Application of mass defect filtering and statistical analysis for non-target data mining of gas and soot data from a study testing different firefighting methods

(1)

Application of mass defect filtering and

statistical analysis for non-target data mining of

gas and soot data from a study testing different

firefighting methods

Danielle Ydstål

Supervisors: Anna Kärrman and Florian Dubocq 2020-10-11

(2)

2

Abstract

Due to the high temperatures during a fire event, a large variety of compounds are formed or released from burning materials, all of which have a varying degree of environmental effects. In an incidental fire there are several variables that are important for which and how much combustion products that are formed, including the burning material, ventilation (presence of air), and temperature.

The aim of this project is to evaluate if there is a difference between formed fire residues in gas and soot using four different fire extinguishing techniques. A non-target approach is used with gas chromatography connected with ultrahigh-resolution mass spectrometry. Unlike target analysis, non-target analysis enables identification of not only known chemicals, but also previously unknown chemicals. However, one of the major challenges in non-target analysis is how to handle the large amount of data generated in order to identify important markers for the current research question.

Mass defect filtering is used to interpret the complex mass spectral data. Plotting the mass defect against the measured m/z allows you to visualize a high number of mass spectral peaks, linking homologues and congeners. The plot is based on a specific mass scale and can be used to find m/z that belong to compounds of a specific compound group. Statistical methods such as Principle Component Analysis (PCA) are also useful as it extracts and displays systematic variation in a data set, which can be used to find interesting variables.

Mass defect filtering proved to be useful for the detection of a number of different compound groups: Alkylated hydrocarbons, halogenated compounds and PAHs. There were several differences in the composition of the gas versus soot. Gas had little variation between the samples whereas soot varied more depending on firefighting method used. Despite the fact that the chemical composition of gas and soot does differ between the four firefighting techniques, the variations in wind conditions made it hard to draw any conclusions regarding how the different firefighting techniques affect the compound formation and to what extent.

(3)

3

Introduction

One of the goals of the Swedish Civil Contingencies Agency (MSB) (former Swedish Rescue Service Agency) is to limit damage to properties and environment (Lönnermark et al., 2007). Their research covers, among other things, health and environmental effects during and after incidental fires. Due to the high temperatures reached during a fire event, a large variety of compounds are formed or released from burning materials, all of which have a varying degree of environmental effects (Lönnermark et al., 2007). The combustion gases formed during a fire event can travel long distances through the atmosphere and eventually be deposited back to the ground through dry and/or wet deposition. The distribution and potential environmental hazards of fire residues calls for firefighting techniques that enable fast and efficient fire extinguishing, which makes firefighting techniques an important research area.

New firefighting methods, some using different additives, are constantly being developed (Bjurlid et al., 2017). At the same time, new chemicals are being used in building materials and we are increasing our use of for example electronic components and plastics. Our increasing use of chemicals can, during incidental fire events, affect compound formation both from thermal decomposition and de novo synthesis. Incidental fire events can give rise to a broad range of combustion products, one example being brominated dioxins (Bjurlid et al., 2017). Therefore, there is also an increasing need for identification of a broader range of compounds in fire residues.

It is possible that some important compounds are excluded when a pre-selection of chemicals are analyzed in so called target analysis, due to the selectivity of the chosen sampling method, extraction and analysis. Unlike target analysis, non-target analysis enables identification of not only known chemicals, but also previously unknown chemicals. One of the major

challenges in non-target analysis is how to handle the large amount of data created, in order to identify important markers for the current research question. One approach for finding

important markers is data mining using mass defect filtration, which is the method used in this study together with statistical analysis.

Aim and objectives

The aim of this project is to evaluate if there is a difference between formed fire residues in gas and soot using four different fire extinguishing techniques. A non-target approach was used with gas chromatography connected with ultrahigh-resolution mass spectrometry to address the following objectives:

• What kind of non-target data analysis methods are useful for the identification of unknown compounds in fire residue samples?

• Are there unknown compounds/compound groups in the fire residue samples? • Are there differences between which compounds are identified in soot and gas

samples?

• Does the chemical composition of the gas and soot differ depending on firefighting method used?

(5)

5

Background

The main mechanisms behind fire extinguishment are a significant decrease in temperature of the flames and that the concentration of pyrolysis gases becomes low (Särdqvist 2002). There are different methods for obtaining this, the most common is water applied in different ways (jet, spray or fog) but powder and foam are also frequently used. Temperature is important when it comes to formation of compounds from the burning materials. Typical incineration processes are controlled and optimized to yield a complete combustion and minimizing the formation of by-products. In an incidental fire there are several variables that are important for which and how much combustion products that are formed, including the burning material, ventilation (presence of air), and temperature (Blomqvist et al., 2002). Substances present in materials can also become volatile due to high temperature and be emitted into air. Different firefighting methods influence the rate at which a fire is extinguished and how fast the temperature drops and will thus affect the formation of by-products.

A large range of substances have been identified in emissions from fires, depending on the materials involved. Two of the most common incineration products from synthetic and natural materials are polycyclic aromatic hydrocarbons (PAHs) and volatile organic compounds (VOCs). When plastics and electronics, which contain halogenated compounds, are present during a fire event, complex mixtures of unidentified and environmentally persistent

contaminants are formed (Myers et al., 2014b). One of the known compound group that can be formed is chlorinated dibenzo-p-dioxins and furans (PCDD/Fs).

Dioxins are extremely toxic to humans and the environment and they can be formed from PCBs, chlorophenols and chlorobenzene during a fire. PCBs can be found in for example certain plastics and solvents for painting. Chlorophenols may be present in wood due to wood impregnation agents (which are banned in Sweden, but not in all countries)(Bengtsson & Antonsson, 1993).

VOCs are a wide variety of compounds that can easily be distributed throughout the

environment, mainly in air, due to their low boiling point and high vapor pressure. These are properties that allow them to be transported long distances. VOCs are emitted from textiles, paints and other building materials and are thus expected to be present during an incidental fire.

PAHs are a group of organic compounds with generally high melting and boiling points (Abdel-Shafy & Monsour, 2016) that are formed through incomplete combustion of organic materials such as wood and plastics. The more complete combustion, the lower the formation of PAHs (Bengtsson & Antonsson, 1993). PAHs are ubiquitous in the environment as they are commonly found in both air, soil and water. However, air is the most common transportation medium. In air, they are found in vapor phase or solid phase (sorbed onto particulate matter). The vapor pressure of PAHs is low, but varies with the size of the compound where lower mass commonly means higher vapor pressure. The higher the vapor pressure, the more likely the compound is to be found in the vapor phase. Thus, the concentrations of lower molecular weight PAHs is usually higher in vapor phase (Abdel-Shafy & Monsour 2016). PAHs are known carcinogens (Bengtsson & Antonsson, 1993; Abdel-Shafy & Monsour, 2016) and have

(6)

6 mutagenic and immunosuppressant effects (Abdel-Shafy & Monsour, 2016). Generally, when speaking of PAHs, the elemental composition is limited to carbon and hydrogen, but there are for example chlorinated aromatic hydrocarbons that can form if PVC plastics or other

chlorinated substances are present during combustion (Bengtsson & Antonsson, 1993).

Non-target analysis

When focus is on a number of selected compounds, such as PAHs, VOCs or dioxins, a selection of sampling, extraction and instrumental methods are done prior to analysis of the sample. This means that information on all untargeted and unknown compounds are lost even though they may be present in far higher levels than the selected compounds (Tengstrand et al., 2012). Non-target analysis is a method for identifying compounds without performing a preselection when it comes to sample preparation and instrument analysis. It is commonly used as a qualitative method and it enables identification of unknown compounds without analytical standard. Instead, to aid identification, the exact mass, molecular formula and structure is used (Krauss, Singer and Hollender, 2010). Non-target data can also be used to extract specific target compounds, and the term used to describe this is suspect screening (Zushi, Hashimoto and Tanabe, 2016). There are many potentially hazardous compounds that are found in the environment that lack analytical standard, in which case, non-target analysis can be used.

When studying organic pollutants one is faced with a wide variety of different compounds with different volatility and polarity. In that case, separation of the compounds using chromatographic techniques such as liquid chromatography (LC) and gas chromatography (GC) can be useful. Both LC and GC can be coupled with mass spectrometry (MS), which involves ionizing the analytes using an ion source before separating the ions based on their mass-to-charge (m/z) ratio using a mass analyzer and, lastly, detecting the ions in a detector. The detection of the ions is performed both qualitatively (by m/z) and quantitatively (by abundance). Depending on the type of MS used, different sensitivity and mass accuracy can be achieved. High resolution mass spectrometry (HRMS) is popular for environmental studies as it is suitable for both targeted and non-targeted analysis due to the high mass accuracy and sensitivity in full-scan acquisition mode. It can for example be used for structure elucidation and in the petroleum industry. However, HRMS has recently been replaced by the new concept of ultra-high-resolution MS. The resolving power of the MS is proportional to its capacity to discriminate between signals from ions of similar m/z in a mass spectrum. (Hernández et al., 2012)

In order to identify a compound from mass spectral data without an analytical standard one has to be able to distinguish the analyte from the matrix background. A high mass spectral accuracy is of importance, since the exact mass is used to determine the elemental

composition of the compound (Krauss, Singer and Hollender, 2010). To achieve the mass accuracy needed for identification, high resolution mass spectrometry (R<10,000 (Cariou et al., 2016)), such as Time-of-Flight (ToF) or ultrahigh-resolution mass spectrometry, such as Fourier transform ion cyclotron resonance (FT-ICR) and Orbitrap mass spectrometry, can be used. Such instruments provide full-scan data and operate in a mass resolution of up to one

(7)

7 million (FT-ICR) (Zushi, Hashimoto and Tanabe, 2016). A high mass resolution means more peaks can be resolved and a more accurate mass measurement can be performed, separating compound which differ with only a few ppm in m/z. One draw-back with high resolution mass spectrometry data is the large data set generated. Environmental samples may have several thousands of data points since each molecular ion present can yield a large number of fragments when using hard ionization techniques such as electron ionization. The large number of data points makes interpretation of the results difficult, even more so as one need to separate the molecular ion from the fragment ions.

However, one advantage of electron ionization is that highly reproducible spectra are generated. Such spectra can be used for identification of compounds in the sample by

searching mass spectral libraries for a match (Hernández et al., 2012). Large libraries of mass spectra can certainly be a useful tool when performing non-target analysis, since it can be otherwise time-consuming and laborious to filter out the exact mass, assign an elemental formula and search for possible structures, all in complex matrices.

To sort a large set of data and extract potentially useful information, manual data mining methods can also be applied. Data mining allows for detection of relationships and patterns within a data set, which can lead to the discovery of previously unknown information

(Thomas et al., 2006). One tool for data mining of mass spectral data, especially for detection of halogenated compounds, is using mass defect.

Mass defect

Previous studies have shown that mass defect filtering has been very useful when categorizing complex environmental samples (Myers et al., 2014b), especially when it comes to thermal decomposition of fluoropolymers. Mass defect filtering is used to interpret the complex mass spectral data generated using high/ultra-high resolution techniques. In the field of mass spectrometry, mass defect is defined as the difference between the exact (monoisotopic) mass of any species and its integer mass (Thurman and Ferrer, 2010).

The International Union of Pure and Applied Chemistry (IUPAC) mass is used to calculate a molecular mass. That mass scale is based on 12C being 12.0000 Da, but a mass spectrum can be rescaled from the IUPAC mass scale to a scale with different base. One such scale is Kendrick’s mass (KM) scale which is based on CH2 being 14.0000 Da. The rescaling is done

using Equation 1 below (Hughey et al., 2001).

(1) The KM is then used to calculate the Kendrick’s Mass Defect (KMD) as seen in Equation 2 (Hughey et al., 2001).

(2) However, the mass scale can be based on other compounds/compound groups than KM scale (CH2) to get a mass defect which is specific for that compound/compound group. The

𝐾𝑀 = 𝐼𝑈𝑃𝐴𝐶 𝑚𝑎𝑠𝑠 ×𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑚𝑎𝑠𝑠 𝑒𝑥𝑎𝑐𝑡 𝑚𝑎𝑠𝑠

(8)

8 nominal mass divided by the exact mass (in Equation 1) is called mass defect ratio and is specific to the mass scale used. If another scale than Kendrick’s mass scale is used the denotation KM can be replaced by mass scale and KMD replaced by mass defect (MD). For identification of brominated compounds for example, -H/+Br scale can be used instead of CH2 (Ubukata et al., 2015). The -H/+Br and -H/+Cl scales are close enough to be used

interchangeably, since their mass defect ratios are 0.0011486 and 0.0011476 respectively. An overview of some mass defect scales is given in Table 1.

Plotting the MD against the measured m/z allows you to visualize a high number of mass spectral peaks, linking homologues and congeners (Myers thermal et al., 2014). Each plot, based on a specific mass scale, can be used to find m/z that belong to compounds of that specific compound group. The m/z with no particular MD, belonging to compounds

containing only H, C, N and O, are clustered closely in the lower part of the plot and can thus be filtered out (Cariou et al., 2016). This is useful for identification of halogenated

compounds since they have a negative mass defect which places them away from the main cluster, even though they can span the whole MD range depending on the remaining chemical composition. The mass defect plot has proven useful for the discovery of novel halogenated compounds (Myers et al., 2014b).

Homologue series are important in the identification process. They are composed of a number of molecules (homologues) that have the same MD, and are thus aligned on a horizontal line in the plot, but differ from each other with a specific m/z (Hughey et al., 2001). The

homologues have the same chemical formula, except for a scale-specific part. For detection of brominated compounds, based on a –H/+Br scale, the part of the molecule that differ between homologues is -H/+Br, which equals a m/z difference of ±78. In the same way, the

homologue series of a PAH would differ m/z ±50 between two homologues, which is the sum of an aromatic ring being added or removed from another aromatic ring resulting in addition or subtraction of C4H2.

For monoisotopic species (A), each homologue is composed of one m/z. Compounds that have stable natural isotopes however have two or more m/z for each homologue. These isotopes can be seen as A+1 or A+2 ions, forming a pattern of m/z values with 1 or 2 Da apart from the most abundant ion (A) (McLafferty and Turecek, 1993) at approximately the same MD. Bromine and chlorine show distinct isotopic signals at the A+2 ion (Krauss, Singer and Hollender, 2010). They both have two isotopes with significant abundance that differ with two neutrons, which corresponds to a nominal mass of 2 Da (Cariou et al., 2016). The isotopic pattern is very distinct for the specific atom and differ depending on the number of that atom that is present in the molecule, see Table 1 for more details. For example, assume a compound containing three Br atoms are present. The molecular ion (A) will be found at roughly the same MD as the A+2, A+4 and A+6 ions, all with two m/z apart. The intensities of these four points will be approximately in the ratio 1:3:3:1. If both Cl and Br are present in a molecule the ratios can be multiplied (Cl (3:1) x Br (1:1) = ClBr (3:4:1)) to create the expected new ratio, as both Cl and Br has an A+2 isotope.

By studying high resolution mass spectral data one can learn a lot of valuable information about an unknown compound. Isotopic abundances and fragmentation patterns can give

(9)

9 information about the composition of a compound – allowing one to determine the molecular formula and possibly even the structure (McLafferty and Turecek, 1993).

Table 1. Examples of mass scales used in mass defect filtering, together with the compound groups they can be used to detect, which area of the plot they commonly occupy, homologue series pattern and isotopic pattern. The isotopic pattern of aliphatic hydrocarbons has been excluded here as it is not used as a primary aid for identification.

Compound group Mass scale Area of plot Homologue series Isotopic patterns

Aliphatic

hydrocarbons CH2 MD<0.1 m/z difference of 14 -

Chlorinated -H/+Cl MD varies depending on

remaining structure m/z difference of 34

Cl: 3:1 Cl2: 9:6:1 Cl3: 27:27:9:1 Cl4: 81:108:54:12:1

Brominated -H/+Br MD varies depending on

remaining structure m/z difference of 78

Br: 1:1 Br2: 1:2:1 Br3: 1:3:3:1 Br4: 1:4:6:4:1 Aliphatic fluorinated CF2 MD varies depending on

remaining structure m/z difference of 50 -

Fluorinated HF

MD varies depending on remaining structure and how partially fluorinated

the compound is m/z difference of 20 - PAHs C4H2 MD>0.9 m/z difference of 50 - Halogenated PAHs -H/+Br or -H/+Cl MD≈0.6-0.8 m/z difference of 34 (Cl) or 78 (Br) Statistical analysis

Full-scan mass spectral data that contains a large number of variables (m/z) are often very complex and hard to interpret. Therefore, application of statistical methods is necessary. Principle Component Analysis (PCA) is a projection method that is commonly used in

multivariate data analysis. It extracts and displays systematic variation in a data set, which can be used to find interesting variables. A multivariate data table containing mass spectral data can be represented as a low-dimensional plane, usually consisting of 2-5 dimensions, to get a better overview of the data. In this way, trends, outliers or groupings of observations (for example different firefighting methods) or variables (for example mass spectral data points) can be disclosed. Also, associations between the variables and the observations can be studied. This is done using scores plots (which display the observations) and loading plots (which display the variables). Those two plots are based on a number of principal components (PCs). The first PC is computed by drawing a line through the data in the low-dimensional space which intersects origin (if mean centered). The direction of the line should be adjusted to fit the data as well as possible and thus maximizes the variance of the co-ordinates on that line. The second PC is orthogonal to the first, the third PC is orthogonal to the first two, etc. (Eriksson, 2013).

(10)

10 In some cases the results from the PCA is not fully satisfactory. This could be due to a large number of clusters that may not be distinctly separated from each other. In these cases, hierarchical cluster analysis is useful, as it can detect not only prominent but also subtle clusters (Eriksson, 2013). One way to visualize the clusters produced by cluster analysis is through a heatmap dendrogram (Ivanisevic et al., 2014). A heatmap dendrogram aids in comparison of samples through clustering and detection of the features that have the highest impact on the clustering. The heatmap displays the lowest values in the dataset as one color, for example blue, and the highest values as another, for example red, and all other values are assigned a color in between to form a color gradient. The dendrogram is a result of a

(11)

11

Materials and method

Samples, sampling and sample handling

Details on the tests performed in 2015 to get the soot and gas samples that yielded the data used in this study are presented in Table 2. The samples were collected and processed as described in Bjurlid et al. (2017) and analyzed using FTICR-MS by Ontario Ministry of the Environment and Climate Change, Toronto, Canada. No part of the sampling, sample

preparation or analysis was done during the present study. However, the sample extracts were re-run in this study together with analytical standards to confirm the presence of tentatively identified compounds.

Table 2. Details for each test, specifying sample matrix, firefighting method and extinguishing agent.

Test Sample

matrix Firefighting method Extinguishing agent

1 Soot Nozzle Water spray

Gas

2 Soot Coldcut cobra cutting extinguisher Water fog

Gas

3 Soot Compressed air foam system (CAFS) Foam

Gas

4 Soot Coldcut cobra cutting extinguisher + additive Water fog with X-fog additive Gas

5 Soot Coldcut cobra cutting extinguisher + additive Water fog with X-fog additive Gas

In short, the test site was the training facility Guttasjön, Södra Älvsborgs

Räddningstjänstförbund (SERF) in Sweden. The tests were performed during three days in January 2015 using a 20 feet container with double plasterboards on the interior walls. The different techniques were tested using an identical set of living room interior and furniture, with devices such as television, laptop and other electronic equipment. The room was heated using a heat cannon to better represent a real apartment (as the outdoor temperature was subzero at this time) and one door of the container was open during the whole fire and extinguishing event. The ignition was done using ignition wood blocks, which is the same technique as the test facility uses. Extinguishing was performed by a fire-fighter trained in all four techniques, who used their regular routines for each technique.

Samples were collected separately during the fire phase (from ignition to flashover) and the extinguishing phase (from about 10 seconds after flashover, during 5 minutes). Only the soot and gas samples collected during the extinguishing phase were used in this study.

For sampling of combustion gases, active sampling using a low volume pump was performed. Air was pumped through connective glass tubing through the container wall 2.2 meters above the floor. The glass tubes were connected to a sampler containing XAD-2 adsorbent (20/50, 400/200 mg, Supelco) on the outside of the container. The particle phase of the air was filtered out using a Teflon pre-filter.

(12)

12 For the sampling of soot particles, pre-cleaned plates made of stainless steel were used, one for the fire phase and one for the extinguishing phase. The samples were collected in the top part of the doorway of the container and the soot was wiped off the whole 0.21 m2_plates

using Kleenex tissues.

Chemicals and materials

A detailed description of the chemicals and materials used for sampling, extraction and clean-up of the gas and soot samples can be found in Bjurlid et al. 2017, as it was not done as part of the present study.

However, additional confirmatory analysis was made in the present study, using the same sample extracts as prepared and analyzed in Bjurlid et al. 2017, together with standards. The standards used were Tetrabromobisphenol A (TBBPA) in methanol (Wellington) and a halogenated PAH standard containing 19 chlorinated and 19 brominated PAHs in toluene (previously prepared as described in Jin et al., 2017).

Extraction and clean-up

As mentioned, the sample preparation was not performed during the present study, but previously performed as described by Bjurlid et al. 2017. In short, soot and gas samples were extracted using Soxhlet extraction with toluene during 24 h reflux and the clean-up was performed in a series of steps involving three different open columns: multilayer silica, AlOx and active carbon. The extracts were then stored in -20 ºC until analysis using FTICR-MS by Ontario Ministry of the Environment and Climate Change, Toronto, Canada.

Analysis

GC-FTICR-MS

The full scan analysis was previously performed using Gas Chromatography Fourier Transform Ion Cyclotron Resonance Mass Spectrometry (GC-FTICR-MS) by Ontario

Ministry of the Environment and Climate Change, Toronto, Canada. For separation, a Varian CP-3800 GC equipped with a DB-5HT GC column (15 m × 0.25 mm ID × 0.10 µm, J&W Scientific, USA) was used. To obtain the ultrahigh resolution mass spectra used in this study, a Varian 920 FTICR-MS, positioned in a Varian 9.4 Tesla superconducting magnet, was used. The information about the resolution was not noted at the time of analysis and thus the

information has been lost, but previously published articles where the author has been using the same instrument state a resolution of 40 000 at m/z 400 calculated at full width half maximum (FWHM), and there is no reason to suspect it to be lower. The ionization technique used was electron ionization (EI) in positive mode.

The instrument was tuned before analysis and a lock mass was used during the analysis. Post-calibration was performed on the data used to create mass defect plots, in order to optimize the mass accuracy. Two different masses, both originating from substances from the

(13)

GC-13 column (column bleed) were used for the post-calibration, (m/z 281.05114 and m/z

419.31560).

GC-EI-Q-Orbitrap

To confirm the presence of tentatively identified compounds, the gas and soot extracts analysed by Ontario Ministry of the Environment and Climate Change, Toronto, Canada, were also run together with standards in the present study using Gas Chromatography Electron Ionization Q Exactive Orbitrap Mass Spectrometry (Q Exactive GC, Thermo

scientific, Bremen, Germany). Electron Ionization was used in positive mode at 70 eV. A full scan with the m/z range 50-600 was performed with a resolution of 120,000 FWHM at m/z 200.

Identification and structure elucidation

Step 1 – mass defect plot filtration. The post-calibrated (using column bleed signals) high

resolution mass spectral data retrieved from the analysis made by Bjurlid et al. 2017 was rescaled in Excel either using CH2 mass defect scale to detect hydrocarbon features, or other

mass defect scales to detect halogenated and polycyclic aromatic compounds. These scales were calculated according to Equation 1, where IUPAC mass is the measured m/z. As the mass defect of Br and Cl is almost identical, the –H/+Br scale was used to find both Br and Cl-containing compounds. The new mass scale was then used to calculate the MD (Equation 2).

One mass defect plot was generated for each mass scale, where each signal in the GC-MS analysis is plotted with the MD on the y-axis, measured m/z on the x-axis, and intensity shown as the size of the bubble. The m/z values that were found in the relevant parts of the plots (specified in Table 1) were studied further. For example in the CH2-based scale, the

lowest part (between MD 0 and 0.1) of the plot is relevant, as compounds composed of only C and H have a very low KMD. In addition to what specific part of the plot the m/z occupies, the isotopic patterns (for bromine and chlorine containing compounds) and homologue series are important.

Step 2 – mass spectral interpretation. The m/z values found in step 1 were studied further

in their corresponding mass spectra using the program Omega FTMS (Varian). In this step, the data used was raw data files that had not been post-calibrated. From the raw data, the mass spectra of the m/z found in step 1 was extracted from the selected ion chromatogram. A mass accuracy tolerance of up to 50 ppm was necessary since raw data files had not been used during step 1. The mass error tolerance level needed to be carefully selected for each signal since a too low tolerance tended to filter out the chromatographic peak belonging to the molecular ion and a too high tolerance resulted in that more than one peak was included and the peaks were not baseline separated. The data used did not have a combined peak spectra, so all m/z features, both molecular ions and fragment ions, were studied individually.

The mass spectra was used to get the accurate mass, confirm isotopic patterns observed in the mass defect plots, look for fragment ions and neutral loss fragments. During the EI ionization

(14)

14 method, multiple fragment ions are generated, thus the molecular ion intensity can be low, sometimes even below the detection limit. The fragment ions are fragments of the molecular ion formed in the ion source. The neutral loss fragments are neutral fragments that are lost by fragmentation of the molecular ion or a fragment ion. They are found by looking at the mass difference between the molecular ion and a fragment ion. Both the fragment ions and the neutral loss fragments hold information on the composition and structure of the molecular ion.

Step 3 – elemental composition calculation. A molecular formula was suggested based on

the accurate mass obtained in step 2. All the molecular formulas given in the results were determined from the ultrahigh resolution data by using the “Elemental Composition

Calculator” tool in the software MassLynx (Waters Corporation). The maximum mass error tolerance was ±10 ppm, the Double Bond Equivalent (DBE) between 0 and 15 and included were carbon, hydrogen, nitrogen (0-5), oxygen (0-10), phosphorous (0-3), fluorine (0-5), bromine (0-4) and chlorine (0-4).

Step 4 – proposing a possible structure. Based on the molecular formula determined in step

3, possible structures were proposed using databases such as Chemspider and Comptox by the United States Environmental Protection Agency (EPA). Still, to confirm the structure,

additional analysis using different complementary ionization techniques could be used to get a strong signal from the molecular ions and to further study the fragmentation pattern. Also needed to confirm a structure are analytical standards.

(15)

15

Statistical analysis

The full-scan raw data was analyzed using XCMS Online (The Scripps institute). The settings used for preprocessing of the data and the statistical analysis are seen in Table 3. For

multigroup comparison, that is tests comparing more than two different groups, Analysis of Variance (ANOVA) was used. For pairwise comparison, the Welch t-test was used (unpaired parametric, unequal variances). The remaining settings were the same for both multi- and pairwise comparison. The robustness of the results may however be limited due to the small number of samples analyzed.

Table 3. Settings for preprocessing of the data and statistical analysis. Multigroup and pairwise comparison was made using XCMS Online. Included in the table are also descriptions of each setting used.

Parameter Description Multigroup Pairwise

General Polarity Mode of data acquisition Positive Positive

Feature detection

Method Feature detection method Matched

filter

Matched filter FWHM Full width at half maximum of matched

filtration Gaussian model peak 30 30

S/N threshold Signal to noise ratio cutoff 3 3

Max Maximum number of peaks per extracted ion

chromatogram 10 10

Step Step size to use for profile generation 0.1 0.1

Steps Bin size (m/z).

Larger = smoother, smaller = finer 2 2 m/z difference Minimum difference in m/z for peaks with

overlapping RTs 0.01 0.01

Retention time correction

Method Retention time correction method Obiwarp Obiwarp ProfStep Step size (m/z) to use for profile generation

from the raw data files 0.1 0.1

Alignment

Grouping

method Density Density

Bw Bandwidth: Tolerated RT deviations (sec) 5 5 m/z width

Width of overlapping m/z slices to use for creating peak density chromatograms and

grouping peaks across samples

0.025 0.025

Minfrac

Minimum fraction of samples necessary in at least one of the sample groups for it to be a

valid group

0 0

Statistics

Statistical test Statistical test method. Can be paired/unpaired and parametric/non-parametric ANOVA parametric Welch t-test parametric p-value threshold

Highly significant features have a p-value

lower than this 0.01 0.01

Fold change threshold

Highly significant features have a fold change

greater than this 1.5 1.5

Diffreport value

Into= integrated peak intensities are used

(16)

16 The analysis using XCMS Online presents the results in a table, cloud plot, PCA and

heatmap. In this work only the PCAs and heatmaps generated were used. The scores plots of the PCA, which show the observations (in this case firefighting methods), was used to find clustering among the firefighting methods and evaluate if the replicates were grouped together. The loading chart, which display the variables (measured m/z), was used to spot outliers and in connection with the scores plot the outliers could be connected to a certain firefighting method. The heatmaps were used in order to find clusters that were not clear in the PCA to better be able to compare the observations.

QA/QC

The data used in this study originated from soot and gas from five identical fires that were extinguished using four different methods. One extinguishing method were conducted twice and acts as a quality control for the fire test and also for the sampling and extraction. All soot and gas samples were injected three times, in order to study and evaluate errors in the

instrumental analysis, except soot test 5 which was only injected twice.

In the structure elucidation of the non-target data the protocol suggested by Schymanski et al., 2014 was used. A mass accuracy tolerance of 5 ppm was generally used for producing

molecular formulas used for tentative identification. For identification with confidence level 1 standard materials were used.

Analytical standards of several compound classes were run together with the extracts to confirm the presence/absence of certain compounds/compound groups after non-target analysis. This increased the level of certainty of the identification.

(17)

17

Results and discussion

Different methods were used for data mining of the mass spectral data from the gas and soot samples in order to elucidate possible unknown compounds and to assess if there are any differences between the content in gas and soot as well as between the different firefighting methods used. The methods used are described under identification and structure elucidation, and include (i) mass defect plot filtration; (ii) mass spectral interpretation; and (iii) elemental composition calculation. In addition, a possible structure was proposed using different databases. The compound identification section is followed by the statistical analysis.

Data filtration, identification and structure elucidation

The structures presented are based on the molecular formula, isotopic pattern, fragment ions, and double bond equivalent, and should be considered tentative (level 3-4). XCMS used for data preprocessing was not suited for combining the individual m/z features to a peak spectrum and thus it was not possible to use spectral libraries for identification. Instead the exact mass of the molecular ion was used to find the molecular formula using elemental composition calculator. Some structures were found in libraries and in other articles and could be assigned level 2 confidence. Standards are needed to achieve a higher degree of confidence and to confirm the structures.

Alkylated compounds

Hydrocarbons are abundant in many materials and are expected to be detected in the data from all fire samples used in this study. However, the CH2 scale was only applied for soot test

3 since hydrocarbons are expected to provide little information to the present study aims and they do not have a unique isotopic pattern, unlike halogenated compounds. The CH2 scale is

nevertheless interesting in a mass defect filtering aspect, which is why it is included for one sample.

The plot using Kendrick mass scale (CH2) to detect alkylated compounds in soot test 3 can be

seen in Figure 1. Here, the focus was on homologous series showing the CH2 pattern (m/z

±14) in the low region of the mass defect plot. Low intensities caused difficulties in

identifying all ions from the plot in the mass spectra, but the ions with high enough intensity were identified and used to find and fill out the gaps in the homologue series marked in Figure 1A. However, the identified ions are all even electron ions and thus have odd masses, and according to the nitrogen rule, hydrocarbons with an odd mass cannot be a molecular ion unless an odd number of nitrogen atoms are present. The presence of nitrogen is unlikely since the mass defect is low. Since hydrogen is the only atom that contributes to the mass defect in hydrocarbon species the addition of nitrogen would have increased the mass defect. It is reasonable to argue that the ions identified are most likely fragment ions of the molecular ions, and the ionization method was too hard to see any molecular ions.

The homologue series with the lowest degree of unsaturation and thus also the lowest mass defect was CnH2n-2 which is the general formula of alkynes. Most likely the signals

(18)

18 correspond to alkenes with different number of double bonds. The lowest mass defect series (series 1 with a MD≈0.022) has two degrees of unsaturation, which may correspond to two double bonds. The second lowest series (series 2 with a MD≈0.036), has three double bonds, the next one after that four (series 3 with a MD≈0.050), and the last one included in these results has five (series 4 with a MD≈0.062), see Figure 1.

Figure 1. (A) Mass defect plots using the CH2 mass scale showing alkylated hydrocarbons in soot test 3 (extinguishment using foam) between m/z 150-320. The black box in (A) is zoomed in in (B), where the loss/addition pattern is shown together with the tentatively identified molecular formulae, see Table 4.

The tentatively identified molecular formulae of alkylated compounds in soot from test 3 are presented in Table 4. Several homologues with a varying degree of alkylation and saturation could be tentatively identified, however, none containing the molecular ion.

(19)

19

Table 4. Tentatively identified alkylated hydrocarbons (based on the molecular ion, with a mass error tolerance of <5 ppm) detected from mass defect plots based on a CH2 mass scale and identified based on position in the mass defect plot, elemental composition and double bond equivalent.

Homologue

series Molecular formula Theoretical mass

Error from measured mass (ppm) DBE 1 C11H19 151.1487 -0.7 2.5 4 C12H15 159.1174 0.0 5.5 4 C16H23 215.1800 0.0 5.5 2 C14H23 191.1800 -1.6 3.5 3 C14H21 189.1643 -1.1 4.5 2 C12H19 163.1487 0.0 3.5 Halogenated compounds

The brominated compounds found in soot had a varying mass defect and m/z, but generally, the higher the m/z the higher the mass defect. No brominated compounds were found in gas. Brominated compounds can be easily identified in the plot due to their characteristic isotopic patterns (Figure 2). A total of 17 different ions containing bromine could be detected in the mass defect plots. However, only nine seem to be the molecular ion according to the elemental composition. The remaining eight ions were fragment ions formed in the ion source. No previously unknown compound containing bromine could be identified. The technique has however proven efficient in filtering out large sections of the data set to be able to study a specific compound group, thus being unbiased to whether the compound is a target, suspect or unknown compound.

(20)

20

Figure 2. Mass defect plots using the mass scale for brominated compounds showing brominated and chlorinated compounds in soot test 3 (extinguishment using foam) between m/z (A) 150-350; (B) 350-550. The tentatively identified molecular formulae of the circled ions is presented together with possible structures in Table 5.

The plots based on –H/+Br were used to search for both brominated and chlorinated

compounds in all samples. The mass defect plot of soot test 3 based on the -H/+Cl mass scale can be seen in Figure S1 in supplemental information.

Only two compounds containing chlorine could be identified and they were found in soot test 3 and 4, at m/z 283.82157 and 402.87787 (soot test 3). Both compounds contain two bromine atoms and one chlorine, resulting in the isotopic ratio 3:7:5:1. The isotopic pattern matches that of two bromine and one chlorine very well, except for the lack of A+6 ion, giving it the ratio 3:7:5. The lack of A+6 ion is probably explained by the low overall intensity of the molecular ion. The mass defect of the two chlorine containing compounds is the same as that of homologue series 1 and 2 (Figure 2), which are composed of mainly brominated

compounds.

The two different mass scales (HF and CF2) used to identify fluorinated compounds gave no

results, which was expected since no fluorinated firefighting foams were used, and there were no other known sources of fluorinated compounds present in the fire. Mass defect plots using

0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5 150 228 306 M D ( -H/ +B r) m/z

(A)

Series 1 --> 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5 350 428 506 M D ( -H/ +B r) m/z

(B)

Series 3 -->  Series 2

(21)

21 the two scales for detection of fluorinated compounds in gas test 3 can be seen in

supplemental information Figure S2 and S3.

The tentatively identified halogenated compound are presented in Table 5. All of the

identified halogenated compounds are molecular ions, which means they were present in the sample and not formed in the ionization source. Structures for two of the halogenated

compounds are missing, one contains two bromines and one is a mixed brominated and chlorinated compound (mass spectra shown in Figure 3). The suggested molecular formula (with mass error <5 ppm) suggest a similar structure to that of dibromobisphenol.

Tetrabromobisphenol A (TBBPA) was tentatively identified in soot test 3 and 4 using the exact mass and elemental composition, and its presence could be confirmed, using an

analytical standard, in soot 3, 4 and 5. The mass spectra of TBBPA in the analytical standard (A) and in soot test 3 (B) are shown in Figure 3.

Figure 3. The selected ion chromatogram and mass spectra of Tetrabromobisphenol A (m/z 539.7565) in (A) the standard and (B) soot test 3. Present are also a fragment ion (m/z 524.7321). Both the retention time and fragmentation pattern match.

(22)

22 Tetrabromobisphenol A is a common flame retardant and tri- and dibromobisphenol A, as well as dibromophenol, are combustion products of tetrabromobisphenol A (Ortuño et al., 2014). Since tetra- tri- and dibromobisphenol A share the same chemical composition and structure except for a consecutive loss of bromine, they are part of the same homologue series (Figure 2). In the mass defect plot of soot test 3 (Figure 2) that homologue series (series 3) can be detected in low intensity together with a more intense homologue series (series 2) with slightly lower MD and mass. The series with the lower MD could, based on the tentative identification of the tetrabromobisphenol series, be identified as even electron ions (thus not the molecular ions) of tetra-, tri- and dibromobisphenol A, with a loss of CH3. This indicates

that the ions were formed in the ion source and not during the fire. However, one of the ions in that series was found in two peaks in the chromatogram, the same peak as

dibromobisphenol A (suggesting that it is a fragment ion formed in the ion source) and one with slightly shorter retention time than dibromobisphenol A (suggesting that it is not a fragment ion of dibromobisphenol A). Several of the tentatively identified compounds in Table 5 (indicated) have been found in previous combustion or pyrolysis studies (Ortuño et al., 2014; Tostar, 2016).

(23)

23

Table 5. Tentatively identified halogenated compounds (with a mass error tolerance of <6.1 ppm) detected from mass defect plots based on a –H/+Br mass scale and identified based on position in the mass defect plot, isotopic pattern, elemental composition and double bond equivalent.

Name Possible structure Molecular formula Theoretical mass Identified in sample 4,4'-(2,2-Propanediyl)bis(2,6-dibromophenol) (Tetrabromobisphenol A)[1,2] C15H12O2Br4 539.7571 Soot test 3 Soot test 4 2,6-Dibromo-4-[2-(3-bromo-4-hydroxyphenyl)-2-propanyl]phenol (Tribromobisphenol A)[1,2] C15H13O2Br3 461.8466 Soot test 3 Soot test 4 4,4'-(2,2-Propanediyl)bis(2-bromophenol) (Dibromobisphenol A)[1,2] C15H14O2Br2 383.9361 Soot test 3 Soot test 4

Unknown even electron ion - C14H11O2Br2 368.9126

Soot test 3 Soot test 5 Unknown even electron ion - C14H10O2ClBr2 402.8736

Soot test 3 Soot test 4 2,4,6-Tribromophenol C6H3OBr3 327.7734 Soot test 1 Soot test 3 Soot test 4 Soot test 5 2,6-Dibromophenol C6H4OBr2 249.8629 Soot test 1 Soot test 3 Soot test 4 Soot test 5 2,6-Dibromo-4-chlorophenol C6H3OClBr2 283.8239 Soot test 3 Soot test 4 2,6-Dibromo-4-isopropenylphenol [1] C9H8OBr2 289.8942 Soot test 3 Soot test 4 [1] Ortuño et al., 2014 [2] Tostar 2016

(24)

24

PAHs

The PAHs, which are found in the top part of the C4H2 mass defect plot, show no isotopic

pattern (Figure 4). The m/z ±50 pattern, where m/z ±50 resemble an aromatic ring being added/removed from another aromatic ring, is used to find homologues. The m/z with the highest mass defect is usually the molecular ion, and the signals with lower mass defect represents the consecutive loss of H. This loss of a hydrogen from the molecular ion could have happened both in the fire situation and in the electron impact ion source. Comparing retention times between signals can be done to learn if the ions were formed during the fire (different retention times) or formed in the EI source (same retention times).

Filtration using mass defect plots was highly helpful for PAHs as well, resulting in homologue series of PAHs that could be tentatively identified with good mass accuracy (below 5 ppm).

Figure 4. Mass defect plots using the C4H2 mass scale in soot test 1 (extinguishment using water spray) between m/z 150-300. The tentatively identified molecular formulae of the detected ions are presented together with possible structures in Table 6-10.

The mass defect plot of the standard containing halogenated PAHs (19 chlorinated and 19 brominated) that was run to confirm the absence of halogenated PAHs in soot and gas is shown in Figure 5. The plot was based on the –H/+Br mass scale and as expected,

halogenated PAHs display the same isotopic patterns as other halogenated compounds, but they are found higher in the plot than the brominated and chlorinated compounds seen in Figure 2 since they have a higher MD.

0,92 0,93 0,94 0,95 0,96 0,97 0,98 0,99 1 150 200 250 300 M D ( C4 H2 ) m/z Series 1 Series 2 Series 3 Series 4 Oxy-PAHs

(25)

25

Figure 5. Mass defect plots using the –H/+Br mass scale in the halogenated PAH standard between m/z 180-380. Many of the compounds in the standard have identical m/z and MD since they are structural isomers, which makes the plot appear to contain fewer compound than it does since the data points overlap. No Halogenated PAHs could be identified in the gas and soot samples. Due to problems with the exportation of the raw data file the intensity could not be included in this plot.

The PAHs from the mass defect plot (Figure 4) that could be tentatively identified are shown in Table 6-10. The table include molecular ions but possibly also fragment ions formed in the ion source. The chromatograms and mass spectra of the interesting m/z were hard to interpret due to the presence of isomers. In order to distinguish a molecular ion from a fragment ion, the retention times were compared. If two m/z fragments have exactly the same RT pattern they are likely to belong to the same compound. Also helpful is if the molecular ion can be seen in the mass spectra. For the soot and gas samples it was difficult to obtain the molecular ion using a low mass tolerance due to low signal strength of the molecular ions

Several PAHs identified from the mass defect plot were also detected in previous target analysis (unpublished data) and are indicated in Table 6-9. Also indicated in the table is references to pyrolysis studies where the PAHs were detected. The mass scale of C2H4 proved

to work well for filtering out PAHs from the non-target data set.

The proposed structures that have no support in literature are suggestions that need to be confirmed using analytical standards. Still, they cannot be classified as unknown compounds as many of the structures that are proposed in Table 6-10 are likely structures found in common databases. On the other hand, a number of tentatively identified compounds lack support in the literature, indicating that target methods are not inclusive enough and it is fair to suspect that unknown yet possibly interesting compounds are omitted in target analysis. Homologue series 1 (seen in Figure 4), is presented in Table 6. Each m/z ±50 step (indicated with a color change in the table) in the series includes multiple compounds with varying degrees of unsaturation.

(26)

26

Table 6. Tentatively identified PAHs in homologue series 1 (Figure 4) (with a mass error tolerance of <5 ppm) detected from mass defect plots based on a C4H2 mass scale and identified based on position in the mass defect plot, elemental composition and double bond equivalent. The footnotes indicate previous studies the compounds have been identified in.

Name Suggested structure(s) Molecular

formula DBE Theoretical mass Identified in sample Acenaphthylene[1,2] C12H8 9.0 152.0626 Soot test 1-5 Gas test 1-5 Acenaphtene[1,2] C12H10 8.0 154.0774 Soot test 1 Soot test 5 Gas test 1 Gas test 3 Fluoranthene[1,2] Or Pyrene[1,2] C16H10 12.0 202.0783 Soot test 1 Soot test 3-5 Gas test 1 1-Phenylnaphthalene 5,10-Dihydroindeno[2,1-a]indene 2,3-Dihydrofluoranthene C16H12 11.0 204.0938 Soot test 3 Benzo(b)fluoranthene,[1,2] Benzo[j]fluoranthene[1] Benzo(k)flouranthene[1,2] Benzo[e]pyrene[1] Benzo(a)pyrene[1,2] Perylene[1] C20H12 15.0 252.0939 Soot test 1 Soot test 3-5 Gas test 1 Gas test 3 1,2-Dihydrocyclopenta[ij]tetraphene C20H14 14.0 254.1096 Soot test 1 Soot test 3-4 Gas test 1 7,12-Dimethyltetraphene[1]

6-Ethylchrysene[1] C20H16 13.0 256.1252 Soot test 1

3,3',4,4'-Tetrahydro-1,1'-binaphthalene C20H18 12.0 258.1409 Gas test 1

[1] Identified in target analysis of the gas and soot (unpublished) [2] Detected in soot by Sánchez et al., 2012

(27)

27 Homologue series 2 (seen in Figure 4), is presented in Table 7. Each m/z ±50 step (indicated with a color change in the table) in the series includes multiple compounds with varying degrees of unsaturation.

Name Suggested structure(s) Molecular

formula DBE Theoretical mass Identified in sample Fluorene[1,2] C13H10 9.0 166.0783 Soot test 1-2 Soot test 5 Gas test 1-5 Benz[e]indan C13H12 8.0 168.0939 Soot test 5 Gas test 3-4 11H-Benzo[a]fluorene[1] 1-Methylfluoranthene[1] C17H12 12.0 216.0939 Soot test 1 Soot test 4 1,2-Cyclopentenophenanthrene C17H14 11.0 218.1096 Soot test 1 Soot test 4 7-Methylbenzo[pqr]tetraphene[1] or

3-Methylcyclopenta[ij]tetraphene C21H14 15.0 266.1096 Soot test 1

3-Methyl-1,2-dihydrocyclopenta[ij]tetraphene or 2,3-Dihydro-1H-cyclopenta[k]tetraphene C21H16 14.0 268.1252 Soot test 1 7,8,12-Trimethyltetraphene C21H18 13.0 270.1409 Gas test 1

7-tert-Butyl-1-methylpyrene C21H20 12.0 272.1565 Gas test 1

(28)

Name Suggested structure(s) Molecular formula Theoretical mass DBE Identified in sample Cyclopenta[fg]acenaphthylene C14H8 176.0626 11.0 Soot test 1-2 Soot test 4-5 Gas test 2-3 Gas test 5 Phenanthrene[1,2] or Anthracene[1,2] C14H10 178.0783 10.0 Soot test 1-5 Gas 1-5 9,10-Dihydroanthracene or 9,10-Dihydrophenanthrene C14H12 180.0939 9.0 Soot test 1-2 Soot test 4-5 Gas test 1-5 1,2,3,4-Tetrahydrophenanthrene C14H14 182.1096 8.0 Soot test 1-2 Soot test 4-5 Gas test 2-5 Benzo[ghi]fluoranthene or Cyclopenta[cd]pyrene C18H10 226.0783 14.0 Soot test 1 Soot test 3-5 Gas test 1 Benz(a)anthracene[1,2], Chrysene[1,2], Triphenylene[1] or Benzo[c]phenanthrene[1] C18H12 228.0939 13.0 Soot test 1 Soot test 3-5 Gas test 1 5,12-Dihydrotetracene C18H14 230.1096 12.0 Soot test 1 Soot test 3 Indeno(1,2,3-cd)pyrene[1,2] or Benzo(g,h,i)perylene[1,2] C22H12 276.0939 17.0 Soot test 1 Soot test 3-5 Dibenzo(a,h)anthracene[1,2] C22H14 278.1096 16.0 Soot test 1 Soot test 3-5

(29)

Table 9. Tentatively identified PAHs in homologue series 4 (Figure 4) (with a mass error tolerance of <5 ppm) detected from mass defect plots based on a C4H2 mass scale and identified based on position in the mass defect plot, elemental composition and double bond equivalent. The footnote indicate a previous study the compounds have been identified in.

Name Suggested structure(s) Molecular formula Theoretical mass DBE Identified in sample 4H-Cyclopenta[def]phenanthrene[1] or 6H-Cyclopenta[d]acenaphthylene C15H10 190.0783 11.0 Soot test 1 Soot test 3-5 2-Methylphenanthrene[1] or 2-Methylanthracene[1] C15H12 192.0939 10.0 Soot test 1 Soot test 3-5 Gas test 1 4H-Cyclopenta[def]chrysene or 3H-Benzo[cd]pyrene C19H12 240.0939 14.0 Soot test 1 Soot test 4 Gas test 1 7-Methyltetraphene[1] or 3-Methylchrysene[1] C19H14 242.1096 13.0 Soot test 1 Soot test 4 4-(Propan-2-yl)pyrene or 12,17-Dimethylgona-1(10),2,4,6,8,11,13,16-octaene C19H16 244.1252 12.0 Soot test 1

(30)

30 The tentatively identified oxygenated PAHs, also called Oxy-PAHs (seen in Figure 4), are presented in Table 10.

Table 10. Tentatively identified oxygenated PAHs (Figure 4) (with a mass error tolerance of <5 ppm) detected from mass defect plots based on a C4H2 mass scale and identified based on position in the mass defect plot, elemental composition and double bond equivalent.

Name Suggested structure(s) Molecular formula Theoretical mass DBE Identified in sample 9H-Xanthene or Benzophenone C13H10O 182.0732 9.0 Soot test 1 Gas test 1 Gas test 3 1,2,3,4-Tetrahydro-9H-fluoren-9-one C13H12O 184.0888 8.0 Soot test 1 Soot test 5 Gas test 3 9(10H)-Anthracenone C14H10O 194.0732 10.0 Soot test 1-2 Soot test 5 Gas test 1-2 Gas test 4-5 (4-Methylphenyl)(phenyl)methanone C14H12O 196.0888 9.0 Soot test 1-2 Soot test 5 Gas test 1-5

Di-2-naphthylmethanone C21H14O 282.1045 15.0 Gas test 1

2-Methoxy-1,1'-binaphthalene C21H16O 284.1201 14.0 Gas test 1

(31)

31

Comparison of gas and soot

Several differences between gas and soot can be observed from the mass defect plots and lists of identified compounds (Table 4-10). Generally, the data of the soot samples contained a larger number of m/z than the gas samples. This could reflect the formation of more compounds that sorb to particles than compounds that are dissolved in the gas phase, or it could reflect a bias in the sampling procedure. Brominated compounds could be identified in all soot samples except soot test 2, while there are no brominated compounds in any of the gas samples. The lack of brominated compounds in gas is most likely due to low volatility of such compounds.

Chlorinated compounds were only found in soot 3 and 4. If PVC plastics were present during the fire, they may be the source of chlorine. However the fire tests were done with furniture and electronics and there were probably very little chlorinated materials. The chlorinated compounds identified seem to be substituted products of Tetrabromobisphenol A (TBBPA) formed in the fire and since TBBPA was only identified in soot 3 and 4, that may explain why chlorinated compounds were only found in those samples.

PAHs are semi-volatile and are thus expected to be present in gas phase, but they can also be found as particulate matter, as demonstrated in Table 6-10. The main difference between gas and soot when it comes to the PAHs is that gas contains few of the heavier PAHs that have m/z >200 (fire rings or more), while soot contains almost the whole range of masses from m/z 150-300. This is also expected since the lighter PAHs have a lower boiling point and higher vapor pressure, making them more volatile.

The results of a PCA of all soot and gas samples are seen in Figure 6 (scores), Figure 7 (loadings) and Figure S4 in supplemental information (scree). Most of the variance and thus separation is explained by PC1. Soot test 3 and 4 seen in the left part of the plot are separated from the remaining samples, indicating that they are correlated to each other but not the remaining samples. The gas samples are poorly separated in PC1, signifying that they are similar. On PC2, the gas samples are clustered into two subgroups; gas test 1 and 3 and gas test 2, 4 and 5. This is interesting since test 2, 4 and 5 were all using the same firefighting technique: coldcut cobra cutting extinguisher (Table 2). Test 2 is without additive and test 4 and 5 are with additive. In this plot there is no clear separation between with and without additive. Also present in the test 2, 4 and 5 gas cluster is soot test 2 which means that soot and gas for test 2 has similar composition. Also, based on the results from the tentative

identification of halogenated compounds in the samples, it was possible to conclude an absence of any halogenated compound in soot test 2. The same was true for all gas samples. This implies that halogenated compounds have a great impact in the PCA model.

(32)

32

Figure 6. Scores plot of all soot and gas samples displaying clustering of the samples. The position of each sample in the scores plot is decided by the loadings in the loading chart.

The loading chart display the measured m/z and show which of the m/z that influence the scores of the samples. The distance and direction from the origin decides how much the m/z influence the sample score in the scores plot and in what way. The outliers found in a specific area of the loading chart have strong correlations to the samples found in the same area in the scores plot. The outlier that contribute the most to the clustering of the gas samples and soot test 2 in the top right corner is m/z 195.1 T20.9, which has not been detected as an interesting compound by the mass defect filtration. Below it in the bottom right corner, which is

associated with gas test 1 and 3, is m/z 181.1 T15.6 which likely belongs to

Trihydrophenanthrene as a fragment ion. The compound that is the most accountable for the position of soot test 3 and 4 is Fluoranthene and/or Pyrene (202.1 T14.5).

(33)

33

Figure 7. Loading chart of all soot and gas samples. The distance and direction from the origin decides how much the m/z influence the sample score in the scores plot and in what way. The outliers found in a specific area of the loading chart have strong correlations to the samples found in the same area in the scores plot.

Compound formation depending on firefighting method

Once again, by visual inspection of the mass defect plots and lists of identified compounds (Table 4-10) one can argue that the chemical composition of the samples seems to differ between the tests. As mentioned, there were no halogenated compounds in any of the gas samples or in soot test 2. The rest of the samples contained a varying composition of

brominated compounds, where soot test 3 and 4 contained the largest number of halogenated compounds overall. Also, test 3 and 4 were the only ones containing chlorinated compounds. The number of identified PAHs varied between the samples. Test 1 (soot and gas) contained the largest number of PAHs, followed by test 4 (soot and gas) and test 5 (soot and gas) Figure 8). The lowest number of PAHs was found in test 2 (soot and gas) and 3 (soot and gas).

(34)

34

Figure 8. Bar plot showing the contribution of halogenated compounds, PAHs and oxy-PAHs to the overall composition of each sample. Observe that it is not a measure of concentration or amounts in each sample, only the number of compounds.

In the scores plot (Figure 9) of all soot samples the instrumental injections of each sample are grouped together and all samples are clearly separated from each other. However, despite all samples being separated on PC1, soot test 3 and 4 are very similar on PC2. Soot test 4 and 5, which are replicates of the same firefighting method, are expected to be grouped together. Although they are fairly close on PC2, they are separated on PC1. This non-grouping indicates that there are differences in their compound formation. The cause may lie in the external conditions, which will be discussed later.

Figure 9. Scores plot of all soot samples displaying clustering of the samples. The position of each sample in the scores plot is determined by the loadings in the loading chart.

The loading chart of all soot samples can be seen in Figure 10. Two important markers for PC1, that were shown to separate all soot samples (Figure 9) were tentatively identified as (in the top right part of the plot) Tribromophenol (m/z 329.8 T10.0 & 331.8 T10.0 which are the

0 5 10 15 20 25 30 35 N o . o f id en tifi ed comp o u n d s Sample Sum hologenated Sum Oxy-PAHs Sum PAHs

(35)

35 two most intense ions of the compound) & Dibromophenol (251.9 T5.6 which is the most intense ion of the compound), as well as (in the bottom right part of the plot) Fluoranthene and/or Pyrene (202.1 T14.5).

Figure 10. Loading chart of all soot samples. The distance and direction from the origin decides how much the m/z influence the sample score in the scores plot and in what way. The outliers found in a specific area of the loading chart have strong correlations to the samples found in the same area of the scores plot.

The repeated injections of each sample are grouped together in the scores plot (Figure 11) for the gas samples as well, except for the gas test 4 injections. The gas test 4 injections are fairly close in PC1, but separated in PC2.

Figure 11. Scores plot of all gas samples displaying clustering of the samples. The position of each sample in the scores plot is decided by the loadings in the loading chart.

(36)

36 There are few outliers in the loading chart (Figure 12) that may be accountable for the split of the two gas test 4 injections, and it may also not be dissimilarities between the two injections that cause the variability in PC2, but rather the similarities between all samples. The

intensities are low across all gas samples, making the slightest deviation stand out clearer for the gas samples than for the soot samples. Thus, the loading chart of all gas samples is not analyzed further.

Figure 12. Loading chart of all gas samples. The distance and direction from the origin decides how much the m/z influence the sample score in the scores plot and in what way. The outliers found in a specific area of the loading chart have strong correlations to the samples found in the same area in the scores plot.

(37)

37 It could have been favorable to replace PC2 by PC3 or PC4 to get a different clustering. In this case however the overall clustering did not improve by doing so (Figure 13-14).

Figure 13. Scores plot of all gas samples displaying PC1 plotted against PC3. The clustering of the instrumental injections did not improve compared to the clustering seen using PC1 and PC2 (Figure 11).

Figure 14. Scores plot of all gas samples displaying PC1 plotted against PC4. The clustering of the instrumental injections did not improve compared to the clustering seen using PC1 and PC2 (Figure 10).

(38)

38 Due to the behavior of the gas test 4 injections, which are supposed to be identical, the results from the gas samples are ambiguous, and they were therefore not used for drawing any conclusions regarding the difference in compound formation between the different firefighting methods.

The PCA of all soot samples revealed unexpected clustering results since the two test replicates (test 4 and 5) were not grouped together. To be able to compare the different firefighting methods, more information is needed about the replicates. A pairwise analysis, comparing those two tests, should have revealed that they are similar and the variance low. However, they were separated on PC1 (Figure 15). PC2 on the other hand was not able to separate them equally well.

Figure 15. Scores plot of pairwise comparison of soot test 4 and 5 displaying clustering of the injection replicates and separation of the two samples. The position of each sample in the scores plot is decided by the loadings in the loading chart.

The fact that the two replicate tests are separated indicate that there are differences in the compound formation not explained by the firefighting method used. By using the knowledge about the samples and sampling it is possible to present a possible explanation for this difference. This difference may be explained by the varying weather conditions, specifically wind speed and direction, which changed between the two test days. It has been discussed in Bjurlid et al. 2017 that the weather conditions affected the rate of temperature increase after ignition, which in turn affected compound formation, see Figure 16.

(39)

39

Figure 16. Temperature curves of all tests borrowed with permission from Bjurlid et al. 2017 showing temperature change from ignition. The test number and wind conditions are specified by each curve.

Since the two fire events took place during different weather conditions, it is plausible to assume that the study is sensitive to fluctuations in the external conditions. The outliers (the compounds that differ between the different tests) are therefore more likely compounds that are formed to a different extent depending on heat development and thus wind conditions rather than the firefighting method used. In Figure 17, the temperature curves of the

extinguishing phase show correlations between test 4 and 5 (the two replicates), which is not reflected in the results from clustering in the PCA of all soot samples. This is another

indication that the firefighting method that was used is not the sole factor in compound formation.

Application of mass defect filtering and statistical analysis for non-target data mining of gas and soot data from a study testing different firefighting methods