Non-target screening of organic compounds in offshore produced water by GC×GC-MS

Chemicals and reagents

Benzoic acid, phenol, cycolhexanecarboxylic acid, octanoic acid, dichloromethane (LiChroSolv, Merck, Darmstadt, Germany), n-hexane (SupraSolv for gas chromatography MS, Merck, Darmstadt, Germany), magnesium sulfate (ReagentPlus, Redi-Dri, Sigma–Aldrich, St. Louis, MO, USA), N,O-bis(trimethylsilyl)trifluoroacetamide containing 1% of trimethylchlorosilane (BSTFA+TMCS, Supelco, Bellefonte, PA, USA) were used as received. Deuterated internal standards (naphthalene-d₈, acenaphthene-d₁₀, phenanthrene-d₁₀, chrysene-d₁₂, Supelco analytical standards) were used to monitor retention time shifts. ¹D retention index calibration was performed using a linear C7 to C30 saturated alkanes mixture (Supelco, TraceCERT, Sigma–Aldrich, St. Louis, MO, USA). High purity water was obtained from a Milli-Q Advantage A10 unit. All chemicals and reagents were used as received.

Sampling and sample preparation

Produced water samples were donated by Mærsk Oil & Gas (now Total E&P). The samples were acquired from a producing field (water-injected) located in the Danish region of the North Sea. The sampling campaign took place between June 2018 and February 2019. The samples were taken at irregular intervals with no replicates. Water samples were collected at the test separator on the production platform following protocols established by the operator. The test separator is a simple gravity-based three-phase separator where the water settles below the oil and can be sampled. No cleaning or further processing of the samples was carried out at the platforms. The samples were received in plastic bottles (1 L), aliquoted (500 mL), and immediately treated with dilute hydrochloric acid (18%, one mL per 100 mL sample) for a final pH < 2. Nine samples from the received batch were selected for analysis. The samples were selected due to the absence of dispersed oil droplets as evaluated by visual inspection (using microscope). The samples were stored in the dark at 4 °C until extraction and analysis.

Three aliquots (50 mL) of each produced water sample were extracted using separate glassware. The aliquots were filtered through a 0.45 µm PTFE-filter to remove solids and particles. Each 50 mL aliquot was extracted with DCM (50 mL). The organic phase was washed with saturated aqueous sodium chloride (50 mL) and carefully removed in vacuo. The residue was reconstituted in n-hexane (1.5 mL) and dried over MgSO₄. An aliquot (1,000 µL) of the sample was transferred to a two mL vial, combined with deuterated internal standards (for monitoring of retention time stability), combined with BSTFA+TMCS (50 µL) and incubated at 70 °C for 30 min, whereafter, it was allowed to return to room temperature. The derivatized sample was further diluted 10-fold with n-hexane and immediately analyzed on the GC×GC-MS.

Extraction recovery and reproducibility

A model produced water containing five representative model compounds (benzoic acid, phenol, 2-naphthoic acid, cyclohexanecarboxylic acid, and octanoic acid, each at 5 ppm, total organics 25 ppm) in synthetic formation water (see Supplemental Information) was prepared to establish variability in the sample preparation protocol and instrumental analysis. The concentration of total organics was chosen to emulate typical levels encountered at production platforms. The model water was extracted four times in two batches following an identical procedure as for the produced water samples. Three procedural blanks were prepared to establish background levels and experimental sources of contamination.

GC×GC-MS analyses

GC×GC-MS data were acquired using an Agilent 7890B GC coupled to a 7200B QTOF high-resolution mass spectrometer (Agilent Technologies, Palo Alto, CA, USA). The system was equipped with a Zoex ZX-2 thermal modulator (Zoex Corporation, Houston, TX, USA). The separation was achieved using a combination of an Agilent DB-5MS UI (¹D, 30 m, 0.25 mm i.d., 0.25 µm d_f) and a Restek Rxi-17Sil MS (²D, 2 m, 0.18 mm i.d., 0.18 µm d_f) capillary columns connected using a SilTite µ-union. The oven was temperature programmed as follows; 1 min hold-time at 50 °C, ramp to 320 °C, 3 °C min⁻¹ in constant flow mode (1 mL/min). The modulation period was set to 3 s with a 400 ms hot-jet duration. The MS transfer line was held at 280 °C. The MS acquired spectra in electron ionization mode (70 eV) with a mass range of 45–500 and an acquisition speed of 50 Hz. The instrument was operated in its 2 GHz sampling rate mode to increase the dynamic range. Automatic mass calibration was performed for every 5^th sample (approximately 7.5 h).

Data processing

Baseline correction (Reichenbach et al., 2003), peak detection and library search were performed using GC Image v2.8.3 (Zoex, Houston, TX, USA). Mass spectra were matched against the NIST Mass Spectral Library (National Institute of Standards and Technology, 2017 edition; NIST v17, Software Version: 2.3) with a minimum match factor of 700. All compound tables were exported as comma-separated texts for external processing. A data processing workflow was implemented in Python (Python Software Foundation. Python Language Reference, version 3.7.4. Available at http://www.python.org). The script is available as Supplemental Information deposited in a Zenodo repository (DOI 10.5281/zenodo.4009045).

Sample preparation and analyses

Previous studies on dissolved organics in oilfield produced water have employed LLE or solid-phase extraction (SPE) (Thomas et al., 2009; Kovalchik et al., 2017; Barros et al., 2018; Samanipour et al., 2019). For produced water, LLE using DCM has been shown to have a similar recovery as SPE (Samanipour et al., 2019). The main difference was observed in the size distribution, where larger species tend to have a higher recovery using SPE due to their low solubility in DCM. LLE is non-discriminative in comparison to SPE which is used to fractionate compound classes based on adsorption characteristics. Thus, it allowed us to extract the broad range of organics present in produced water. Furthermore, the Norwegian Oil and Gas specialist network recommends LLE for the quantification of phenols in produced water (Norwegian Oil & Gas, 2012). Thus, it would allow us to ‘see’ what is missing using routine targeted analyses. Approximately 30 samples were received as part of the campaign.

The water samples had varying levels of oil, likely due to the sampling and status of the test separator. Samples that contained a clear separate layer of oil, including smaller amounts, were excluded from the study to minimize the risk of contamination (Fig. 1). Optical inspection of these samples showed that they contained large amounts of dispersed oil droplets, even when sampling below the oil layer (Fig. 2). After exclusion, nine samples were chosen to be included in the study with extraction and characterization. All samples were filtered through a 0.45 µm PTFE-filter to remove insoluble material and particles. A small aliquot (50 mL) of each sample was extracted in triplicate using an equivalent volume of DCM.

The dissolved components of produced water were largely expected to be oxygenated organics, i.e. alcohols (mainly phenols) and acids. To increase the volatility of the aforementioned compounds, the samples were silylated. After removal of the solvent in-vacuo, each sample was reconstituted in 1.5 mL of n-hexane. A one mL aliquot of the reconstituted extract was treated with BSTFA-TMCS and incubated at 70 °C for 30 min. After derivatization, the samples were further diluted 10-fold, meaning that the samples ultimately were concentrated approximately 3-fold (from 50 mL to 15 mL). The final concentration factor was chosen as a compromise between the detection of trace compounds and avoiding column and detector overload. Due to the large concentration range of our analytes, it would be beneficial to run each sample at multiple dilution factors. However, due to the long run time (90 min) and the number of extraction replicates (3) this was not feasible due to time constraints. It is important to remember that this strategy has two effects; (1) high concentration compounds may overload the detector with spectral skewing and poor library match as result, and (2) trace-compounds may be diluted to a level below the limit of detection.

Three procedural blanks were extracted and analyzed to determine background levels and potential sources of contamination. Minor amounts of fatty acids were detected in addition to a series of polysiloxanes (discrete peaks, not common column bleed). The source of the latter could not be identified but as it lacked retention in the ²D it did not interfere with our analytes of interest. To establish recovery and repeatability, a model produced water was prepared by spiking four organic acids and phenol into synthetic formation water (see Experimental section and Supplemental Data). All model compounds were detected in one or more produced water samples. The model water was extracted in four replicates. The recovery values were calculated based on peak volumes in comparison to those obtained by analyzing pure stock solutions (Table 1). The recovery varied from 36% for benzoic acid up to 90% for phenol. Considering the multi-step sample extraction, including a derivatization reaction, this was deemed acceptable. The relative standard deviation of the peak volumes (measured after baseline correction using the GC Image package) of model compounds as detected in a representative produced water sample varied from 2% to 23% (calculated from three extraction replicates).

Chromatography

Crude oil is an ultracomplex mixture of saturated and aromatic hydrocarbons with a smaller fraction of N,S,O-containing compounds (Marshall & Rodgers, 2004; Hsu et al., 2011; Palacio Lozano et al., 2020). This complexity will be reduced but reflected in the produced water. Based on previous studies of Danish oils we know that the dominant oxygenated species belong to the O1 and O2 classes with a large diversity in aromaticity (Sundberg & Feilberg, 2020). Thus, the compositional variation in the produced water was assumed to be dominated by carbon number and level of saturation. Ultimately, the boiling point range was assumed to be larger than the variation in saturated versus aromatic structures. Therefore, we choose to use a non-polar column in the ¹D (providing the highest separation power) with a shorter medium polarity column in the ²D. Conventional polar columns are based on polyethyleneglycol (PEG) chemistry and are incompatible with silylation reagents. Therefore, we choose to use a 50%-phenyl-type column where aromaticity is the largest factor affecting retention. By using this column combination, the retention in both dimensions will increase with aromaticity. For example, cyclohexane acetic acid has a retention of 17.2 min/0.84 s in ¹D/²D as compared to 24.6 min/1.33 s for benzeneacetic acid (measured as the corresponding trimethylsilyl esters). In contrast, alkylation of a core aromatic or saturated structure will only affect retention in the ¹D. For example, phenol and its alkylated homologs (methyl, ethyl and propyl) have ¹D retention times of 13.4, 16.9, 19.9 and 23.3 min, respectively, where the retention in ²D is within 0.89 to 0.95 s. By investigation of a typical chromatogram, two things become obvious; (1) the desired separation of saturates, mono- and diaromatics is achieved and (2) a large portion of the ²D space is relatively uncopied due to the small number of polycyclic species where the majority of aromatics are benzene derivatives. A representative chromatogram with selected analytes marked is presented in Fig. 3.

A 3 °C min⁻¹ temperature gradient was found to be the optimal compromise between peak width, resolution and run time. At this rate, the typical peak width was 10 s in the first dimension. Thus, using a modulation period of 3 s allowed us to obtain the recommended minimum of three modulations per peak (Murphy, Schure & Foley, 1998). Due to instrumental complexity and long run times, retention time shifts are commonly observed both in inter-sample and inter-batch runs. Deuterated PAH standards were used to monitor retention times over time. The ¹D/²D retention time variability is presented in Table 1. Both ¹D and ²D were shown to be stable over the whole analyses run, covering 27 seven injections (nine samples with three extraction replicates) and 7 days.

Non-target screening and compound identification

Approximately 1,500 compounds were detected in each sample. A data processing workflow was implemented to sort, organize and score the data. A schematic representation is presented in Fig. 4. The associated files are available as Supplemental Information. The detected features were matched against the NIST EI Mass Spectral Library (2017; NIST v17, Software Version: 2.3). All compounds with a match factor below 700 were removed. To increase the annotation confidence, two additional factors were included; retention index (Kovats) and the presence of the molecular ion. To quantify the identification confidence the following scoring rules were implemented:

1. A match factor above 800 gives a score of 10 points.

2. A match factor between 700 and 800 gives a score of 5 points.

3. A retention index match within 50 units gives a score of 5 points.

4. The detection of the molecular ion within 20 ppm mass accuracy gives a score of 5 points.

A total score of above 10 was (i.e. match factor of a minimum of 700 and either a retention index and or molecular ion match) was classified as a probable match, whereas a total score below 10 was classified as a tentatively identified structure (Schymanski et al., 2014). To reduce experimental errors, only features that were present in all three replicates were included in the final table of compounds. Furthermore, all duplicate compounds were removed. This is a crude step that inherently removes, for example, isomeric species which would not be differentiated using automatic library search. However, for our purpose the identification of isomers is not an inherent goal. The aim of the study was to identify the presence of broad compound-class types and dominant species. The correct identification of isomeric species or species with highly similar fragmentation patterns requires manual intervention and is beyond the scope of this study.

The workflow reduced the number of features from approximately (1) 1,500 detected to (2) 200 library hits with match factor > 700, to (3) 50–70 when with duplicates based on name were removed. Ultimately, the merging of inter-sample features resulted in 120 unique hits across all samples. Of those, 42 had the maximum score of 20 (i.e. match factor > 800, molecular ion detected and retention index within 50 units). Near all (87%) identified compounds are oxygen-containing, with amines, sulfides and hydrocarbons being the remaining constituents. Only 15 compounds (after removal of internal standards and background species) were present in all samples (Table 2).