. 22 °C; 1 atm). For CP-MIMS, the system should be refreshed periodically with clean solvent flushes and/or membrane replacements, particularly after complex or high-concentration samples (e.g., roadway runoff). The membranes employed here exhibited stable analytical performance for weeks to months (hundreds of samples) and were generally only replaced as part of a preemptive cleaning routine or due to operator error during setup (e.g., membrane inadvertently dislodged).

Detection was carried out with an electrospray ionization (ESI) triple quadrupole mass spectrometer (QSight 220, PerkinElmer, Waltham, MA) in positive-ion mode. Global instrument parameters include: nebulization gas = 120 psi, hot-surface induced desolvation (HSAID) source = 320 °C, and capillary voltage = +4.0 kV. For MS/MS monitoring of 6-PPDQ, atypical qualifier/quantifier MS/MS ion ratios were observed using the dominant m/z 299 → 215 transition. This may be due to co-occurring isomeric ozonation products of 6-PPD recently characterized by Zhao et al.³² Instead, 6-PPDQ was monitored using m/z 299 → 256 as quantifier ion and both 299 → 241 and 299 → 100 as qualifier ions. MS/MS scan parameters for quantifier and qualifier ion transitions of other PPDQs are provided in Table S1. A dwell time of 1000 ms was used for each MS/MS transition. Additionally, full-scan mass spectra were collected between m/z 100 and 500 with step size 1 m/z and 3 ms per step.

Autosampler Construction and Operation

The autosampler was constructed from a base-model 3D printer (Ender 3 V2, Creality, Shenzhen, China). Conversion to a CP-MIMS autosampler was performed by removal of the print nozzle and fan. A 1/4″ hole was drilled through the aluminum cooling housing, allowing the CP-MIMS probe to slide in and be secured using the existing set screw. This places the CP-MIMS probe at approximately the same XY coordinates as the print nozzle would be, providing 220 mm × 220 mm area of programmable XY motion. The Z-stop sensor was moved up such that the bottom of the probe would not contact the print bed. Photos of the cooling housing, raised Z-stop sensor, and overall setup are shown in Figure S4. A 6 × 6 sample tray was 3D-printed with polylactic acid (PLA) filament (Creality) to cover the print bed surface and accommodate 40 mL TOC vials (1 1/8″ holes). The sample tray was secured to the print bed with tape for early experiments, and later with a 3D-printed base, which held the sample tray in place. These prints were made using a larger working area 3D printer (CR-10 V3, Creality); however, in principle, the trays could be printed on the smaller Ender 3 by printing the trays vertically rather than laid flat. The XY coordinates for the four corner sample slots were measured manually and the remaining slots were interpolated from these coordinates.

To control the autosampler, custom Matlab script (version 2021b; MathWorks, Natick, MA) was written which takes a user-defined runlist, coordinate map, sampling time, and rinse time and writes G-code for a given run. For routine sample analysis, only full trays were measured, containing one rinse solution (HPLC-grade methanol), 5 calibration standards, 25 samples, and 5 calibration verification standards. The system was programmed to give sample/standard membrane exposure for 2.5 min, stirring the solution by moving the probe in a square motion about the center of the vial. A 1 min rinse in the methanol vial with the same stirring motion was programmed to start the run and after each sample/standard analysis. Overall, this gave a total measurement time of ca. 4 min/sample. A full calibration was performed at the start and end of each run (Figure 1). Additional QA/QC samples included a calibration standard and deionized water blank interspersed after every batch of 5 samples. This process includes a total of 44 analyses and takes ca. 3 h to run, allowing 3–4 trays to be processed per day (i.e., 132–176 total analyses).

Data Processing

Matlab scripts were developed to import a raw data file and sample info file and output quantitative results. A flow diagram outlining the overall process is available in Figure S5. Briefly, a CSV file was exported from the MS vendor software (Simplicity 3Q, version 1.4.1806.29651; PerkinElmer, Waltham, MA) and imported into Matlab alongside an info file with sample list and calibrant concentrations. Peaks were picked and identified from the raw data based on the donor phase internal standard peak (6-PPDQ-d₅) and specified runlist. Calibration models were then constructed using the maximum signal intensity after smoothing in each sample/standard window normalized to that of the internal standard in the same window. The intra-run limit of detection (LoD) was calculated using the blank calibration standard (n = 6) and lowest calibration standard exhibiting a signal-to-noise ratio ≥3 (n = 2). A more robust evaluation of the LoD and limit of quantitation (LoQ) was also performed using n = 7 analyses of a blank, 25 ng/L, and 60 ng/L combined standard of the PPDQ suite. The intra-run LoD was used as reporting limit for environmental samples as it accounts for day-to-day variability in performance. Data is flagged for a possible interferent if the relative abundance of the qualifier ion transitions deviates significantly from expected values. The sample flag threshold for qualifier ion ratios was set at three standard deviations observed for calibration check solutions, applied to samples that exhibit 6-PPDQ concentrations ≥50 ng/L (i.e., above LoQ).

Commercial Analysis with Liquid Chromatography–Mass Spectrometry

Water samples (n = 12) collected from the Nanaimo area on Vancouver Island were sent to SGS AXYS Analytical Ltd. (Sydney, BC) within 48 h of sampling for commercial analysis of aqueous 6-PPDQ. Samples were preserved with dichloromethane upon arrival and held at 4 °C in the dark for 23–24 days until analysis. The LoD was 0.05 ng/L in a sample size of 1 L. Spike recovery experiments at 80 ng/L 6-PPDQ for a blank stream sample (n = 5) yielded 103 ± 1% recovery. Analysis was carried out according to the methods outlined in Lo et al.⁶ with minor modifications. Briefly, 6-PPDQ-d₅ was added to aqueous samples prior to liquid–liquid extraction (LLE) with dichloromethane. The extracts were then analyzed on a Waters ACQUITY UPLC I-Class System and a Xevo TQ-S tandem mass spectrometer operated in positive-ion mode. A Waters ACQUITY UPLC BEH C18 (1.7 μm, 2.1 mm × 50 mm) column was used for analytical separation, with an ACQUITY UPLC BEH C18 Vanguard (1.7 μm, 2.1 mm × 5 mm) employed as guard column. Gradient elution was performed with UPLC-grade water containing 0.1% formic acid (solvent A) and 1:1 acetonitrile/methanol (solvent B). The gradient started at 70% A and was held for 1 min, and then increased to 100% B by 10 min. This mobile phase composition was maintained for 2 min and then returned to original conditions over another minute for a total time of 13 min. The column was then held at these conditions for another minute to equilibrate before the next injection. 6-PPDQ was monitored using two MRM transitions at m/z 299 → 241 (quantifier ion) and 299 → 215 (qualifier ion).

Results and Discussion

Development of a Low-Cost Autosampler for CP-MIMS Analysis of PPDQs

Rapid analysis techniques offer significant advantages, including higher sample throughput and reduced cost. However, as the sample analysis time decreases, manual operation becomes impractical. In anticipation of large sample sets collected during multiple rain events, the need to automate CP-MIMS became apparent. Early efforts in our lab employed a rotary tray autosampler from a flow injection analyzer.³³ While reasonably effective, this approach was limited by the lack of full user control (e.g., limited sampling time), inflexible geometry (e.g., sample bottle types), and did not accommodate stirring. To leverage the high sample throughput of the technique, we have developed a custom autosampler from a repurposed 3D printer, enabling fully automated data acquisition and processing. Others have used this approach for similar challenges including spatially resolved mapping of uneven surfaces for mass spectrometry imaging³⁴ and nucleic acid isolation/amplification.³⁵ The printer employed here (Creality Ender 3) met our basic criteria of being low cost (<$500 CAD) and supporting precise motion (±0.1 mm). Further, the heated print bed lends itself to future temperature-controlled experiments to probe physicochemical properties and/or improve analytical sensitivity.^36,37 With relatively minor physical modification of the printer (replacement of print nozzle with CP-MIMS immersion probe, reposition Z “zero” position; Figure S4) and a simple Matlab program to translate user inputs into printer instructions, a functional autosampler was constructed for CP-MIMS (Figure 1).

A typical signal chronogram is shown in Figure 1. The system demonstrated stable operation for 3 h, and we have conducted up to 4 of these runs in a single day (100 samples, 176 total analyses after incorporation of calibrators, QA/QC, and blank samples). Sample analysis is bracketed by a full calibration suite at the beginning and end of the run, with a calibration check and DI blank measured after every 5 samples. Currently, the system is operated with a ca. 4 min duty cycle consisting of 2.5 min in a sample solution, 1 min in a methanol rinse, and roughly 30 s to transition between vials. This can be reduced to as little as 2.5 min/sample for steady-state membrane transport measurements or even faster using a non-steady-state approach.^38,39 Beyond the eased burden on the operator, an additional advantage of automating data acquisition is that signal peaks are more reproducibly structured (width, distance) facilitating automated data analysis. The colored peaks in Figure 1 are identified and processed automatically based on the specified acquisition order and time stamp of the internal standard peak. The combination of automatic data acquisition and processing has been transformative, with same- or next-day reporting possible for the large number of samples associated with each rain event. This is particularly important for community-based projects where engagement with citizen scientists and other project partners is crucial. This basic platform can be expanded to any membrane permeable analyte, allowing for a simple, high-throughput, and automated workflow for the analysis of a wide range of trace organic contaminants.

Adaptation to PPDQ Analogues and Laboratory Experiments

Electrospray ionization (ESI) in positive-ion mode is a particularly sensitive technique for primary and secondary amines,⁴⁰ including PPDQs. However, it can be prone to ion suppression/enhancement effects due to both the sample matrix and long-term signal drift. While membrane extraction largely excludes matrix components that can affect ionization, signal drift across long run times necessitates use of an internal standard to compensate.³⁹ During our previous study, isotopically labeled standards of 6-PPDQ were not commercially available, so quantitation relied on a standard addition methodology to compensate for the above effects.²⁹ In this work, 6-PPDQ-d₅ and ¹³C₆-6-PPDQ were included in the aqueous sample phase and methanolic acceptor phase, respectively. Normalizing signal response to either internal standard improves calibration stability (Figure S2); however, the sample-phase internal standard was favored where possible as it captures both sample partitioning and ionization effects.

Given the similar structural features of the PPDQ analogues (Figure S1), their physicochemical properties¹⁵ are favorable for membrane transport through poly(dimethylsiloxane) (low pK_a of the conjugate acid, high K_ow). After initial MS/MS optimization (Table S1), the limit of detection (LoD) was assessed by replicate analyses (n = 7) of a blank, 25 ng/L, and 60 ng/L standard. LoDs and limits of quantitation were in the low ng/L range (Table S1) for 6-PPDQ (LoD: 6.0 ng/L; LoQ: 20 ng/L), IPPDQ (11; 36), CPPDQ (2.0; 6.6), DTDPQ (3.3; 11), and 77PDQ (4.8; 16). DPPDQ showed the poorest performance with an LoD of 28 ng/L (LoQ: 94). We attribute this to less efficient ionization by ESI due to the lower basicity of the amine moieties in DPPDQ, as electron density can be more delocalized into the phenyl rings. While DTDPQ may also be expected to exhibit this behavior, similar sensitivity to 6-PPDQ was observed for DTDPQ. Steric hindrance of the ortho-methyl groups may reduce the pi-orbital overlap and reduce electron delocalization into the aromatic system.⁴¹ Due to the proximity of parent [M + H]⁺ ions (e.g., m/z 299 and 297 for 6-PPDQ and CPPDQ, respectively) and the potential for source fragmentation of 6-PPDQ into IPPDQ, we conducted a control experiment and observed no signal intensity in the CPPDQ or IPPDQ MS/MS channels in response to several spike additions of 6-PPDQ (Figure S6).

The detection limits achieved here are within an order of magnitude of those achieved by conventional workflows^2,15 (e.g., LC-MS) and are below the acute toxicity values reported for aquatic organisms (41–95 ng/L LC₅₀ for coho salmon).^2,6 We believe that given the simplicity and high-throughput capabilities of CP-MIMS, it provides a “fit-for-purpose” complement to conventional workflows. For instance, CP-MIMS provides an excellent avenue to probe the behavior of 6-PPDQ under controlled conditions as illustrated with the following two examples:

• (1)The aqueous stability of 6-PPDQ remains an open question in the literature and has important implications for sample hold times, operation of exposure tank toxicological studies, and optimization of engineered mitigation strategies. An early report from Hiki et al. suggested a half-life as short as t_1/2 = 33 h,³⁰ which would challenge comprehensive environmental monitoring campaigns significantly. To verify this result, we monitored the 6-PPDQ concentration in several fortified samples over 2 weeks at room temperature (Figure 2). Minimal loss of 6-PPDQ was observed beyond the precision limits of the method (ca. ± 15%). Real-world samples stored at 4 °C exhibited similar stability over the short term, with significant losses only occurring after months of storage (55–65% concentration reduction over 3.5 months; 4 °C). While these results do indicate some loss of 6-PPDQ during extended storage, it is considerably slower than that observed by Hiki et al.³⁰ However, when we repeated the experiment with a series of standards left uncapped (open to the air), we observed loss of 6-PPDQ with t_1/2 = 51 h (Figure 2 C,D). Given that the partitioning rate between water and air will depend on a number of factors (e.g., surface area, agitation, temperature), this result is in better agreement with that reported by Hiki et al.³⁰ When we examined the rate of loss across the series of PPDQ analogues (Table S2) in the uncapped vials, 77PDQ exhibited the shortest half-life (t_1/2 = 14 h) and DTDPQ had the longest (t_1/2 = 108 h). It is unclear whether loss of PPDQs from the aqueous phase is due to physical and/or chemical processes. In either case, the rapid loss of PPDQ analogs from aqueous solutions open to the atmosphere has important implications for the environmental fate and distribution of these contaminants, as well as the design of experiments involving PPDQs.
• (2)Given the relatively high octanol-water partition coefficient (log K_ow = 3.98) of 6-PPDQ,¹⁵ some fraction is expected to sorb to sediments rather than remaining in the aqueous phase. Because CP-MIMS provides an online membrane cleanup, the aqueous phase concentration can be probed directly from a heterogeneous slurry containing sediment or sorbent.²⁹ Capitalizing on this, we set up a series of samples with the same initial concentration of 6-PPDQ and increasing loadings of a low (1.85%)- and high (5.96%)-organic-content sediment. After 2 days of equilibration under gentle shaking, the remaining fraction of 6-PPDQ in the aqueous phase was measured directly within the heterogeneous sample (Figure S7). Relative to negative control (Ottawa sand), significant sorption occurred for both sediments, with a greater extent observed for the high organic content sediment (up to 90% removal from the aqueous phase). Preliminary organic carbon–water partition coefficients log K_OC (L/kg) can be derived from these data, yielding values of 2.8 ± 0.8 and 3.6 ± 0.5 for the low- and high-organic-carbon sediment, respectively. These values are in good agreement with both calculated values (3.14)⁴² and experimental reports (3.2–3.5);⁴³ however, given the relatively short equilibration time (2 days) and modest extent of sorption for the low-organic-carbon sediment, caution should be exercised when interpreting these values. We are currently conducting a more exhaustive study to probe heterogeneous partitioning behavior using longer equilibration times and a broader array of sorbent materials.

Quantitative Performance for PPDQ Analogues

With any new analytical tool, performance should be thoroughly evaluated in order to understand the reliability of results in real-world samples. In this work, performance was assessed with calibration check solutions interspersed throughout regular samples. Additionally, spike recovery experiments were evaluated in a series of surface water samples.

Figure S8 shows the results for the calibration check solutions (n = 81; ca. 60 ng/L of each PPDQ). 6-PPDQ showed excellent performance, with a 73/81 (90%) calibration check solutions exhibiting bias <±30% and a relative standard deviation of ±21%. The most similar structural analogues IPPDQ and CPPDQ also performed well, both with 64/81 (79%) exhibiting bias < ± 30%. However, DPPDQ, DTDPQ, and 77PDQ showed poorer performance with inconsistent (DPPDQ, DTDPQ) or systematically biased recoveries (77PDQ). For 77PDQ, the observed concentration of the calibration check solution drops over the course of individual sample runs (Figure S9), with the first calibration check solution of each run exhibiting an average bias of 7 ± 28% and the final one (ca. 110 min later) exhibiting an average bias of −54 ± 44% (n = 10 runs). In contrast, for 6-PPDQ, these values were 10 ± 26 and −2.0 ± 26%, respectively (n = 15 runs). This suggests loss of 77PDQ from the uncapped vials during analysis, which is consistent with the uncapped vial stability study where 77PDQ exhibited the shortest half-life (t_1/2 = 14 h, Figure 2D). The more rapid loss observed here may be attributed to the continuous agitation on the autosampler tray. For DPPDQ and DTDPQ, some sample batches produced relatively reproducible and accurate quantitation of the calibration check solution (e.g., November 1st run with 9.5 ± 8.7% bias for DTPDQ and −7 ± 46.8% for DPPDQ, n = 5) punctuated by later batches exhibiting more inconsistent and/or inaccurate results (e.g., November 22nd run with −29.6 ± 40 and 61 ± 119% for DTDPQ and DPPDQ, respectively). We realized that the periods of good performance corresponded to when fresh acceptor phase had been prepared. We believe this is due to the sensitivity of these analogues to the formic acid concentration and the limited stability of formic acid in methanol, as it can be slowly esterified to form methyl formate.³¹

From this point onward, care was taken to ensure formic acid was added to the acceptor phase immediately before analysis. This approach improves the performance of these compounds as demonstrated by a spike recovery experiments for five real-world samples using fresh acceptor phase (Figure S10 and Table S3). Water quality parameters for these stream samples are available in Table S4. A low- (60 ng/L) and mid-concentration (125 ng/L) spike of the PPDQ suite were added to samples, and the recovery was measured against an unfortified sample. 6-PPDQ, IPPDQ, and CPPDQ again show excellent performance, exhibiting mean recoveries of 109 ± 14, 105 ± 13, and 100 ± 16%, respectively. With the fresh formic acid in the acceptor phase, DTDPQ now also shows excellent recoveries (mean of 106 ± 11%). For DPPDQ, the lower-concentration spike did not always produce a signal above the limit of detection. However, the mid-concentration spike shows good performance with recoveries between 66 and 120%. The loss of 77PDQ to the atmosphere was also apparent in this experiment, with the recovery dropping steadily as the run progressed.

Based upon the above results for the calibration check solutions (Figure S8) and spike recovery experiments (Figure S10 and Table S3), we are confident in the quantitative results for 6-PPDQ, IPPDQ, and CPPDQ in environmental samples. For DPPDQ and DTDQ, performance was significantly improved by the inclusion of fresh formic acid in the acceptor phase. However, because this was only implemented partway through the sampling campaign, our reporting for these analogues in this work should be considered semiquantitative. Similarly, experimental results suggest significant loss of 77PDQ to the atmosphere even during the analysis of a single autosampler tray (Figure S9). As such, reporting for 77PDQ using CP-MIMS should also be considered semiquantitative or even be relegated to presence/absence. While not evaluated systematically here, this effect can be mitigated by reducing the time samples are open to the atmosphere as was employed for Figure 2C. Alternatively, inclusion of an isotopically labeled standard for 77PDQ could compensate for this effect. We are working to obtain isotopically labeled standards for the full suite of PPDQ analogues to improve quantitative reporting.

Spatiotemporal Monitoring of PPDQs

Roadway runoff is an intrinsically diffuse problem, benefiting from local expertise to identify threatened streams and potential inputs. This information can then be employed by regional authorities and municipalities to mitigate the risks of 6-PPDQ entering waterways and coastal environments. We partnered with 10 volunteer groups across coastal British Columbia, including local First Nations and stream keeper groups as part of a citizen science campaign to collect stream samples before, during, and after rain events. Where available, roadway runoff samples were also collected near stream sites, resulting in ca. 75 samples per rain event and 192 samples (168 stream samples and 24 roadway runoff samples) over the Fall 2022 season.

Concentrations of 6-PPDQ in stream water samples were relatively low (<150 ng/L; Table S5), and 6-PPDQ was detected in 20 of the 50 samples (40%) collected during rain events. Four of the 14 streams studied (28%) exhibited concentrations exceeding the LC₅₀ for coho salmon (41–95 ng/L)^2,6 during at least one rain event. Substantial variability in concentration was observed both between sites (Figure 4) and for the same site over time (Figure 5). Where appreciable 6-PPDQ was observed, a distinct concentration pulse was generally observed across the before-during-after stream sample series. Smaller and/or more urbanized creeks were more prone to high concentrations, even during relatively small rain events. IPPDQ and CPPDQ were also observed, often co-occurring with 6-PPDQ at somewhat lower concentrations (20–80 ng/L; Tables S6 and S7). Higher concentrations were observed in the roadway runoff samples (n = 24; Figure S11), with the average 6-PPDQ concentration (360 ng/L) an order of magnitude higher than in surface waters (28 ng/L) collected during a rain event. These roadway runoff samples also exhibited concentration variability both site-by-site and across different rain events, but the average concentrations of 6-PPDQ (360 ng/L), IPPDQ (110 ng/L), and CPPDQ (200 ng/L) are generally in line with those observed by Cao et al.¹⁵ In that work, runoff water in Hong Kong was measured, exhibiting median concentrations of 1120 ng/L (6-PPDQ), 560 ng/L (IPPDQ), and 60 ng/L (CPPDQ).

Figure 4.

Map of eastern Vancouver Island, British Columbia (BC), Canada, and the lower mainland of BC summarizing observed concentration of 6-PPDQ in streams over the fall 2022 sampling campaign. Dots are sized and colored according to the maximum concentration of 6-PPDQ observed at each site across all sampling times. Indicated sites A–D are Millstone river up (A) and downstream (B) of a highway, Northfield creek (C), and Cougar creek (D).

Figure 5.

Concentration of 6-PPDQ over time for the 4 stream sites (A–D) indicated in Figure 4. The shaded gray area indicates the precipitation over time (reverse axis) for the nearest weather station at each site. Sites A and B are along the same river (Millstone), up- and downstream of a highway, respectively. While the upstream site (A) exhibits no 6-PPDQ above the limit of detection, the downstream site (B) reaches concentrations ≥50 ng/L during a ≥15 mm rain event. (C) and (D) are small urban creeks in a small (Nanaimo; Northfield) and large (Vancouver; Cougar Creek) city in British Columbia. In these urban creeks, even small rain events (5–10 mm) introduce significant levels of 6-PPDQ. Bars represent the average of n replicates (n indicated next to error bar), and the error bars represent the standard deviation across those replicates. Lighter gray shading in the background indicates periods where no samples were collected, most notably over the period of October 29–November 19.

To validate the quantitative results obtained with CP-MIMS, 12 duplicate samples were collected at select sites for additional conventional analysis of 6-PPDQ using a liquid–liquid extraction–liquid chromatography–tandem mass spectrometry method (LLE-LC-MS/MS). Across these 12 samples (10 surface water samples and 2 roadway runoff samples), an excellent correlation was observed between the LLE-LC-MS/MS method and CP-MIMS (Figure 3; R² = 0.993). For the roadway runoff samples, CP-MIMS exhibited a negative bias of 22 and 30% relative to the LLE-LC-MS/MS results. This difference may be attributable to the fact that CP-MIMS probes the concentration of analytes that are free in solution. Given that these samples were unfiltered and that conventional LLE methods are more exhaustive, it is likely that they are recovering analyte bound to fine suspended solids (including tire particulates) prevalent in roadway runoff samples. It should be further noted that the stream water samples were measured within 5 days of collection (i.e., minimal if any loss expected; Figure 2B) while the roadway runoff was held for up to 100 days (some instability expected). When we compare quantitation for only the stream samples (inset in Figure 3B), excellent quantitative agreement is observed (n = 10; slope = 1.01, R² = 0.992). Residual plots for both the “all samples” and “surface waters” plots are available in Figure S12. We believe that this cross-validation reaffirms the value of CP-MIMS as a platform to quickly probe and/or answer sample-intensive research questions. As with any direct mass spectrometry method, we recommend prudence when interpreting results given the greater risk of interference in complex, real-world samples. In this work, multiple MS/MS transitions are monitored for each PPDQ analog. Atypical ion ratios between these transitions can indicate the presence of an interferent. This revealed that the m/z 299 → 215 MS/MS transition previously employed for quantitation²⁹ can lead to positive bias in the quantitation of 6-PPDQ. As such, we no longer recommend this transition for analysis of 6-PPDQ using CP-MIMS. Interference at m/z 299 → 215 may be due to the presence of isomeric ozonation products of 6-PPD. Zhao et al. recently characterized several such products, revealing three 6-PPDQ isomers.³² Product ion scans for all three isomers exhibited m/z 215 fragment ions, but do not exhibit the fragment ions employed herein for MS/MS quantitation (m/z 256) or qualification (m/z 243 and 100). CP-MIMS can be employed as a rapid “first-tier” analysis to identify and quantify a large suite of samples, allowing for more efficient allocation of LC-MS resources to samples with highly consequential or suspect results (Figure S13). Here, we have reported all quantitative results and flag those with unusual qualifier ion ratios (blue italics in Table S5).

Figure 3.

Quantitative agreement between CP-MIMS (blue, right bars) and a commercial LLE-LC-MS/MS method (gray, left bars) for analysis of trace 6-PPDQ. Good correlation is observed when comparing all samples (n = 12; R² = 0.993), and the two methods exhibit quantitative agreement for surface waters (n = 10; slope = 1.01, R² = 0.992). The numbers above CP-MIMS results in (A) indicate the number of replicate measurements taken for that sample.

Figure 4 presents a geospatial map summarizing quantitative data for 6-PPDQ in streams across the sampling region. To visualize local “hotspots,” points are sized and colored according to the maximum concentration observed for 6-PPDQ at each site. Similar plots for IPPDQ and CPPDQ are available in Figure S14. Numerous factors are expected to contribute to the PPDQ concentration in a given stream at a particular point in time, including the surrounding topology, overall stream flow,⁴⁴ and traffic density. Here, larger streams and those outside of city centers were generally observed to have lower 6-PPDQ concentrations (<20 ng/L); however, those within urban areas reached concentrations up to 180 ng/L. The time-course data for 4 stream sites (indicated as A–D in Figure 4) are provided in Figure 5. The shaded area indicates the precipitation from the nearest weather station for each site. A and B are sites along the same stream (Millstone), upstream (A) and downstream (B) from a highway. C and D are smaller, more urbanized streams in Nanaimo (Northfield creek) and Vancouver (Cougar Creek), BC, respectively. During the same rain events, these sites show significantly differing behavior. Throughout the first (∼6 mm, October 23–24th) rain event, the larger system (B; Millstone) shows low concentrations of 6-PPDQ (<15 ng/L) while the more urbanized sites C and D rise to ca. 35 and 110 ng/L, respectively. However, during the second, larger (∼20 mm, October 27–28th) rain event, the concentrations at all three sites (B–D) exceeded the coho LC₅₀ by a factor of ca. 2–4.5. Through the third rain event captured (November 23–24th), only the two small, urbanized creeks (C and D) showed appreciable 6-PPDQ. The site upstream of the highway (A; Millstone) remains below the detection limit during all three rain events supporting the source as roadway runoff. At this stage, it is unclear whether the differences in concentration we observe in different streams are due to road runoff inputs (e.g., traffic density) and/or stream hydrology (e.g., flow). Further, the concentration of 6-PPDQ throughout a rain event is dynamic.⁴⁵ Given the nature of collecting samples through a citizen science network, there is some variability in precise sampling time relative to the rain event, which confounds detailed comparisons between sites. We are currently expanding sampling campaigns for wider surveillance to identify persistent hotspots and characterize site-specific flush dynamics for 6-PPDQ. Capturing this type of fine spatiotemporal scale data would be both time- and cost-prohibitive without the high-throughput measurement capacity afforded by CP-MIMS.

Conclusions

Direct mass spectrometry measurements employing tandem MS provide unique opportunities for high-throughput analysis, enabling greater spatial and temporal resolution in environmental assessments. Here, CP-MIMS is used to provide quantitative information on PPDQs at low ng/L levels at a throughput of up to 100 samples/day with only minor sample preparation (i.e., addition of internal standard). Quantitative comparison between a conventional method (LC-MS) and CP-MIMS for a series of surface waters shows excellent agreement (n = 10, slope: 1.01, R² = 0.992). The high-throughput capability of CP-MIMS is applied to answer fundamental questions about this emerging class of toxins (sample storage time, partitioning behavior), revealing significant loss to the atmosphere for PPDQs (t_1/2 = 14–108 h). These results suggest the need for careful experimental design and ongoing stability evaluation when studying 6-PPDQ and analogues to ensure reliable results. The method is employed to analyze 192 real-world stream and road runoff samples, representing the first spatiotemporal survey of PPDQs on Vancouver Island and the lower mainland of British Columbia (Canada). The resulting spatial and temporal variability of large sample sets can be followed up with conventional analysis, if and when required.

An overarching objective of this work is to provide molecular-level information when and where it is needed through the development of simple, high-throughput analytical techniques. Future work includes adapting the simple workflow enabled by CP-MIMS to fieldwork, allowing for real-time streamside measurements to characterize pulse dynamics and/or stormwater inputs.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsestwater.3c00275.

• Full analyte names and structures (Figure S1), calibration data using different internal standard strategies (Figure S2), workflow for probing loss of PPDQs to the atmosphere (Figure S3), photos of autosampler construction and setup (Figure S4), overview of automated data processing (Figure S5), MS/MS cross-talk evaluation (Figure S6), preliminary sorption data (Figure S7), bias on calibration check solutions (Figure S8), bias on calibration check solutions as a function of analytical run time (Figure S9), spike recovery experiments (Figure S10), roadway runoff concentrations (Figure S11), residual plot for comparison of LC-MS and CP-MIMS (Figure S12), proposed workflow for complementary use of CP-MIMS and LC-MS (Figure S13), and “hotspot” mapping for IPPDQ and CPPDQ (Figure S14), MS/MS instrument parameters (Table S1), tabulated atmospheric loss data (Table S2), tabulated spike recovery results (Table S3), water quality data for spike recovery experiments (Table S4), quantitative results for 6-PPDQ (Table S5), IPPDQ (Table S6), and CPPDQ (Table S7) (PDF)

The authors declare the following competing financial interest(s): Chris G. Gill and Erik T. Krogh hold patent US 9583,325 issued to none.

Special Issue

Published as part of the ACS ES&T Watervirtual special issue “3D Printing Technologies for Environmental and Water Applications”.