Rapid analysis of spilled petroleum oils by direct analysis in real time time-of-flight mass spectrometry with hydrophobic paper sample collection

Background Oil spills are widespread and can cause devastating environmental consequences. Rapid oil identification is critical to find the origin of the spill, monitor the environment, and lead to informed mitigation measures. The current standard methods in oil spill identification are precise and reliable, but require extensive sample preparation, long instrument runs, and time-consuming data processing. Direct analysis in real time time-of-flight mass spectrometry (DART-ToF MS) has been employed to screen for spilled petroleum oils, with results obtained in mere hours. The present study introduced an innovative, simple, and fast oil sampling method using hydrophobic filter paper and demonstrated its compatibility with DART-ToF MS analysis. Motor oils, jet fuels, marine diesels, crude oils, intermediate fuel oils, heavy fuel oils


Background
The demand for petroleum oil products persists despite efforts to mitigate greenhouse gas emissions by various renewable energy initiatives.The continuing widespread use of petroleum oils leads to extensive marine transport of these products.In fact, the Port of Vancouver in the Pacific Northwest has seen a six percent increase in mass of petroleum products handled between 2020and 2022(Vancouver Fraser Port Authority 2023).This region is home to a multitude of sensitive marine and intertidal species whose livelihoods are already threatened by the effects of climate change (Sutherland et al. 2013;Morales-Caselles et al. 2017).More oil shipments increase the risk of marine oil spills causing devastating effects on water quality and wildlife due to the high toxicity related to petroleum products (Ruberg et al. 2021).Petroleum oil contains significant levels of polycyclic aromatic hydrocarbons (PAHs), with some of the PAHs being carcinogenic and inhibitory to the development of several marine larvae and young fish (Boffetta et al. 1997;Morales-Caselles et al. 2017;Donohoe et al. 2021).Crude oil exposure is additionally associated with bird mortality and chronic flight impairment, digestive gland tissue damage in krill, mortality in brine shrimp, and impaired embryo-larval development in mussels and sea urchins (Perez et al. 2017;Moodley et al. 2018;de Santana et al. 2021).Therefore, to address an oil spill incident, a quick response is vital to contain the spread, protect wildlife, and maintain water quality.
Each oil type behaves differently in marine environments due to their unique chemical compositions.For example, in five days marine diesel can evaporate by 50% and disperse, whereas intermediate and heavy fuel oils do not evaporate as readily nor disperse (Emergencies Science Division, Environment Canada 1999).Aquatic toxicity also varies among oil types, with diesel showing higher toxicity than intermediate and heavy fuel oils.Furthermore, the application of chemical dispersants has the potential to increase ecotoxicity (Koyama and Kakuno 2004), and new biodispersants are emerging alternatives with unknown toxicity to aquatic species (Okeke et al. 2022).Considering the vast array of circumstances and potential emergency responses, an efficient and accurate oil type identification method is essential for developing suitable emergency measures to remediate spills.
Natural degradation processes, known as weathering, alter the physical and chemical compositions of spilled oils following their exposure to marine environments (Stout and Wang 2007;Reyes et al. 2014;CEN 2023).Clearly, weathering presents a significant challenge in oil spill identification due to the transformation and loss of many characteristic compounds.The weathering processes vary over time following a spill: dissolution, evaporation, and photodegradation dominate the initial stage of spilled oil.In the later stage, biodegradation influences the oil weathering process, decomposing high molecular weight PAHs and low molecular weight alkanes (Cook et al. 2020).Accounting for weathering effects in analytical chemistry oil identification methods ensures accurate matching of oil spills to their source.
The current standard for oil spill identification is described in the European Committee for Standardization (CEN) EN 15522-2 Oil Spill Identification guidelines (CEN 2023).Part of this internationally accepted gas chromatography (GC) mass spectrometry (MS) method compares biomarker ion response ratios between oil spill samples and potential oil sources.The indicative biomarkers are compounds in oil that are the most resistant to weathering processes.The CEN method contains 78 standard diagnostic response ratios between biomarkers.If the diagnostic ion ratios of the majority of the biomarkers between the source and spill differ by less than 14%, the spilled oil is deemed a match to the source oil.Over recent years, the methodology has benefited from the application of higher resolution quadrupole time-offlight (GC-QToF) instruments (CEN 2023;McCallum et al. 2023;Filewood et al. 2022).Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS) has also been used for successful oil analysis, and determination of spill origin (Terra et al. 2014;Hegazi et al. 2016).Overall, the oil biomarker ratio analysis is detailed, thorough, and suitable for legal cases (CEN 2023).Nevertheless, the CEN methodology is time consuming (up to three days in our hands when dealing with a median size oil spill incident of 16 samples) and needs extensive training for less experienced analysts.Sample preparation is long, labor-intensive, and tedious.Data collection is slow, with the GC-MS instrument taking one hour to run each sample, and a high level of expertise is required to process the data (CEN 2023).These limitations highlight the demand for a rapid and streamlined oil identification method for emergency spill response procedures.
Direct analysis in real time time-of-flight mass spectrometry (DART-ToF MS) is a powerful technique with the potential to address the above-mentioned challenges presented by the biomarker ratio analysis.Chemical characterization by DART-ToF MS requires minimal to no sample preparation.The instrument can quickly screen multiple samples and provide instantaneous spectral data (Easter and Steiner 2014;Espinoza et al. 2014).Several substances have been previously characterized by DART-ToF MS, including wood species, fleece fibers, and pharmaceutical drugs (Price et al. 2020(Price et al. , 2022;;Easter and Steiner 2014).The general DART-ToF MS forensic method entails comparing sample mass spectra to a library of reference spectra and conducting multivariate statistical analysis.For sample identification, a single mass spectrum measured at low collision energy provides a "fingerprint" of the material composition that can be searched against a database.Ion presence and relative abundance for replicate measurements of each sample and reference file are compared visually using heat maps (Price et al. 2022).The protocol for identifying wood by DART-ToF MS includes searching the Forensic Spectra of Trees (ForeST) database for most likely matching species, creating a heat map for the potential matches, then performing multivariate statistical analysis to match the sample to its assignment (Price et al. 2022).The instrument's data acquisition speed and streamlined data processing software make the DART-ToF MS an ideal tool for rapid chemical analysis, particularly for wood species identification (McClure et al. 2015;Price et al. 2022).Due to the presence of a multitude of biomarkers in fossil fuel oils and the timber products upon which the DART-ToF MS wood forensics method is based, this method is considered suitable for oil spill forensics.
Oil samples have previously been analyzed and matched to their source by DART-ToF MS (Tikkisetty et al. 2023a, b).A previous study by the authors' lab demonstrated the detection of biomarkers from DART-ToF MS data and the successful classification of unknown petroleum oil samples by heat map and multivariate statistical analysis (Tikkisetty et al. 2023a).Another study expanded the DART-ToF MS method to correctly match oil samples from actual oil spills to the correct source candidate (Tikkisetty et al. 2023b).The quality and authenticity of olive oil samples have also been determined by DART-ToF MS (Vaclavik et al. 2009).The results show great promise for oil spill forensics by DART-ToF MS.
Aquatic oil spill samples are generally collected in glass containers, often by immersion or dipping of glass bottles, followed by transport to the lab for analysis.The current study proposes an alternative and simpler field sampling method by using hydrophobic filter paper instead of a glass container for sampling, which is particularly suitable for DART-ToF MS analysis.Following an oil spill, hydrophobic filter paper can be immersed directly into the water at the spill location.The specialized paper repels seawater and retains the spilled oil.After sampling, the filter papers can be stored in a plastic bag and transported to a laboratory for DART-ToF MS analysis.While glass containers are bulky and fragile, filter paper samples remain flexible and easy to transport.Meanwhile, the authors were aware of the potential cross contamination risk presented by placing oiled paper samples within plastic bags and address the issue in this study.
The current study introduces a rapid screening methodology for oil type identification which is based on combining DART-ToF MS analysis and hydrophobic filter paper sampling.Reference petroleum oils were sampled using the hydrophobic filter paper technique and characterized via DART-ToF MS to generate Source oil data.Oil spills were simulated for five types of oil under two experimental situations, i.e., by creating small oil slicks in glass beakers and by exposing oil slicks to environmental conditions with an outdoor microcosm weathering experiment.The prepared oils were sampled using hydrophobic filter paper and analyzed with the DART-ToF MS.The first goal of the study was to develop a method for rapid oil sample collection and DART-ToF MS data acquisition using hydrophobic filter paper.The second goal was to verify the ability of the DART-ToF MS to classify un-weathered and weathered hydrophobic filter paper-collected oil slick samples using multivariate statistics.The final goal was to demonstrate the robustness of the screening method by examining various factors which may impact the outcome of the analysis such as oil weathering, plastic bag storage, and paper sample holding time in front of DART unit.

Reagents and samples
OmniSolv grade dichloromethane (DCM) and n-Hexanes were purchased from VWR (Mississauga, Canada) and were used to prepare a binary solvent of 40% DCM and 60% hexanes.Polyethylene glycol 600 (PEG 600) used for mass spectral calibration was supplied from Tokyo Chemical Industry (Tokyo, Japan).Whatman 1PS phase separator filter papers (70 mm diameter), high quality filter papers permeated with a stabilized silicone, were obtained from Millipore Sigma (Oakville, Ontario).The filter paper circles were cut into 16 sector-shaped pieces with 3.5 cm radius using stainless steel scissors.Twentynine oil samples were included in the classification models: six motor oils, marine diesels, and crude oils, four diluted bitumen, three heavy fuel oils, two jet fuels and two intermediate fuel oils.The motor oil category refers to petroleum-based engine lubricant products in the wider category of lubricant oils.Additional information about the oils is included in the Supplemental Information (SI) (Table S1).Sand filtered seawater was sourced at a depth of 33 m from the Burrard Inlet, BC, Canada, via an on-site pumping system at Pacific Environmental Sciences Centre (N 49.308051, W-123.0014) on December 15th, 2023, which was used to prepare the microcosm oil weathering experiment.From this location the seawater has a general salinity of 26 to 32 parts per trillion, hardness of 5000 to 6500 mg/l CaCO 3 , and pH of 7.6 to 8.0.

Oil source sample preparation and data collection
To prepare the Source oils, 0.4 ml of the original oil sample was dissolved in 4 ml a binary solvent of 40% DCM and 60% hexanes and stored in a clear 4 ml vial with a Teflon-lined lid (Millipore Sigma, Oakville, Ontario).The "Source" oil samples differ from the slick and weathered slick samples by the preparation methods summarized in Table 1.For sampling, the tip of a filter paper piece was inserted into the 4 ml vial and moved back and forth for three seconds to immerse it in the Source sample.Each Source replicate was run on the DART-ToF MS immediately after collection.A solvent blank was run by collecting mass spectral data for the binary solvent only, instead of an oil solution.Eight replicates were collected from all Source samples.Three solvent blank replicates were collected and run in between each oil category to monitor for background noise.

Oil slick sample preparation
Oil slicks were simulated in the laboratory by diluting 1 ml of select petroleum oils in 10 ml of a binary solvent of 40% DCM and 60% hexanes and pipetting the oil solution on top of 350 ml of seawater in a pre-cleaned 600 ml beaker (VWR, Missisauga, Ontario).The oil samples that were used to create slicks are referred to by their oil type and are listed in Table 2.The prepared samples were left in a fume hood for 30 min for volatile solvents to evaporate, creating a thin oil slick with a depth of one to two millimeters on top of the seawater.The 40% DCM and 60% binary solvent was selected to ensure that all oils would dissolve in the solvent and that the oil solution would form a uniform slick on the surface of the seawater.A seawater blank was prepared by placing 350 ml of seawater in a beaker.The room temperature in the laboratory was approximately 22 °C during the slick preparation and sampling.

Oil slick sample collection
Slick samples were collected within one hour of the slick preparation.A piece of hydrophobic filter paper was gripped by plastic tweezers at the arc.The tip was immersed vertically to a 0.5 to 1 cm depth into the oil slick, moved forwards and backwards, and retracted from the slick over the course of five seconds (Figure S1).The sample was then placed in a storage container.Three sets of five replicates from each sample and the seawater blank were collected: one set was placed in clear 7 ml glass vials with Teflon-lined lids (VWR, Mississauga, Ontario) for immediate transport to the DART ToF MS for data collection, one set was placed in the 7 ml glass vials for storage, and one set was placed in polyethylene plastic bags (16.5 × 8.2 cm (L × H), SC Johnson, Brantford, Ontario) for storage.Each glass vial held one filter paper sample (Figure S2).Five filter paper replicates of the same spilled sample were spaced out in a plastic bag from which the air was pressed out of before sealing (Figure S3).The filter papers remained spaced out in their bag during storage so that each filter paper and oil residue remained separated.All stored sampled remained inside of their container in a refrigerator at 6 ± 2 °C for 7 days prior to DART-ToF MS analysis.

Microcosm weathering conditions
The weathered slick oils were subjected to outdoor winter conditions of the British Columbia West Coast to induce weathering processes including dissolution, evaporation, and photodegradation occurring during a microcosm experiment.The previously prepared slick and seawater blank beakers were placed inside an aquarium tank [26 × 32 × 51 cm (W × H × L)] and the tinted tank lid was raised to create a 4 cm gap to allow air movement (Figure S4).The tank was moved to an outdoor area where it remained throughout the weathering process with placement under a high cover to prevent rainfall effects and allow direct sunlight exposure.The weathering experiment was conducted from December   2024).Weathered slick samples were collected like the slick samples by inserting the tip of the filter paper sector into the oil slick and moving it forward and backwards.Two sets of three replicates were collected from each sample and the seawater blank: One set was stored in glass vials and the other set was stored in plastic bags.All containers were stored in a refrigerator at 6 ± 2 °C for 10 to 19 days prior to DART-ToF MS analysis.

DART-ToF MS data acquisition
An AccuTOF-DART 4G mass spectrometer (JEOL USA, Inc., Peabody, MA USA) was used to record the mass spectra from the oil samples.Slick samples were removed from their container and gently wiped with a clean Kimwipe Delicate Task Wiper (Fisher Scientific, New Hampshire, United States) to remove the excess oil.Weathered slick samples were not wiped at this stage of the procedure due to a lack of excess oil.Each slick filter paper replicate was held vertically in the DART heated gas stream at 400 °C for ten seconds using metal tweezers while the AccuTOF component collected its mass spectrum (Figure S5) (refer to SI Table S2 for instrument conditions).When necessary, some weathered slick samples were held for less than 10 s due to the collected oil slick undergoing rapid expansion and dispersion in the gas stream.
At the beginning and end of each sample analysis run, mass calibration was performed by the instrument.Calibration was based on mass spectra collected for polyethylene glycol 600 (PEG 600) introduced to the instrument using closed end borosilicate glass melting point tubes (Fisher Scientific, New Hampshire, United States).For runs of slick and weathered slick samples, two seawater blank replicates were run prior to the oil replicates, and one was run after the sample.These blanks were used for baseline subtraction (detailed in next Section) and to visually monitor background noise.Using a DART source heater temperature setting of 400 °C, samples were analyzed in positive-ion mode via a DART-SVP ion source (IonSense, Saugus, MA, USA).The DART source heater temperature of 400 °C was selected based on previous studies (Tikkisetty et al. 2023a, b).The spectra were acquired over the range of m/z 70 to 1000 and were recorded at a rate of one scan per second.The helium flow rate for the DART-SVP source was factory-preset.
Prior to analysis, the ion source was positioned to leave a 2 cm gap between the insulator cap and orifice 1.Additional instrument parameters are listed in SI Table S2.
Eight replicate scans were collected for each Source oil sample for data processing and, due to limited availability, five replicates were collected from each slick sample and three replicates were collected from each weathered slick sample to monitor variation between samples.

Heat map generation
Total ion current chromatograms (TICCs) were collected and processed using msAxel@LP data processing software (JEOL USA, Inc., Peabody, MA, USA).The PEG 600 mass spectrum was used to compensate for drift in the sample data, and mass calibration was subsequently applied.Sample mass spectra were averaged over a time interval of approximately ten seconds corresponding to the maximum sample signal in the TICC.Baseline subtraction was performed when extracting each slick and weathered slick mass spectra using a corresponding seawater blank from the beginning of the run.Spectra were saved as centroided text files to allow for heat map construction using Mass Mountaineer (https:// massm ounta ineer.com).Heat maps summarize the mass spectra of multiple samples by visually depicting the relative abundance of ions (Price et al. 2022).In heat map figures, the x-axis represents the m/z values, and the pigmentation intensity increases with the relative ion abundance.Each row of a heat map corresponds to one sample replicate.For visual clarity, intensity can be adjusted for visual assessment without change to comparative relative intensity.Heat maps of the Source oils were generated to facilitate visual assessment of the mass spectra.Heat map inspection informed the selection of classes for the twostage classification models to be detailed in the following sections.Statistically significant ions were identified using variance analysis for the spectra of the Source sample heat map, which were used for the ensuing multivariate statistical analysis.

Multivariate statistical analysis
Ions were selected from the Source oil mass spectra to create classification models in the Mass Mountaineer software.Four models were created in total, i.e., one stage 1 model and three stage 2 models, the latter being used to classify closely similar oils.A principal component analysis (PCA) plot was constructed to demonstrate the similarity and difference between model classes based on their mass spectra collected from the DART-ToF MS.Discriminant analysis of principal components (DAPC) models with 10 mmu tolerance were built using the principal components from the corresponding PCA plot.
Leave-one-out-cross-validation (LOOCV) and external validation was performed on each DAPC model to assess its efficacy.External validation entailed randomly removing 30% of the samples from the stage 1 model's training data set and 20% of the samples from the smaller stage 2 models' training data sets, recalculating the models, and classifying the removed samples as if they were unknown.
The DAPC models were then employed to categorize the slick and weathered slick oil samples.The details of each statistical model are listed in SI Table S3.

Source oils mass spectra and heat map comparison
Previous studies have employed heat maps to visually examine the chemotype of petroleum oil samples and contrast different categories (Tikkisetty et al. 2023a, b).This "fingerprinting" technique has also contributed to successful identification of unknown wood species Distinct chemotypes were visually discernable from the heat map (Fig. 1).The motor oil sample spectra displayed relatively high ion abundances at specific m/z.These sharp characteristic ions may be attributed to the sensitivity of the DART-ToF MS to certain lubricant additives in the oil base (Shahnazar et al. 2016;Mangas et al. 2014;Chua et al. 2020).The jet fuel and marine diesel spectra populated a relatively low m/z range of approximately 100 to 300 m/z due to the removal of heavier hydrocarbons during refining process.Jet fuel mass spectra displayed a strong presence of expected lower mass ions and additional background (refer to filter paper and solvent blank responses).Background responses were attenuated under lack of ionization competition and were also detected on the heat map in motor oils and marine diesels at lower relative intensities.Heavy fuel oil displayed a characteristic bimodal distribution of ions between approximately 100 to 900 m/z.Intermediate fuel oil occupied a similar m/z range to heavy fuel oil and did not display as defined of a bimodal distribution.Crude oil and diluted bitumen spectra populated the greatest ion range, with some replicates containing ions ranging from approximately 100 to 1000 m/z.The overlapping ion ranges of crude oil and diluted bitumen are attributed to the congruous components of the two oil types.

Source oils multivariate statistics and model validation
The oil sample classification was executed in two stages to prevent similar oils from misclassification due to the complexity and similarity of many types of oil products.Each stage had its corresponding model, and model classes were determined based on similarities between oil spectra with the information from the heatmap (Fig. 1), as well as those used in a previous oil forensics study (Tikkisetty et al. 2023a).The stage 1 model contained four categories: motor oil, crude oil/diluted bitumen, heavy fuel oil/intermediate fuel oil and marine diesel/jet fuel.Consecutively, the samples were assigned to a stage 2 model depending on their stage 1 assignment.Multiple stage 2 models were built and used to process the assigned oil typing further.For example, one stage 2 model separated samples being assigned to the crude oil/diluted bitumen category during stage 1 by further assigning them to either crude oil or diluted bitumen classes.Another stage 2 model separated heavy and intermediate fuel oil.The last stage 2 model separated marine diesel and jet fuel.However, stage 1 classifications of motor oil were considered final due to the distinct spectral data of the Source motor oils and their uniqueness in comparison to the other Source oils.Each stage 2 model was constructed using the same Source oil data as the stage 1 model.For example, the Source spectral data used for the crude oil and diluted bitumen stage 2 model corresponded to the crude oil/diluted bitumen category of the stage 1 model and was represented in the crude oil and diluted bitumen sections of Fig. 1.
To construct each model, Mass Mountaineer software selected a set of 2000 features from recognized peaks in the corresponding Source mass spectra that were extracted for model building.Analysis of variance was performed for each m/z value between the two assigned classes with the largest difference between mean abundances for each m/z.The software then eliminated feature m/z's with p values over 0.05.This procedure was employed in established and published procedures for wood species identification (Espinoza et al. 2014;McClure et al. 2015) as well as oil spill forensics (Tikkisetty et al. 2023a, b).Features outside of the 70 to 500 m/z range were excluded because ions under 70 m/z are not detected by the instrument and mass spectral peaks above 500 m/z overlapped considerably when present in oil spectra.A heat map displaying the extracted masses used in the stage 1 model is included in the Supplementary Information (SI Figure S14a).
The features remaining after processing were used to generate a principal component analysis (PCA) threedimensional scatter plot for the comparison of oil categories and individual replicates (Tikkisetty et al 2023a, b).Clustering in PCA plots indicated similarities between points with the corresponding clusters and assigned oil categories illustrating statistical differentiation between the categories.A PCA plot was first constructed for stage 1 oil categories (Fig. 2).The assigned oil classes exhibited clusters with varying degrees of inter-class separation and intra-class variance, with the motor oil (light blue) and marine diesel/jet fuel (green) categories forming compact and separate clusters.The crude oil/diluted bitumen (dark blue) and heavy fuel oil/intermediate fuel oil (red) classes included more variance and overlap, yet still formed clusters.Based on these results, the two latter categories were expected to present a greater identification challenge due to their similar, isomeric, and isobaric hydrocarbon and hetero-hydrocarbon compounds, indicated by the slight intersection of clusters as well as the largely overlapping ion ranges in the heat map (Fig. 1).While the PCA plot successfully illustrated similarities and differences between oil categories, discriminant analysis of principal components (DAPC) was favored for classification due to its enhanced ability to differentiate between assigned classes.
From the selected Tier 1 model data, 75 principal components which contained 95.42% of the variance were used for DAPC.The number of principal components was selected to cover over 75% of the variance but less than 100% to prevent model overfitting.Plot distance between assigned classes was maximized and plot distance inside of each class reduced via the DAPC method.This type of model was employed to classify slick and weathered slick samples based on the improved distinction between classes.The DAPC scatter plot displayed compact clusters for all assigned oil classes (Fig. 3).The model was validated by LOOCV and by external validation (SI Figure S15 and SI Figure S16).The LOOCV score of 97.77% and external validation score of 96.23% validated the stage 1 model which accurately categorized data points within the four categories.The stage 2 models were constructed and validated in a similar manner with more specific Source data sets and achieved satisfactory validation scores (SI Figures S14b-d

Oil slick sample classification
The oil slick samples were analyzed on the DART-ToF MS immediately after their collection and results displayed consistent mass spectra.Five slick samples and associated duplicates (2), each analyzed as replicates (× 5), displayed visually similar mass spectra between each oil slick sample (Fig. 4).Replicates showed slight variation in the presence of heavier ions for the heavy fuel oil, diluted bitumen, and crude oil.Slight variation between replicates was inevitable due to differences in sampling and the positioning of the filter paper in the DART-ToF MS gas stream with ensuing differences in the relative amount of molecules ionized.The observed mass spectral range of the slick samples and their intrasample consistency (Fig. 4) corresponded closely to those of each respective Source sample (Fig. 1).Based on this heat map data it was concluded that the filter Similarly, statistical stage 1 and stage 2 modelling analysis of the experimental oil slick samples achieved satisfactory classification results, summarized in Table 3, and detailed in the Supplementary Information (SI Figure S26, SI Table S4, and SI Table S5).All replicates including duplicates were classified as the correct stage 1 category: crude oil/diluted bitumen, heavy fuel oil/intermediate fuel oil, marine diesel/jet fuel or motor oil (Table 3).The motor oil classifications were deemed conclusive, while the other oil types proceeded to their corresponding stage 2 model assessment.The marine diesel, heavy fuel oil, heavy fuel oil duplicate, and diluted bitumen replicates were all classified as their respective type by respective corresponding stage 2 model.In contrast, 20% of the crude oil replicates and 100% of the crude oil duplicate replicates were misclassified as diluted bitumen at this second stage.Overall, the DAPC classification models were effective, but the stage 2 separation of crude oil and diluted bitumen was found to be unreliable.This was unexpected since previous studies have shown the ability  of DART-ToF MS and the glass capillary technique to identify weathered spilled oils to their source (Tikkisetty et al. 2023a, b;Brunswick et al. 2024).However, one such study, with a similar two-stage model procedure, considered a stage 1 classification of crude/diluted bitumen to be final due to their mass spectral similarities (Tikkisetty et al. 2023a).Crude oil and diluted bitumen contain similar proportions of light ends, BTEX (benzene, toluene, ethylbenzene, and xylenes) and sulfur as well as display similar micro carbon residuals (Zhong et al. 2022).Thus, a chemotype-focused method, such as GC-MS, would more effectively distinguish oils with resembling chemical compositions.The considerable overlap observed between characteristic ions of crude oil spectra and diluted bitumen spectra inevitably caused the stage 2 misclassifications.It is well recognized that crude oil and diluted bitumen contain the very similar compounds in varying relative abundances (National Academy of Sciences 2016; Natural Resources Canada 2018; Zhong et al. 2022).The heat map in Figs. 1 demonstrated this chemotypic similarity between the ion ranges of crude oil and diluted bitumen Source replicates.The resemblance was further demonstrated by the similar m/z ranges populated by the crude oil and diluted bitumen slicks in Fig. 4. Nonetheless, the stage 1 model classified motor oils and effectively narrowed down all slick replicates to their correct dual-type category.Application of a stage 2 model was able to distinguish between marine diesel and jet fuel and additionally heavy fuel oil from intermediate fuel oil (Table 3).Both attributes can assist in a rapid emergency spill response procedure by narrowing down the oil type.

Paper based oil slick sample storage
Additional samples from the experimental oil slick that were stored for a week prior to DART-ToF MS analysis also showed satisfactory classification results.Storage in a glass vial was compared to the plastic bag storage method to explore any classification differences that could be attributed to the different container materials, such as leachate contamination from the polyethylene bag (Xu et al. 2020).A heat map of the stored oil slick samples displayed general consistency between replicates in comparison to their Source counterparts (SI Figure S27).Like their immediately analyzed counterparts, all stored replicates including duplicates were classified as the correct stage 1 category (Table 4).Additionally, the marine diesel, heavy fuel oil, heavy fuel oil duplicate and diluted bitumen replicates were all classified as their respective type by respective corresponding stage 2 model.At the second stage, 20% of the crude oil glass vial replicates, duplicate glass vial, and duplicate plastic bag replicates were incorrectly assigned to the diluted bitumen class (Table 4).Compared with immediate analysis (Table 3) the results for crude oil appeared improved following storage.The reason for this may be a combination of factors but there was suspicion that the data acquisition technique may affect the result, and this was investigated and discussed in the next section.Clearly, with the two-stage approach, short term plastic bag storage of paper-based samples would not impact the outcome of the DART-ToF MS oil classification analysis.
The consistently high classification accuracy for stored samples indicated that the DART-ToF MS method can realistically analyze hydrophobic filter paper oil samples collected in the field.The storage period of one week in the fridge mimicked the scenario of a sample being transported from the field to the laboratory in a cooler as well as sample turnaround time with DART-ToF MS analysis.Both glass vials and plastic bags yielded similarly high classification accuracy, indicating minimum impact to the results from the difference of container composition, plastic additives, or chemical leaches.With a run of 35 replicates requiring under 30 min of instrumentation and 30 min of data processing, and under one hour of the ensuing statistical classification, the DART-ToF MS Motor oil 100 Final classification achieved at stage 1 allowed for efficient, accurate and streamlined oil typing.Although less precise than the standard GC-MS and GC-QToF MS method, the paper sampling and DART-ToF MS analysis produced the results in a fraction of the time of the standard method and demonstrated the potential to identify all oil slicks to the category level and most oil slicks to the oil type level quickly and reliably using both glass and plastic container storage.Rapid oil typing is critical for environmental spill triage and confirmation by GC-MS can be performed subsequently, if needed.

Method development considerations
During method development it was noted that the filter paper time held in the DART gas stream affected the instantaneous data collected from the sample for certain oil slick replicates.As some filter paper samples were held for longer in the gas stream, more high m/z ions appeared in the mass spectra.These changes were examined and compared by extracting mass spectra at consecutive subintervals of approximately three to four seconds taken from the ten-second range of the actual TICC signal.Examining the diluted bitumen slick replicate, the abundance of heavier ions in the m/z 300-1000 range was relatively lower during the first subinterval than during the second and third subinterval and during the overall interval which was averaged over the course of ten seconds (Fig. 5).The crude oil slick mass spectral response increased in overall ion abundance from the first to the second sub-interval of its data collection time (SI Figure S28).The heavy fuel oil slick replicate increased in heavier ion intensity from the first to the second subinterval of its data collection time and increased in lighter ion intensity from the first to second and second to third subinterval (SI Figure S29).These replicates displayed consistent mass spectra at between ten to fifteen seconds in the gas stream during further exploration.By comparison, the marine diesel and motor oil samples displayed consistent mass spectra regardless of the sample DART ionization durations (SI Figure S30 and Figure S31).Nonetheless, it was important to hold all samples in place for ten seconds to acquire a fuller range of ions in the mass spectra and to maintain overall consistently, lest a shorter spectrum average fail to capture ions in a certain range.As such, the probability of correct model assignment for each replicate was increased by ensuring consistent, representative, and efficient mass spectra collection.The gradual increase in ions of high molecular weight compounds in the oil mass spectra may relate to the desorption and adhesion of various oil chemical components during DART ionization.The primary ionization mechanism of the DART-ToF MS entails the protonation of water vapor molecules in the sample gap by a heated stream of inert gas, then the proton transfer from water molecules to the analyte in the sample gap (Cody and John Dane 2010).Thus, it is possible that the less volatile molecules in petroleum oil require increased exposure to the gas stream to undergo positive ionization, desorb from the filter paper, and appear in the mass spectrum.Asphaltenes are complex and nonvolatile constituents of petroleum oil which are present in greater proportions in heavy crude oils and diluted bitumen (National Academy of Sciences 2016).It has been established that petroleum oils vary in their adhesion to pipeline walls and steel needles, and that their adhesion increases with asphaltene content (National Academy of Sciences 2016; Lyu et al. 2022;Jokuty et al. 1995).Through molecular dynamics simulation, models demonstrated that high asphaltene content leads to the formation of asphaltene aggregates which strengthen the adhesion of oil droplets to pipeline walls (Lyu et al. 2022).Therefore, asphaltenes may have had a similar effect on the adhesion of oil to and the desorption of oil from the hydrophobic filter paper, resulting in gradual release of the heavier ions for crude oils, diluted bitumens, and heavy fuel oils.It is further recognized that any glass capillary or filter paper matrix would also be preferentially absorbing the heat of the gas stream and again reducing the capacity of the system to volatilize the oil components.Thus, the importance of exposing each replicate to the gas stream for ten seconds was important to overcome any lower volatility and adhesion of heavier oils' chemical constituents, enabling collection of full mass spectra.Such adhesion and low volatility effects may be bypassed when the oil sample is diluted in a volatile solvent, such as the 40% DCM 60% hexanes binary solvent used in the study.In fact, the heavier ions appear quickly when collecting data from diluted Source oils (data not shown).However, solvents would be absent from natural spill samples, and the effects of heavy molecule adhesion can therefore be accounted for by ensuring that filter papers are exposed to the gas stream for a sufficient time during the data collection stage of oil slick classification.This allows the thicker film of oil on the filter paper to be fully desorbed.
As a result of method development study, every oiled filter paper piece from the simulated oil spill sampling was gently wiped with a clean Kimwipe to remove excess, wet oil prior to running the sample on the DART-ToF MS.This was performed to minimize the risk of excess oil expanding rapidly and forming a mist in the gas stream and contaminating the DART-ToF MS orifices, an issue which was observed in preliminary trials but was avoided by the addition of this step.This pre-wiping allowed the analyst to hold each filter paper in the sample gap for the full ten seconds.The filter paper pieces that were stored in vials retained more excess oil when removed from their container because the vials did not conform to the paper's shape.Plastic bags were sealed tightly such that the sides of the bag were in contact with the oil on the paper during storage and most of the excess oil remained adhered to the bag when the sample was removed.The wiping was less necessary for samples stored in plastic bags but was performed, for sampling consistency.For Fig. 5 The mass spectra obtained from the 10 s average and the three subintervals of the TICC signal of the first diluted bitumen slick replicate future streamlining, plastic bag storage was the most effective in minimizing excess oil, thus, reducing the need for wiping.

Weathered oil slick sample classification
The oil slicks were exposed to outdoor environmental conditions as a microcosm to explore the effects of weathering on DART-ToF MS classification of filter paper oil samples.Oils were sampled at weekly intervals over four weeks.Storage times for each sampling round were summarized in SI Table S7.Six time-point replicates were collected from each sample; three were stored in glass vials and three were stored in plastic bags.Mass spectral data were collected by DART-ToF MS, extracted in msAxel, compiled into heat maps, and classified by the previously constructed stage 1 and stage 2 DAPC models.Heat maps displaying all weathered spectra were provided in the Supplementary Information .Each round of classification took approximately two hours in total, including sample collection, data collection by DART-ToF MS, data processing in msAxel and multivariate statistics in Mass Mountaineer.
Classification by the two-stage model procedure resulted in successful identification for most of the weathered slick samples.All weathered slick replicates were assigned to their correct category by the stage 1 model (Table 5).The stage 1 model effectively categorized weathered oil samples and identified the weathered motor oil, whose stage 1 classification was considered final.After undergoing stage 2 classification, the marine diesel, heavy fuel oil, heavy fuel oil duplicate, crude oil, and crude oil duplicate replicates were all correctly identified as their respective oil type, while diluted bitumen results were less accurate (Table 6).Additional details of stage 1 and 2 weathered slick classification are available in the Supplementary Information (SI Figure S37 and Tables S7 and S8).The successful identifications were not compromised by the number of days weathered and days stored, nor the type of storage container, indicating that the sampling, data acquisition, and classification procedures were versatile and robust.
It is noted in Table 6 that the outdoor weathered diluted bitumen replicates were misclassified under stage 2 classification, with confidence of identification reduced over a longer time weathered.These misclassifications were likely caused in part by the loss of the low mass components, attributed to the addition of distils to the bitumen, to evaporation and decomposition by weathering.It is also noted that the formation of more solid diluted bitumen oil slicks during the weathering process further increased oil adhesion to the filter paper and inhibited the desorption of heavier ions.As a result, the weathered diluted bitumen mass spectra more closely resembled the crude oil Source data due to the decreased presence of heavier ions in the weathered diluted bitumen slicks when compared to the diluted bitumen Sources.The weathered slick investigation was performed before it was recognized that removal of excess oil by wiping off the excess oil would improve the mass spectral data consistency and therefore may contribute to sample classification accuracy.

Applications of the classification model
The developed classification models can direct spill response procedures by correctly narrowing down the category of a weathered oil and was proven able to identify motor oil, marine diesel, and heavy fuel oil from one another.In the case of a real-life oil spill analysis, the model could be used to categorize the oil spilled because suitable spill response procedures depend on the type of oil spill (Chen et al. 2022;Li et al. 2016).A wide variety of remediation routes are suited to lighter crude oils, such as marine diesel, while heavy and persistent oils can be more difficult to address.Such oils often require the use of high pressure or temperature washing, mechanical or manual removal, surface washing agents on shorelines, or manual recovery offshore (Chen et al. 2022;Li et al. 2016).Thus, by the DART-ToF MS ability to rapidly differentiate between general categories of weathered lighter and heavier oils, the models can inform and direct timely spill clean-up.Combined with the rapid data collection and processing by DART-ToF MS, the procedure is well suited to emergency oil spill response.

Conclusions
The DART-ToF MS combined with hydrophobic filter paper sampling allowed for the efficient identification of oil types of a variety of diluted, slick, and weathered petroleum oils.Mass spectra were quickly extracted and each Source oil produced chemotypes that were clearly visualized by heat map.Source oil chemotypes formed the basis of DAPC classification models which were used to classify oil samples collected from a variety of oil slicks.Most oil slicks and weathered oil slicks were correctly classified as their respective oil type using the current method, with the only misclassifications occurring between the chemically similar crude oil and diluted bitumen oil.The study accomplished three main goals: to develop a hydrophobic filter paper-based method for oil sample collection for DART-ToF MS data acquisition, to classify unweathered and weathered oil slicks as their respective oil type, and to examine the impact of glass and plastic storage containers, sample holding time, and weathering.The two-hour procedure time from sample collection to classification highlighted the potential of this method for rapidly screening oil samples when time is of the essence.The simple sampling methods, extended storage time, and glass and plastic storage container compatibility makes the hydrophobic filter paper a particularly practical oil collection tool; minimal training is required to sample a suspected oil slick and send the samples to the laboratory for DART-ToF MS analysis.
While additional developments may lead to the differentiation of crude oils and diluted bitumen, the developed paper-based sampling in tandem with DART-ToF MS provides a powerful addition to the toolbox of first responders, enforcement officers, and analytical chemists alike for combating devastating oil spill impacts to the marine environment.

(
Espinoza et al. 2014).Thus, heat maps were utilized as the first step in the chemical characterization of the oil samples in the current study and to provide the foundation for oil Source data model construction.Average mass spectra were extracted in the msAxel software for each Source replicate and compiled into a heat map using Mass Mountaineer software.A heat map of all the Source oils run during the study are shown in Fig.1.Seven types of oil are present: diluted bitumen, crude oil, heavy fuel oil, intermediate fuel oil, motor oil, jet fuel, and marine diesel.Twenty-nine samples were included, and eight replicates of each sample were run on the DART-ToF MS using hydrophobic filter paper.Filter papers and filter papers dipped in 40% DCM and 60% hexanes binary solvent were also analyzed to obtain paper blank and solvent blank mass spectra, respectively.Example mass spectra for each oil type and a solvent blank are included in the Supplementary Information (SI FiguresS6-13).

Fig. 1
Fig. 1 Positive ion heat map comparing types of Source oils included in the classification models, plus hydrophobic paper blanks and solvent blanks.Different oil types are separated by colored horizontal lines , and SI Figures S17  to S25).With the completion of model construction, the oil slick and weathered slick classification procedure could begin.

Fig. 2
Fig. 2 Three-dimension PCA scatter plot for the four oil categories included in the stage 1 classification model.The heavy fuel oil/intermediate fuel oil category is abbreviated as HFO/IFO

Fig. 4
Fig. 4 Heat map of all five oil slick samples and replicates which were run on the DART-ToF MS immediately following sample collection

Table 1
Summary of oil sample types and their preparation methods used in the study

Table 2
Summary of the oil samples used to prepare slicks and weathered slicks January 12th, 2024, during which the temperature reached a high of 12 °C and a low of − 1 °C(Past  Weather in North Vancouver, British Columbia, Canada The crude oil (ASMB4) and the heavy fuel oil (BKC1994) samples were duplicated, indicated by the addition of DUP in the abbreviation

Table 3
Rates of correct oil slick classification at stages 1 and 2Five replicates were analyzed for each sample.Heavy fuel oil and intermediate fuel oil model classes are abbreviated as HFO and IFO, respectively

Table 4
Rates of correct stored oil slick classification at stages 1 and 2, separated by oil sample and storage method Five replicates were analyzed for each sample and storage method

Table 5
Results of the stage 1 classification of the weathered slick sample replicates

Table 6
Results of the stage 2 classification of weathered slick sample replicates