We obtained labeled and native DAP metabolites form Cambridge Isotope Laboratories (Andover, MA, USA). including DMP, DEP, DMTP, DETP, DMDTP and DEDTP and the corresponding alkyl-D₆ and D₁₀ labeled analogs for dimethylphosphates and diethylphosphates, respectively. Glyphosate was obtained from Supelco (Bellefonte, PA, USA), and ¹⁵N, 2-¹³C Glyphosate was purchased from Cambridge Isotope Laboratories. HPLC grade solvents used include isopropanol and methanol from Fisher Scientific (Pittsburg, PA, USA). Ultrapure water was generated in house (AQUA solutions, Lab water systems, Jasper, GA, USA).

Glyphosate (10 mg/L) and DAPs stock solutions (1 mg/L) in water were stored at −20 °C in Teflon™ capped silanized glass vials until further use. Ten calibration standards containing known amounts of DAPs and glyphosate were prepared by serial dilution of the stock solutions in water to final concentrations ranging from 0.1 μg/L to 60 μg/L. The lowest concentration calibrator (0.1 μg/L) was used as an “instrument ready check” to confirm low baseline conductivity, good chromatographic resolution and high sensitivity before the start of an analytical run, and was excluded from the 1/x weighted calibrations curves, leaving the calibration in a range between 0.2 μg/L – 60 μg/L. The internal standard stock solutions containing isotope-labeled DAPs and glyphosate were diluted with water in a volumetric flask to reach concentration levels of 75-100 μg/L, depending on the analyte. All involved glassware was silanized, to minimize adsorption of analytes onto the glass surface.

200 μL of the urine sample or aqueous standard was transferred into a silanized vial and 50 μL of internal standard and 200 μL water was added. For matched matrix calibration, 200 μL of blank urine was added instead of 200 μL water. The diluted sample was vortexed for 30 sec before analysis.

As of today, commercial quality control (QC) materials for human biomonitoring do not exist for OP metabolites and glyphosate. Therefore, for QC and method validation purposes, we collected human urine anonymously from male and female adults in Atlanta, GA. No personal information was obtained. We cannot rule out that multiple samples were obtained from the same individuals. Collected samples were stored at −70 °C before analysis. CDC’s Institutional Review Board approved the urine collection and analysis. A waiver of informed consent was requested under 45 CFR 46.116(d).

After screening for endogenous DAP and glyphosate concentrations we pooled urine samples with the lowest concentrations (representing a “blank” urine). We adjusted the concentration of the target analytes in low concentration pool aliquots to 1.0 and 2.0 μg/L for the QC low and 10 to 21 μg/L for the QC high. Two replicates of each QC material were included in all runs to ensure high quality measurements. Each QC material was characterized by repeated measurements to define mean concentrations and 95% and 99% control limits of each target analyte. The calculated mean of the concentrations of the two QCH and two QCL samples analyzed with each analytical batch were evaluated by standard statistical probability rules (Caudill 2008). Furthermore, as part of an in-house proficiency testing (PT) program, we prepared PT samples at 3 concentration ranges. PT_low has concentrations between 0.6 to 1 μg/L, PT_mid 6-10 μg/L, and PT_high 10-20 μg/L. PT samples were prepared in duplicates and characterized by at least 20 repeat measurements and data were evaluated by a PT administrator to determine the mean and standard deviation for each analyte. PT samples were externally blinded and stored at −70 °C. Every six months, a PT administrator randomly selects 5 samples for analysis and evaluates results to ensure long term method ruggedness and accuracy.

In addition to the in-house QC and PT programs, we regularly participate in the German External Quality Assessment Scheme (GEQUAS) for glyphosate and DAP metabolites (see http://www.g-equas.de/) and the External Quality Assessment Scheme for Organic substances in Urine (OSEQAS) for glyphosate (see https://www.inspq.qc.ca/en/ctq/eqas/oqesas/description).

Separation was accomplished with a Dionex ICS-5000+ ion chromatography system (Thermo Fisher Scientific, Sunnyvale, USA) consisting of a column and ion suppressor compartment, an autosampler, a dual pump system (1-channel and 3-channel pump) and an electrochemical KOH generator (Figure 2). In addition, we used three 1-channel auxiliary pumps, one attached to the suppressor, the second one for infusing isopropanol into the mass spectrometer after chromatographic separation and the third for infusing 50% isopropanol into the mass spectrometer electrospray ionization (ESI) source for cleaning purposes while chromatographic separation was carried out. All columns were obtained from Thermo Fisher Scientific. The two column forward-flush assembly consists of a Dionex UTAC-LP2 column for clean-up and enrichment followed by a Dionex AS24 (2 × 250 mm; 4 μm particles) ion exchange column for chromatographic separation. To increase the column durability, we added a guard column (AG24, 2 x 5mm, 4 um). The temperature in the column compartment was adjusted to 30 °C. 20 μL of the processed sample were injected with a constant flow of 1.5 mL/min H₂O. After 2 min, the analytes were transferred to the analytical column using 30 mM KOH. At 7 min, the concentration of KOH was increased to 100 mM and kept for 3 minutes (we also increased the flow rate between 8 and 9 min to 0.5 mL/min). The concentration of KOH was kept at 100 mM for 7 minutes. At 17.1 min the concentration was reduced to 30 mM KOH for 2.9 min (Table 2).

We used an AB Sciex 5500 triple quadrupole mass spectrometer (Applied Biosystems, Foster City, CA, USA) equipped with a TurboIonSpray® source. The ESI source was operated in the negative ion mode at −4500 V, declustering potential (DP) of −80V and entrance potential of −10V. Nitrogen was used as collision gas (CAD cell pressure set to 9 (unitless)), curtain gas 20 psi, turbo heater and evaporator gas each at 60 psi. The ESI source temperature was optimized at 600 °C. The collision cell exit potential was set at −10V. The target scan time was set to 1 second. Resolution of Q1 and Q3 was set to “unit” and the time between transitions was 0.5 ms. Time-programmed multiple reaction monitoring was used, allowing peak monitoring for ±40 seconds from the analyte specific retention time. Specific ion transitions can be found in Table 1.

We purchased 90 urine samples from BioreclamationIVT (Hicksville, NY). The company had IRB approval to collect urine and obtained informed consent from donors who were verbally asked whether they had been in contact with glyphosate within the last 24 hrs of specimen donation (n=40) or whether they had been on an organic diet (n=50). No personal identifiers were provided to CDC.

On-line extraction and liquid chromatography in back-flush mode combined with tandem mass spectrometry have become an integral part of human biomonitoring (Baker et al. 2019, Kato et al. 2005, Kato et al. 2018, Schütze et al. 2016, Valentin-Blasini et al. 2005, Wang et al. 2017, Ye et al. 2005). We investigated several columns including reversed-phase, weak ion exchangers and HILIC columns and tested several combinations for a fast reliable and robust on-line HPLC method for extraction and separation of glyphosate and DAPs but lack of retention, low sensitivity, interferences and carry-over were extremely challenging and neither of those options satisfied the method objectives (data not shown).

Glyphosate is a small molecule, highly polar and forms metal complexes, which complicates its quantification (Subramaniam and Hoggard 1988, Sundaram and Sundaram 1997). The IC system we chose uses polyether ether ketone (PEEK) only materials to prevent carry-over and interaction with metal surfaces. Here we describe an IC extraction/separation using a forward-flush elution mode with a highly corrosive eluent and not volatile, KOH, with mass spectrometry detection, which under regular circumstances would not be compatible with such mobile phase.

Two key components of the system are the hydroxide eluent generator and the 500e eluent suppressor. To generate KOH, deionized water is pumped through the eluent chamber and a DC current is applied between the anode and cathode. This leads to the electrolysis of water at the Pt anode and Pt cathode. While water is oxidized at the anode, O₂ and H⁺ are generated in the reservoir. H⁺ displaces the K⁺ in the reservoir, after which K⁺ migrates across the cation exchange connector into the eluent generation chamber. At the cathode, water is reduced to form hydrogen gas and OH^-, which combines with the displaced K⁺ ion. Formed hydrogen gas is taken out of the mobile phase via a degasser module. The 500e eluent suppressor applies DC current to cleave water into H⁺ and oxygen at the anode and reduces water to OH^- and hydrogen gas at the cathode. The H⁺ ions are then transported across a membrane into the eluent to neutralize the highly basic eluent, while K⁺ is driven towards the cathode and transported into the waste, thus preventing K⁺ build-up and corrosive damage to the mass spectrometry ESI source.

The ion chromatography system consists of a two-column assembly, a UTAC LP2 column used for analyte extraction and clean up and an AG24/AS24 guard/column combination for analytical separation. Sample clean-up was achieved within the first minute of the method runtime. All analytes were well retained and longer washing times had no significant impact on matrix effects (data not shown) but led to slight peak distortion and loss of chromatographic resolution on the analytical column. 30 mM KOH proved to be useful to transfer the analytes from the extraction column forward onto the analytical column. The elution order was DEP, DMP, DETP, DMTP, glyphosate, DEDTP followed by DMDTP (Figure 3). DMP and DEP coeluted, however the analytes do not share the same ion transitions and unlikely interfered with each other. A closer look at the suppressor gradient reveals some “over-suppression.” For example, at 7 min, the KOH concentration is 30 mM at a flow rate of 0.4 mL/min and corresponds to 30 mA for full suppression. However, at 8 min we increase the flow rate to 0.5 mL/min and the current at the suppressor to 124 mA which can suppress 100 mM, however this concentration is generated later in the gradient at 10 min and takes approximately another 120 seconds to reach the suppressor (Table 2). Noteworthy, the suppressor current does not increase linearly, but rather jumps to the next current level at any given time a command is set. Therefore, the suppressor current command needs to be issued before linearly increasing the KOH concentrations to protect the ESI source from KOH influx at any given time, which is the case within this method.

One major concern for ion chromatography is retention time stability. We found no significant retention time changes within the same column regardless of the type and number of urine samples analyzed. To investigate the potential impact of retention time changes between columns for larger human biomonitoring studies such-as the National Health and Nutrition Examination Survey (NHANES) we rotated 6 analytical columns (3 different column lots) and 2 extraction columns in a total of 24 batches (2 runs per combination) and found that retention time differences among columns did not exceed the column specifications provided by the manufacturer.

An integral part of the system is the conductivity detector whose response is affected by all anions and allows us to determine when inorganic matrix components are eluted or whether OH^- is not fully suppressed. Certain interferences can cause ion defocusing in the ESI source that can considerably decrease system sensitivity over time. As can be seen in Figure 3, the analytes (Figure 3a) are separated from most of those anions (Figure 3b), which allowed us to divert these matrix ions into the waste during that time. Furthermore, to increase system stability and sensitivity we used two additional auxiliary pumps (AXP2 and 3). AXP2 pumped 100% isopropanol at 0.1 mL/min to enhance ionization efficiency and AXP3 pumped 50% isopropanol at the same flow rate for additional cleaning while the chromatographic separation was performed (divert position).

The limit of detection (LOD) was established using the Taylor et al. (1987) method, which assesses the relative standard variation at each analyte calibration level. LODs are 0.2 μg/L for all analytes except DEDTP which was set to 0.8 μg/L.

The total precision was estimated as percent relative standard deviation (%RSD) by preparing and analyzing duplicate QClow and QChigh samples in 10 consecutive days using 3 different AS24 columns. Precision ranged from 3.4 to 18.4 %RSD. All values, except those of DMDTP and DEDTP in the QClow (16.6 % and 18.4 %, respectively) were within the 15 %RSD recommended by U.S. FDA guidance document (US-FDA, 2018). A detailed breakdown including intra and inter-day precision can be found in table 3

Because there are no commercial standard reference materials for glyphosate or DAPs in urine, the accuracy of the method was established by spiking two urine specimens at native (zero), low, mid and high concentrations and analyzing them in triplicate in two consecutive days (n=6 per sample and spike level, Table 4). After subtracting native concentrations, the calculated mean relative recovery was between 85-115% for all analytes at all spike levels in both investigated samples. One additional spike level was tested in a blank urine sample and showed an average of 102% (range 92 – 112%) relative recovery for glyphosate at 1 μg/L. Sample and analyte stability were investigated using QC materials under 3 different conditions. First, bench-top stability, which assessed short-term stability for length of time needed to handle study samples (typically at room temperature). Samples were left outside at room temperature overnight and processed for injection the next day. Second, freeze-thaw stability, in which samples were frozen at −80°C and thawed 3 times and later prepared and analyzed. Third, processed sample stability, in which samples were analyzed after remaining in the autosampler for 2 days. We found no significant impact for any of the analytes, all analytes remained stable.

We analyzed 90 samples from subjects with either suspected exposure to pesticides (n=40) or who reported being on an organic diet (n=50). as shown in table 5, glyphosate concentrations varied from <0.2 to 28.6 μg/L, with an 80% detection frequency in the organic diet group and 78% in the glyphosate usage group. Median concentrations (calculated after replacing concentrations <LOD by LOD/SQRT(2) (Hornung and Reed 1990)) are comparable in both subsets, 0.40 μg/L for those potentially exposed vs 0.39 μg/L for those on an organic diet. However, a comparison of the interquartile range shows the influence from usage (0.63 μg/L vs 0.42 μg/L), suggesting exposure to glyphosate in both groups. Our glyphosate results compare well to published data here in the United States. Curwin et al. (2007) reported values for fathers (n=47) and mothers (n=48) from farm and non-farm families. Concentrations of glyphosate ranged between 0.02 μg/L and 18 μg/L, with geometric means for fathers from farm families of 1.9 μg/L and non-farm, 1.4 μg/L. Mothers’ concentrations were 1.2 μg/L for non-farm and 1.5 μg/L for farm families, respectively. Parvez et al. (2018), reported concentrations in the range of 0.5-7.2 μg/L with a mean of 3.4 μg/L in pregnant women (n=71). Internationally, data from German university students showed relatively overall low concentrations of glyphosate ranging from 0.11 μg/L to 0.6 μg/L in 2012 (Conrad et al. 2017).

Similarly, for the DAP metabolites, the organic group had comparatively lower median concentrations for DMP (organic diet 0.27 μg/L vs suspected exposure 0.90 μg/L), DMTP (0.41 μg/L organic diet vs suspected exposure 0.97 μg/L) and DEP (0.86 μg/L organic diet vs suspected exposure 1.86 μg/L). Detection frequency ranged from 54 to 82%, depending on the analyte and group, except for DMDTP and DETP that were detected in only about a quarter of the samples. DEDTP was only detected in one sample, just above the LOD of 0.8 μg/L. Overall, concentrations in the samples analyzed for this study are slightly higher than those reported for NHANES (CDC 2019). Reported 2007-2008 NHANES median and 95^th percentile concentrations were <LOD (0.47 μg/L) and 35.6 μg/L (DMP), 2.10 and 36.8 μg/L (DMTP), <LOD (0.51 μg/L) and 5.60 μg/L (DMDTP), <LOD (0.37 μg/L) and 15.3 μg/L (DEP), <LOD (0.56 μg/L) and 4.35 μg/L (DETP), and both <LOD (0.39 μg/L) (DEDTP). However, factors that may contribute to variations in the urinary concentrations of DAPs include different sampling procedures (e.g., spot urine vs first morning void), differences in year(s) of urine collection, or analytical method differences. Of note, results of accredited external quality assessment schemes (e.g. GEQUAS) showing relatively wide tolerance ranges when determining DAP metabolite concentrations highlight the technical challenges encountered in the quantification of DAPs.

Overall, this method offers high selectivity, excellent precision (<5% RSD) and accuracy (mean relative recovery 99%, range 97-103%) for glyphosate. A fast and simple sample preparation (dilute and shoot) using 200 μL of urine, combined with a run-time of 20 min makes it an excellent choice for large human biomonitoring studies to investigate glyphosate exposure in the general population.

After optimizing every instrumental parameter (e.g. injection volume) and testing several types of columns for robustness, we concluded that only extensive sample preparation, out of scope for this study, is likely to improve sensitivity, which our data showed is enough to quantify glyphosate concentrations in urine samples with environmental exposure levels from the United States. Further investigation showed that the quantification of DAPs in large number of samples for population-based studies (e.g. NHANES) was not as rugged as desired despite successful participation in external performance assessment schemes (i.e., GEQUAS, OSEQAS) and solid results during method validation. The used QC material drifted between 60-140% from the average, first only for DMTP, then DEP followed by DMDTP, and then the rest of the DAP metabolites. Unsurprisingly, the deuterated internal standards were almost baseline separated for some DAP analytes, which can be explained by physicochemical properties of the deuterated analytes compared to their native analogs, and therefore can cause these retention time differences. In combination with a highly variable composition of the urine matrix, the retention time differences between internal and native standard might negatively impact method performance. This effect seems to be enhanced with the batch size and QC placement in the measured batch which is randomized for NHANES. Typical batches during the method validation had a total of 25 samples (including calibrators and QCs), batches containing NHANES samples, however, had 62 samples total, increasing the matrix load dramatically.

It is unclear what matrix components might have caused these problems because the columns were thoroughly flushed. We ruled out the presence of ghost peaks or analyte carry-over by extending the method runtime. Instead, we found the matrix related effect to be reversible by the introduction of methanol. However, this approach would decrease life expectancy of the suppressor while increasing backpressure, and total run time (+60 min). This runtime would far exceed the desired run-time for large human biomonitoring studies significantly. Therefore, we determined that the current method with deuterated internal standards is only suitable for human biomonitoring studies of relatively small sample sizes and is not applicable for NHANES.