Removal of volatile gasoline compounds by indoor potted plants studied by pixel-based fingerprinting analysis

Chemosphere 221 (2019) 226e234

Contents lists available at ScienceDirect

Chemosphere journal homepage: www.elsevier.com/locate/chemosphere

Removal of volatile gasoline compounds by indoor potted plants studied by pixel-based fingerprinting analysis Majbrit Dela Cruz a, b, *, Giorgio Tomasi a, Renate Müller b, Jan H. Christensen a a b

Department of Plant and Environmental Sciences, Faculty of Science, University of Copenhagen, Thorvaldsensvej 40, 1871, Frederiksberg C., Denmark Department of Plant and Environmental Sciences, Faculty of Science, University of Copenhagen, Højbakkegaard All
e 30, 2630, Taastrup, Denmark

h i g h l i g h t s
Potted Hedera helix was exposed to gasoline under dynamic conditions.
Data was analyzed by a non-targeted pixel-based fingerprinting approach.
All gasoline VOCs were reduced in concentration by H. helix.
Preferential removal of some compounds was observed.
The soil microcosm play a substantial role in removal of gasoline VOCs.

a r t i c l e i n f o

a b s t r a c t

Article history: Received 9 August 2018 Received in revised form 15 December 2018 Accepted 18 December 2018 Available online 22 December 2018

Indoor potted plants are able to remove volatile organic compounds (VOC) from air, but only few studies have investigated the removal of compounds in mixtures. Here, we present a non-targeted pixel-based fingerprinting analysis documenting the removal of a complex mixture of gasoline VOCs by Hedera helix under dynamic chamber conditions allowing for air exchange and continuous gasoline exposure. For 15 days, the entire potted plant was exposed to gasoline; subsequently, the epigeous plant parts were removed and the soil microcosm (i.e. soil, plant roots and microorganisms) was exposed to gasoline for another eight days. Quantitative analysis was performed for heptane, 3-methylhexane, toluene, ethylbenzene and m,p-xylenes, and the CHEMSIC method (CHEMometric analysis of Selected Ion Chromatograms) was used for non-targeted pixel-based fingerprinting analysis. The quantitative analysis demonstrated that the presence of potted plants or pots without epigeous plant parts led to a reduction of selected VOCs by 16.7e22.6%. The CHEMSIC method confirmed this and revealed that all gasoline VOCs were reduced in concentration when H. helix was present. The estimate for the total VOC removal was in the range of 11e32%. The removal was highest for samples where the epigeous plant parts were absent and compounds known to be hard to degrade by microorganisms such as dimethylcyclopentanes were removed the least compared to compounds more easily degraded by microorganisms such as heptane when epigeous plant parts were removed. All findings support the conclusion that the soil microcosm was the main responsible for the removal of VOCs. © 2019 Elsevier Ltd. All rights reserved.

Handling Editor: R Ebinghaus Keywords: VOC removal Indoor air quality Hedera helix Chemometrics Hydrocarbon

1. Introduction Volatile organic compounds (VOCs) are everywhere in the indoor environment and several hundred VOCs can co-exist (Edwards et al., 2001). Some VOCs can affect human health negatively, e.g.

* Corresponding author. Department of Plant and Environmental Sciences, Faculty of Science, University of Copenhagen, Thorvaldsensvej 40, 1871, Frederiksberg C., Denmark. E-mail address: [email protected] (M. Dela Cruz). https://doi.org/10.1016/j.chemosphere.2018.12.125 0045-6535/© 2019 Elsevier Ltd. All rights reserved.

formaldehyde and benzene are carcinogenic (World Health Organization, 2010), and recently a study showed that indoor residential concentrations of 18 out of 59 VOCs exceeded guideline values (Logue et al., 2011). In addition, VOCs are often recognized as odors before reaching harmful levels for humans and can in that sense contribute to the perception of poor indoor air quality (Wolkoff et al., 2006). Indoor potted plants can remove VOCs, but often only single compounds or simple mixtures of few compounds have been studied in experimental test systems (Dela Cruz et al., 2014a). The removal of specific VOCs by plants may depend on

M. Dela Cruz et al. / Chemosphere 221 (2019) 226e234

other compounds that are present and interactions are known to exist (Cornejo et al., 1999; Orwell et al., 2006; Porter, 1994; Yoo et al., 2006). These interactions can generate complex removal patterns. For example, when Dracaena deremensis ‘Janet Craig’ plants were exposed to both m-xylene and toluene, m-xylene removal increased significantly (from 0.08 mg∙day1 to 0.24 mg∙day1 and from 0.57 mg∙day1 to 2.36 mg∙day1 at dosages of 0.2 and 1.0 ppm, respectively) compared to when the plants were exposed to m-xylene alone, but interestingly the effect disappeared at higher dosages (10 and 100 ppm) and toluene uptake was unaffected by the presence of m-xylene (Orwell et al., 2006). Similarly, simultaneous exposure to both benzene and toluene at 1 mL L1 caused a decrease in removal compared to exposure to the individual compounds for Cissus rhombifolia, Hedera helix, Spathiphyllum wallisii, and Syngonium podophyllum (Yoo et al., 2006). Only one study has investigated simultaneous uptake of a complex mixture comprising more than five compounds and while it showed that Epipremnum aureum was capable of removing gasoline compounds, it was not specified which compounds in the mixture were removed (Oyabu et al., 2003). With hundreds of compounds co-existing in the indoor environment (Edwards et al., 2001), it is relevant to investigate the effect of potted plants on air contaminated with a complex mixture of VOCs. An example of such a complex mixture is gasoline. Gasoline components such as benzene, toluene, ethylbenzene and xylenes (BTEX) can be found in homes with gasoline-powered equipment or gasoline containers stored in basements, garages, and carports (Batterman et al., 2007; Du et al., 2015; Hun et al., 2011). Exposure to gasoline can lead to genotoxic effects and the main constituent thought to cause these alterations is benzene (Araújo et al., 2010; Benites et al., 2006), which, at higher exposure levels, has also been correlated to an increased reporting of severe asthma (Gordian et al., 2010). The aim of the present study was to investigate the removal of a complex mixture of gasoline VOCs by H. helix, a plant species chosen due to its wide availability in Denmark. This was carried out by investigating both changes in concentration and differential removal of the VOCs. The air samples were analyzed by gas chromatographyemass spectrometry (GCeMS) and the change in the VOC composition (i.e. the chemical fingerprint) by potted H. helix was compared to unaffected, gasoline polluted air. To this end, the CHEMSIC method (CHEMometric analysis of sections of Selected Ion Chromatograms) was used: In this approach, selected ion chromatograms (SICs) are analyzed without the extraction of relative peak areas or concentrations (Christensen et al., 2010; Christensen and Tomasi, 2007; Gallotta and Christensen, 2012). The method comprises a number of preprocessing steps (viz. baseline removal, retention time alignment and normalization) to exclude variation that is unrelated to the chemical composition before the data is analyzed by principal component analysis (PCA) (Jolliffe and Cadima, 2016). One of the main advantages of this approach is that little information is excluded a priori, which allows for a more unbiased and explorative analysis of the data (Christensen et al., 2005). 2. Materials and methods

(Multigreen.dk A/S, Odense, Denmark). The plants were six weeks old, grown in 11 cm pots and consisted of 6e7 plantlets. A plant was placed in two of four glass chambers (plant chambers) (57.5 L, Cichlide Centret Aps, Vallensbæk Strand, Denmark) set up in a climate chamber. The two chambers without plants were used as control chambers to correct for chamber effects and were empty apart from a temperature measuring device (Fig. S1). Light intensity was set to 37 ± 3 mmol m2∙s1 and the day length was 12 h. Temperature was controlled at 20.8 ± 0.5  C. Prior to the experiment, the plants were let to acclimate for 14 days. The samples were then collected over a period of 23 days: after the first 15 days the epigeous plant material was abscised and for the last eight days only the pot with soil and the hypogeal part remained. The leaf area was measured with a LI-3100 area meter (LICOR. Inc., Lincoln, NE, USA) at the time when the epigeous parts were removed. The method of pollution exposure has previously been described (Dela Cruz et al., 2014b). Briefly, air cleaned by a compressed air filter and supplied from a central compressor was directed through a mixing chamber. The mixing chamber was supplied with a 200 mL bottle with gasoline, which could diffuse through a 1.7 cm 12 gauge needle in the lid of the vial. This ensured a continuous flow of gasoline polluted air for the glass chambers. The air flow was regulated at 4.3 ± 0.1 L min1 by a pressure regulator (Gloor 5650, Gloor Bros Ltd., Burgdorf, Switzerland) before entering the mixing chamber. Air entered the glass chambers through an air inlet placed at the top of the removable side glass plate of the glass chambers while the air outlet was placed at the bottom of the plate. The lid was further equipped with an inlet for watering with 15 cm Teflon tubing extending inwards. The lid was held in place with clamps and sealed with rubber foam and gaffer tape. The air leaving the glass chambers was sampled on activated charcoal tubes, that consist of a sampling part and a backup part (ORBO™ 32, (20/40), 100/50 mg, Supelco, Bellafonte, PA, USA), connected to AirChek2000 pumps (SKC Inc., Eighty Four, PA, USA) (Dela Cruz et al., 2014b). After sampling, the activated charcoal tubes were stored at 18  C. 2.3. Sample sets The experiment was repeated twice; each time, a total 56 samples were collected over 23 days: 36 samples for the period when the whole plants were present (epigeous period) and 20 samples for the period when the epigeous plant parts were absent (hypogeous period). Plant and control chambers were sampled equally often. In addition, 12 analytical duplicates were included (Table 1). Sample set labels begin with a P or C to denote plant or control chambers, respectively, followed by a Y or N to denote epigeous or hypogeous period, respectively. Sample set labels for the duplicate samples are prepended with Dup. To avoid ambiguity, sample set labels may further be prepended with Exp1 or Exp2 to denote Exp. 1 or Exp. 2, respectively.

Table 1 Overview of the sample sets. Sample set

2.1. Chemicals The description of chemicals is provided in Supporting Information. 2.2. Experimental set-up Hedera helix ‘Shamrock’ was obtained from a local nursery

227

PY CY PN CN DupPY DupCY DupPN DupCN

Experiment 1

2

18 18 10 10 2 3 1 1

18 18 10 10 2 0 0 3

Chambers

Period

Plant Control Plant Control Plant Control Plant Control

Epigeous Epigeous Hypogeous Hypogeous Epigeous Epigeous Hypogeous Hypogeous

228

M. Dela Cruz et al. / Chemosphere 221 (2019) 226e234

2.4. Chemical analysis The sampling part of the activated charcoal tubes was transferred to 4 mL vials and 2 mL CS2 and 100 mL internal standard mixture was added. For selected samples, the backup fraction was transferred to 4 mL vials and 1 mL CS2 and 50 mL internal standard mixture were added. The samples were ultrasonicated for 30 min and an aliquot of the supernatant was transferred to GC vials. The samples were kept at 20  C until analysis. The sample extracts were analyzed on a GC (7890A, Agilent Technologies, Santa Clara, CA, USA) with an MS detector (5975C, Agilent Technologies, Santa Clara, CA, USA). Selected ion monitoring (SIM) was used to analyze 21 mass-to-charge ratios (m/z) (see Supporting Information). The m/z were selected to cover most compounds in the gasoline mixture to which the plants were exposed (Harley and Kean, 2004). The focus was on compounds with high vapor pressure as the process of exposure was through diffusion. Selected samples as well as a sample of 0.5% gasoline in CS2 were further analyzed on a GC (7890B, Agilent Technologies, Santa Clara, CA, USA) with quadrupole time-of-flight MS (7200, Agilent Technologies, Santa Clara, CA, USA) with similar settings for the GC as above. These analyses were used for tentative identification of compounds using NIST14 MS library search. The NIST hits were evaluated based on the reverse match factor, the match factor, the retention index (RI), and the probability in that order as well as visual inspection of the spectra. The RI of the calibration standards were used as reference points. A total of 210 GC-MS runs was performed which included sample extracts, analytical duplicates, calibration standards, quality control (QC) samples, a 1:1 and a 1:5 dilution of the QC mixture in CS2 (QCD1 and QCD2, respectively), and analytical blanks. Further description of the chromatographic conditions and runs is provided in the Supporting Information.

2.5. Data processing and analysis 2.5.1. Quantitative analysis The VOCs were quantified using a linear five-point internal calibration curve: Heptane and 3-methylhexane were quantified using d16-heptane as internal standard, whereas toluene, ethylbenzene and m,p-xylenes were quantified using d8-toluene, d10ethylbenzene and d10-p-xylene, respectively as internal standards. As the experiments were time series, the collected data points were not independent. Therefore, the VOC concentration differences between control chambers and chambers containing plants were analyzed in R using the lme4 package (version 1.1e10) and a mixed linear model (Bates et al., 2015) with treatment (presence of plants) and condition (presence of epigeous plant parts) as fixed effects. Chamber and sampling day were included as crossed random

effects with experiment nested in sampling day. The delta method (car package, version 2.0e25) was used to estimate removal efficiencies and corresponding standard errors. To estimate the total VOC removal, the area for the total ion chromatogram (TIC) from 9.77 to 24.65 min (corresponding to the chosen SICs specified in section 2.5.2) for each sample was normalized to the area of the nearest QC sample and the percentage removed by H. helix with or without epigeous plant parts was calculated for each sampling day.

2.5.2. Pixel-based analysis For the CHEMSIC method, 17 regions of interest (RoIs) were extracted comprising 15 m/z (Table S2). The SICs for m/z 57 and 91were split in two RoIs to improve the retention time alignment: the Correlation Optimised Warping algorithm (cf. Supporting Information) is biased towards correctly aligning the tallest peaks in a chromatogram (Nielsen et al., 1998); since in these two SICs the peaks in the first part were significantly larger than in the second half, the split effectively removes the bias. No a priori knowledge on gasoline removal by potted plants was applied and the RoIs were defined based on visual inspection of where peaks were present. Compounds eluting before 1,1-dimethylcyclopentane (9.84 min) were found in the backup fraction of the activated charcoal tubes. Therefore, to avoid introducing bias, all the sections eluting before 9.77 min (i.e., the start of the 1,1-dimethylcyclopentane peak) were excluded. The GC-MS data were exported in the netCDF format using the commercial software ChemStation (Agilent technologies, Santa Clara, CA, USA). The sample variation unrelated to the chemical composition of the samples needs to be minimized prior to PCA to facilitate the interpretation of the model parameters. Thus, preprocessing of the chromatographic data included, in order: baseline removal, retention time alignment, and normalization (Christensen et al., 2010; Nielsen et al., 1998; Skov et al., 2006; Tomasi et al., 2004). The SICs were either normalized to internal standard and sample volume or unitary Euclidean norm. The first normalization scheme will focus the analysis on concentration differences between plant and control chambers while with the second scheme focus will be on relative differences in the chemical composition of samples. See supporting Information for further description of the pixel-based analysis. For PCA two options were investigated: a) the PCA matrix was comprised of all the RoIs and b) only selected RoIs were combined for the final PCA models (Table 2). Each model is identified with a label that starts with ‘a’ when all RoIs are included and ‘s’ (selected) otherwise; ‘Y’ and ‘N’ refer to the epigeous or the hypogeous period, respectively (both control and plant chambers), and ‘P’ that only plant chambers are considered (both epigeous and hypogeous period). The model label ends with a number which denotes the

Table 2 Overview of PCA models. Model

aY1 aY2 aN1 aN2 sY1 sY2 sN1 sN2 sP1 sP2

Chambers

Period

Plant

Control

Epigeous

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓ ✓

Training set Hypogeous

✓

✓ ✓

✓ ✓ ✓ ✓

✓ ✓ ✓ ✓

Normalization

RoIs included

IS, volume sampled IS, volume sampled IS, volume sampled IS, volume sampled Euclidean norm Euclidean norm Euclidean norm Euclidean norm Euclidean norm Euclidean norm

all all all all 1m41, 1m41, 1m41, 1m41, 1m41, 1m41,

Exp. 2

✓ ✓ ✓

✓ ✓

Exp. 1

✓ ✓ ✓ ✓

1m55, 1m55, 1m55, 1m55, 1m55, 1m55,

1m69 1m69 1m69 1m69 1m69 1m69

M. Dela Cruz et al. / Chemosphere 221 (2019) 226e234

experiment used as training set for the model. The roles of Exp. 1 and Exp. 2 were different: Exp. 2 was run at a later stage to validate the findings in Exp. 1. Therefore, its samples were used thusly in the data analysis: they were projected on the model calculated on Exp. 1 as well as modelled separately, in order to compare the model coefficients from the two independent experiments.

229

xylenes significantly lower in Exp. 1 than in Exp. 2 (Table S3). The leaf area of the plants in Exp. 1 was 704 ± 1 cm2 whereas it was 437 ± 16 cm2 in Exp. 2, but overall the differences in concentrations between experiments were low (0.8e11.5%, Table S3) and the two experiments were therefore combined for calculation of the percentage removed of the selected VOCs. The percentage removed calculated on the basis of the TIC was in the range of 14e32% for the epigeous period and 11e25% for the hypogeous period.

3. Results 3.2. Pixel-based analysis 3.1. Quantitative analysis The presence of potted plants or pots without epigeous plant parts significantly (p < 0.001) reduced the concentration of the selected VOCs by 16.7e22.6% (Table 3). All removal percentages were higher for the hypogeous period compared to the epigeous period (p < 0.01). Thus, the removal efficiency increased after the epigeous plant parts were removed. The plants showed an adaptation period in the beginning of the exposure, but the removal percentages were stabilized from day five (Fig. S2). The exposure concentrations of heptane, 3-methylhexane and toluene measured in the control chambers were significantly higher and that of m,p-

In model aY1, plant and control samples for the epigeous period were included and the normalization made the PCA focus on concentration effects. PC1 described 96% of the variation in the model and control samples had large positive PC1 score values while plant samples had large negative PC1 score values (Fig. 1A). All but a few compounds had positive PC1 loading coefficients (Fig. S3 and Fig. 1B) which indicated that the control samples had a higher concentration of all of these gasoline compounds compared to the plant samples before the epigeous plant parts were cut off. This indicates that almost all compounds were removed when the potted plant was present. The compounds with negative loading

Table 3 Exposure concentration and percentage removed of the selected VOCs from the gasoline mixture exposed to H. helix. Values are mean ± SE.

Heptane 3-Methylhexane Toluene Ethylbenzene m,p-Xylenes

Concentration in control chambers (mg∙m3)

Removal in epigeous period (%)

Removal in hypogeous period (%)

47.4 ± 1.2 104.5 ± 2.5 385.6 ± 7.5 28.5 ± 0.9 83.8 ± 3.3

17.3 ± 0.6 16.7 ± 0.6 18.4 ± 0.6 18.6 ± 0.8 17.7 ± 0.8

20.6 ± 0.8 19.2 ± 0.8 22.6 ± 0.7 21.8 ± 1.0 20.7 ± 1.0

Fig. 1. A) Scores and B) zoomed loading plot of selected RoIs for PCA of model aY1: both control and plant chambers (before the removal of the epigeous plant parts) with normalization to internal standards and volume sampled. The error bars in A) indicate ±1 standard error. The black dashed line in B) is the average normalized and combined SICs, and the red solid line is the PC1 loading coefficients. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

230

M. Dela Cruz et al. / Chemosphere 221 (2019) 226e234

coefficients had higher concentrations in the plant chambers compared to the controls (Fig. 1B). These compounds were plant or system related as they were not detected in the GC-MS analysis of the 0.5% gasoline in CS2. The projection of Exp. 2 onto the model showed that control samples from Exp. 2 (Exp2CY) had smaller score values than control samples from Exp. 1 (Exp1CY), whereas this was not seen for plant samples (Exp1PY and Exp2PY, Fig. 1A). The explained variance (EV) for Exp2CY was only 28% which indicated that the exposure composition or concentration in Exp. 2 differed slightly from the one in Exp. 1. Conversely, the EV for Exp2PY was 80% which indicated that the removal pattern was similar between the two experiments. This was confirmed by the cosine between loadings of models aY1 and aY2 (0.99), which showed that, in spite of the low signal to noise ratio (S/N), the two experiments yielded the same results when epigeous plants parts were present. The duplicate samples included in the test set clustered together with their respective sample along PC1 (see Fig. S4) which validated the preprocessing. Furthermore, all QC, QCD1, and QCD2 samples clustered together with a PC1 score close to 0 (Fig. 1A), which was expected because both control and plant samples were equally represented in the QC mixture. The error bars of the QC, QCD1, and QCD2 samples were smaller than or similar to the error bars of the control and plant samples (Fig. 1A) indicating that analytical variation was only responsible for a minor part of the variation seen along PC1. Furthermore QC, QCD1, and QCD2 samples were not different (Fig. 1A) indicating that the normalization was successful. To further investigate the difference between Exp1CY and Exp2CY, a sub model of model aY1 with these two sets in the training set and DupCY, QC, QCD1 and QCD2 in the test set was calculated. The model showed that the main difference between Exp1CY and Exp2CY was caused by a higher concentration of early eluting compounds and a lower concentration of late eluting compounds in Exp1CY than in Exp2CY and vice versa (see Fig. S5). The PCA for the hypogeous period and with focus on concentration (model aN1) showed similar results as model aY1 (Fig. S6). Control samples had large positive PC1 score values and plant samples had large negative PC1 score values with PC1 describing 98% of the variation (Fig. S6A). The corresponding loading plot (Fig. S6B) showed large positive PC1 loading coefficients for nearly all compounds, which indicated that their concentration was also lowered when epigeous plant parts were removed. It could be seen that a few more compounds exhibited negative loading coefficients compared to model aY1 indicating that more compounds now had higher concentrations in plant samples than in control samples (Figs. S3B and S6B). The findings were validated by the projection of Exp. 2 onto the model showing similar results as Exp. 1 (Fig. S6A). The EVs for Exp2CN and Exp2PN were 78% and 86%, respectively. The cosine between the two experiments (i.e. model aN1 vs. aN2) was 0.99 which indicated that Exp. 1 and Exp. 2 were identical. The duplicate samples included in the test set clustered together with their respective sample (see Fig. S7), which validated the preprocessing. The QC, QCD1, and QCD2 samples had negative PC1 score values indicating a closer resemblance with plant samples than control samples, but still showing that the normalization was successful and that the analytical variation was not responsible for the difference seen between plant and control samples along PC1 (Fig. S6A). From the visual inspection of PCA models calculated on the single ions, which focused on relative differences in compound removal within each RoI, three RoIs (viz. 1m41, 1m55 and1m69) appeared to yield a meaningful clustering of the samples and a separation between controls and plant chambers (not shown). The corresponding matrices were thus combined and analyzed in models sY, sN and sP. In all of these, the RoIs were normalized to

unitary Euclidean norm, which removed most concentration effects and made the PCA focus on differential removal pattern. In model sY1, PC1 explained 46% of the variance and control samples had positive scores for this component, while plant samples yielded negative ones (Fig. 2A). Thus, the compounds with positive loading coefficients were more abundant in the control chambers than in the plant chambers, which indicated that they were more effectively removed when plants were present. Likewise, those with negative loading coefficients were less effectively removed by the plants. The compounds with the largest loading coefficients, either positive or negative, were tentatively identified (Table S4). The compounds that were removed comparatively more easily were four dimethylcyclopentanes, methylcyclohexane, toluene, and two unidentified compounds (Fig. 2B, and Table 4). Compounds more resistant to removal were 3-methylhexane, heptane, ethylcyclopentane, two trimethylcyclopentanes and five dimethylcyclohexanes. The projection of Exp. 2 onto the model revealed that plant samples (Exp2PY) had smaller scores than plant samples from Exp. 1 (Exp1PY). However, the EVs for Exp2PY and Exp2CY were only 9% and 20%, respectively, and the cosine between the loadings in models sY1 and sY2 was 0.80. The visual comparison between the PC1 loadings of models sY1 and sY2 showed a very similar pattern, except for the loading coefficients for heptane and toluene, but the loadings for sY2 also clearly showed a higher degree of patterns for residual misalignment and peak shape changes (Fig. S8). These observations were consistent with the fact that the normalization to unitary Euclidean norm, by removing concentration effects, greatly increased the similarity between samples (e.g. PC1 of a model sY1 without mean centering explained 99.89% of the variation), which implied that the PCA for model sY1 (and sY2, sN1, sP1, etc.), focused on a small fraction of the total variance, had considerably worse S/N, and were likely more affected by residual misalignments or peak shape changes. However, the consistency between the two experiments, the fact that duplicate samples were clustered together with their respective sample (see Fig. S9), and the small error bars for QC, QCD1, and QCD2 samples, which indicated that the difference between plant and control chambers was not caused by analytical variation (Fig. 2A), suggested that the observed differential removal pattern was real, if very week and more pronounced in Exp. 1 than in Exp. 2, for the preprocessing did not work just as well. In model sN1, which is relative to the hypogeous period and focused on the differential removal of the compounds in the three most relevant RoIs, the control chamber samples (Exp1CN) had positive PC1 scores, whereas the plant samples had negative ones (Fig. S10A). Similarly to model sY1, the mean centering removed the bulk of the variation (99.9%). The first PC of the column centered matrix explained a relatively small fraction of the variance (36%) and the subsequent components explained both chemical variation and changes in peak shapes or the residual misalignment. The negative loading coefficients for 3-methylhexane, heptane, ethylcyclopentane, toluene and five unidentified compounds indicated that there was comparatively more of these compounds in the control samples than in the plant samples (Fig. S10, Table 4); therefore, they appeared to be more easily removed than the remaining compounds. However, the explained variance for samples from Exp. 2 was very small when projected on model sN1 (3.4% and 12% for Exp2PN and Exp2CN, respectively), and the cosine between the sN1 and sN2 loadings was 0.34, which was very low and indicative of substantial differences between the two models. The visual inspection of the loadings confirmed this, and that the main difference between the two experiments was for the five unidentified compounds in 1m69, as these showed more pronounced patterns for residual misalignment and peak shape

M. Dela Cruz et al. / Chemosphere 221 (2019) 226e234

231

Fig. 2. A) Score and B) loading plots for PCA of model sY1: both control and plant chambers (before the removal of the epigeous plant parts) with normalization to Euclidean norm. The error bars in A) indicate ±1 standard error. The black dashed line in B) is the average normalized and combined SICs, and the red solid line is the PC1 loading coefficients. The peak identifications are listed in Table S4. * denotes unidentified compounds with impact on the model. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Table 4 Overview of results from PCA models. Model

aY1 aY2 aN1 aN2

sY1 sY2 sN1 sN2a

sP1 sP2a a

Removal of compounds Compounds removed

Compounds not removed

All All All All

System System System System

gasoline gasoline gasoline gasoline

VOCs VOCs VOCs VOCs

or or or or

plant plant plant plant

related related related related

VOCs VOCs VOCs VOCs

Compounds easy to remove

Compounds difficult to remove

Four dimethylcyclopentanes, methylcyclohexane, toluene, and two unidentified compounds Four dimethylcyclopentanes, heptane, methylcyclohexane, and two unidentified compounds 3-Methylhexane, heptane, ethylcyclopentane, toluene and five unidentified compounds Heptane, ethylcyclopentane, and toluene

3-Methylhexane, heptane, ethylcyclopentane, two trimethylcyclopentanes and five dimethylcyclohexanes 3-Methylhexane, ethylcyclopentane, two trimethylcyclopentanes and five dimethylcyclohexanes Three dimethylcyclopentanes, methylcyclohexane, two trimethylcyclopentanes, four dimethylcyclohexanes, and four unidentified compounds 1,1-Dimethylcyclopentane, 1,1,2-trimethylcyclopentane, 1,1-dimethylcyclohexane, and three unidentified compounds

Compounds easy to remove during epigeous period

Compounds easy to remove during hypogeous period

Four dimethylcyclopentanes, methylcyclohexane, 1,1-dimethylcyclohexane, and five unidentified compounds Four dimethylcyclopentanes, methylcyclohexane, 1,1-dimethylcyclohexane, 1,1-dimethylcyclohexane, and one unidentified compound

3-Methylhexane, heptane, ethylcyclopentane, 1,1,2-trimethylcyclopentane, toluene, three dimethylcyclohexanes and six unidentified compounds 3-Methylhexane, heptane, ethylcyclopentane, 1,1,2-trimethylcyclopentane, toluene, two dimethylcyclohexanes and six unidentified compounds

Models sN2 and sP2 show high degree of residual misalignment or peak shape changes.

changes in Exp. 2 than in Exp. 1 (Fig. S11). Moreover, some inconsistencies could be found in the loadings: e.g. one of the dimethylcyclopentanes (compound no. 3) has negative loading coefficient in 1m41 and positive in 1m55 (Fig. S11). Also for this model, the duplicate samples included in the test set clustered together with their respective sample, but the QCD2 runs were not close to QC and QCD1 (see Fig. S12).

These results confirm that the pattern of differential removal is very weak and hardly discernible from background noise and residual misalignment and peak shape changes, when the epigeous plant is removed. In the last model, sP1, the focus was on the plant samples alone. In this case, PC1 described 55% of the variation (Fig. 3A), and plant samples for the epigeous period (Exp1PY) had positive scores while

232

M. Dela Cruz et al. / Chemosphere 221 (2019) 226e234

Fig. 3. A) Scores and B) loading plot for PCA of model sP1: plant chambers before and after the removal of the epigeous plant parts with normalization to Euclidean norm. The error bars in A) indicate ±1 standard error. The black dashed line in B) is the average normalized and combined SICs, and the red solid line is the PC1 loading coefficients. The peak identifications are listed in Table S4. * denotes unidentified compounds with impact on the model. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

plant samples for the hypogeous period (Exp1PN) had negative ones. The positive loading coefficients indicated that there was relatively more of 3-methylhexane, heptane, ethylcyclopentane, 1,1,2-trimethylcyclopentane, toluene, three dimethylcyclohexanes and six unidentified compounds in Exp1PY than in Exp1PN (Fig. 3B, Table 4). Hence these compounds were more easily removed during the hypogeous period. Like all other models in which the concentration effect was removed, the explained variance for the samples in Exp. 2 when projected on this model was low (7.4% and 19% for Exp2PY and Exp2PN, respectively) and the cosine between the loadings of sP1 and sP2 (i.e. between the two experiments) was 0.68, which indicated that there likely was a significant difference in the differential removal pattern between the two experiments. Visual inspection of the loadings allowed to attribute this difference to residual misalignment or peak shape changes, but also to the five unidentified compounds eluting between 11.5 and 12 min in 1m69 (which showed no influence on the model for Exp. 2), one unidentified compound and 1,1-dimethylcyclohexane (compound 14) eluting after 14 min showing opposite loading coefficients compared to Exp. 1 (Fig. S13). The duplicate samples included in the test set clustered together with their respective sample (Fig. S14), and the small error bars for QC, QCD1, and QCD2 (Fig. 3A) samples, indicated that the difference between plant samples with and without epigeous plant parts was not caused by analytical variation thereby validating the preprocessing and the normalization. 4. Discussion The quantitative analysis revealed that, in presence of H. helix, the concentrations of heptane, 3-methylhexane, toluene, ethylbenzene, and m,p-xylenes were reduced by 16.7e22.6% compared to the controls, which demonstrates that a potted plant can remove

several compounds at the same time. Furthermore, the removal of these compounds was stable over time, apart from a short adaptation period of no more than five days. The pixel-based analysis using model aY1 (plant vs. control chambers with normalization to IS and volume sampled) clearly documents that nearly all compounds detected by the GC-MS analysis were reduced in concentration when plants were present compared to control chambers. This was also confirmed by the fact that the area of the TIC for plant chambers was reduced in the range of 14e32% compared to control chambers. The few compounds for which there was a higher concentration in the plant chambers than in the control chambers were system or plant related as they were not detected in GC-MS analysis of the 0.5% gasoline in CS2. One of the compounds was tentatively identified as a terpene, possibly a-pinene which is a plant derived substance. However, a-pinene has been observed to be removed by H. helix in a previous study (Yang et al., 2009). Gasoline has previously been reported to be removed by plants but it was not specified how many or which of the compounds that were removed (Oyabu et al., 2003). The discovery that all volatile gasoline constituents are removed by potted plants in our study indicates that the improvement of indoor air quality is not limited to only a few compounds. Interestingly, the quantitative analysis revealed that the removal efficiencies for the quantified compounds were higher when the epigeous plant parts were absent than when present. This indicates that the biological removal of VOCs may primarily be carried out by the soil microcosm. Increased removal of formaldehyde by removing leaves of Chlorophytum elatum has previously been reported with an explanation that leaves covering the soil surface hindered diffusion of the pollutant into the soil for microbial degradation (Godish and Guindon, 1989). The leaves of H. helix partially covered the soil surface in this study and could potentially

M. Dela Cruz et al. / Chemosphere 221 (2019) 226e234

hinder the diffusion of the gasoline mixture into the soil. This could also indicate that the process of diffusion into the soil is a rate limiting step for pollutant removal by potted plants. Alternatively, the increased removal of gasoline compounds upon removal of the epigeous plant parts may also be due to a continuous increase in gasoline degrading microorganisms throughout the experiment (Wood et al., 2002). It is important to mention that the removal of VOCs may not be maintained by the soil microcosm without the epigeous plant parts (Wood et al., 2002). This may be the case because the VOCs are of too low concentration to support microbial growth and another substrate for growth is needed before the microorganisms can utilize the VOCs for energy thus removing them from the air (Guieysse et al., 2008). Differences in the relative removal of compounds were observed and depended on whether the epigeous plant parts were present or not. These differences accounted for a minor variation in the data set and were mainly seen in Exp. 1 but also to a limited extend in Exp. 2 despite the change in gasoline composition from Exp. 1 to Exp. 2. The plants in Exp. 2 were smaller than in Exp. 1 which may offer an explanation for the difference between experiments as the support for the microorganisms in the soil microcosm would have been less in Exp. 2 than in Exp. 1. Especially four dimethylcyclopentanes and methylcyclohexane were more easily removed relative to other compounds when the epigeous plant parts were present than when absent. An explanation may be found in their ability to be absorbed by water as the dimethylcyclopentanes and methylcyclohexane are more water soluble and have a lower Henry's law constant than e.g. heptane and 3-methylhexane: the stomata offer a water site for absorption of compounds, which, combined with a slightly hindered diffusion of compounds into the soil, may have caused these compounds to be relatively more removed than other compounds. This explanation, however, cannot account for the behavior of ethylcyclopentane which should be absorbed by water to the same degree as the dimethylcyclopentanes (Fig. S15). This discrepancy may be due to the presence of unknown processes or could be an effect of the low S/N for the data set. The current data set does not allow determining the exact reasons. When the epigeous plant parts were absent, it may be hypothesized that microorganisms in the soil are responsible for the majority of the removal of compounds (Orwell et al., 2004; Wood et al., 2002). Adsorption to particles in the soil and uptake by plant roots may also play a role (Calvet, 1989; Gao et al., 2010; Wild et al., 2005). In microbial biodegradation of gasoline-range petroleum compounds in sediment extracts, heptane was preferentially removed compared to compounds eluting from cyclohexane to methylcyclohexane excluding l,cis-2-dimethylcyclopentane (Thompson, 1983). Similar findings were obtained when plant parts had been cut off, i.e. heptane was more easily removed than dimethylcyclopentanes and methylcyclohexane, compounds eluting between cyclohexane and methylcyclohexane. 3Methylhexane was likewise more easily removed relative to dimethylcyclopentanes and methylcyclohexane. A faster depletion of methylhexanes compared to dimethylcyclopentanes and methylpentanes by microorganisms has previously been reported for crude oils from the Barrow Island oilfield (George et al., 2002). Interestingly it can also be seen that 1,1-dimethylcyclopentane and 1,1-dimethylcyclohexane have rather large loading coefficients compared to the peak height (Fig. 3B and S10B) indicating that these have a relatively high influence on the model and are particularly difficult to remove for the soil microcosm. This could be because of the quaternary carbon, whose presence has been shown to increase recalcitrance for microorganisms in e.g. trimethylalkanes (Solano-Serena et al., 1998, 1999). Furthermore, 1,1dimethylcyclopentane is known to be recalcitrant to

233

biodegradation (Halpern, 1995; Sharaf and El Nady, 2006). The results obtained when epigeous plant parts were absent suggest that microorganisms in the soil are involved in the removal of gasoline compounds. This is to our knowledge the first study describing detailed removal of gasoline VOCs by a potted plant over an extended period of time. The results demonstrate that potted plants can reduce the concentration of all compounds in a complex mixture of VOCs which is of significance in indoor environments where VOCs coexist in complex mixtures. It is not known for how long VOC removal can be sustained once the epigeous plant parts are removed, although this experiment suggests that it may be for at least eight days. Further, the incorporation of potted plants in the indoor environment has a psychological aspect as well which is related to the epigeous plant parts (Bringslimark et al., 2009; Thomsen et al., 2011). Therefore, it seems that the best policy is to keep the epigeous plant parts but keep the soil surface uncovered in order to allow for as much diffusion of VOCs into the soil microcosm as possible. Compounds eluting before 1,1-dimethylcyclopentane were not included in the data analysis. To what degree these compounds are removed by potted plants remains unknown, but this is highly relevant to investigate as they can potentially have a pronounced impact on indoor air quality due to their relatively high vapor pressure. 5. Conclusion Potted Hedera helix exposed to a gasoline mixture under dynamic chamber conditions was able to reduce the concentration of gasoline VOCs. The only VOCs from the pixel-based analysis that did not appear to be removed were plant or system related. A significantly higher percentage of gasoline VOCs were removed when epigeous plant parts had been removed compared to when they were present indicating that the lower part of the potted plant i.e. the soil microcosm is mainly responsible for the observed reduction in concentration of gasoline compounds. The preferential removal of e.g. heptane over dimethylcyclopentanes supports this hypothesis. The fact that removal of VOCs increased with the removal of epigeous plant parts appears likely to be the result of increased diffusion of the VOCs into the soil microcosm or an increase in gasoline degrading microorganisms. Declaration of interest None. Acknowledgement This study was funded by a PhD grant from the University of Copenhagen, Faculty of Sciences. The authors thank Multigreen. dk A/S for supplying the plants. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi.org/10.1016/j.chemosphere.2018.12.125. References Araújo, A.E.O., Mezzomo, B.P., Ferrari, I., Grisolia, C.K., 2010. Genotoxic effects caused by indoor exposure to petroleum derivatives in a fuel quality control laboratory. Genet. Mol. Res. 9, 1069e1073. €chler, M., Bolker, B., Walker, S., 2015. Fitting linear mixed-effects Bates, D., Ma models using lme4. J. Stat. Softw. 67, 1e48. Batterman, S., Jia, C., Hatzivasilis, G., 2007. Migration of volatile organic compounds

234

M. Dela Cruz et al. / Chemosphere 221 (2019) 226e234

from attached garages to residences: a major exposure source. Environ. Res. 104, 224e240. Benites, C.I., Amado, L.L., Vianna, R.A.P., da Graca Martino-Roth, M., 2006. Micronucleus test on gas station attendants. Genet. Mol. Res. 5, 45e54. Bringslimark, T., Hartig, T., Patil, G.G., 2009. The psychological benefits of indoor plants: a critical review of the experimental literature. J. Environ. Psychol. 29, 422e433. Calvet, R., 1989. Adsorption of organic chemicals in soils. Environ. Health Perspect. 83, 145e177. Christensen, J.H., Tomasi, G., 2007. Practical aspects of chemometrics for oil spill fingerprinting. J. Chromatogr. A 1169, 1e22. Christensen, J.H., Tomasi, G., de Lemos Scofield, A., Meniconi, M.F.G., 2010. A novel approach for characterization of polycyclic aromatic hydrocarbon (PAH) pollution patterns in sediments from Guanabara Bay, Rio de Janeiro, Brazil. Environ. Pollut. 158, 3290e3297. Christensen, J.H., Tomasi, G., Hansen, A.B., 2005. Chemical fingerprinting of petroleum biomarkers using time warping and PCA. Environ. Sci. Technol. 39, 255e260. Cornejo, J.J., Munoz, F.G., Ma, C.Y., Stewart, A.J., 1999. Studies on the decontamination of air by plants. Ecotoxicology 8, 311e320. Dela Cruz, M., Christensen, J.H., Thomsen, J.D., Müller, R., 2014a. Can ornamental potted plants remove volatile organic compounds from indoor air? - a review. Environ. Sci. Pollut. R. 21, 13909e13928. Dela Cruz, M., Müller, R., Svensmark, B., Pedersen, J.S., Christensen, J.H., 2014b. Assessment of volatile organic compound removal by indoor plants - a novel experimental setup. Environ. Sci. Pollut. R. 21, 7838e7846. Du, L., Batterman, S., Godwin, C., Rowe, Z., Chin, J.Y., 2015. Air exchange rates and migration of VOCs in basements and residences. Indoor Air 25, 598e609. Edwards, R.D., Jurvelin, J., Saarela, K., Jantunen, M., 2001. VOC concentrations measured in personal samples and residential indoor, outdoor and workplace microenvironments in EXPOLIS-Helsinki, Finland. Atmos. Environ. 35, 4531e4543. Gallotta, F.D.C., Christensen, J.H., 2012. Source identification of petroleum hydrocarbons in soil and sediments from Iguaçu River Watershed, Paran
a, Brazil using the CHEMSIC method (CHEMometric analysis of Selected Ion Chromatograms). J. Chromatogr. A 1235, 149e158. Gao, Y., Ren, L., Ling, W., Gong, S., Sun, B., Zhang, Y., 2010. Desorption of phenanthrene and pyrene in soils by root exudates. Bioresour. Technol. 101, 1159e1165. George, S.C., Boreham, C.J., Minifie, S.A., Teerman, S.C., 2002. The effect of minor to moderate biodegradation on C5 to C9 hydrocarbons in crude oils. Org. Geochem. 33, 1293e1317. Godish, T., Guindon, C., 1989. An assessment of botanical air purification as a formaldehyde mitigation measure under dynamic laboratory chamber conditions. Environ. Pollut. 62, 13e20. Gordian, M.E., Stewart, A.W., Morris, S.S., 2010. Evaporative gasoline emissions and asthma symptoms. Int. J. Environ. Res. Publ. Health 7, 3051e3062. Guieysse, B., Hort, C., Platel, V., Munoz, R., Ondarts, M., Revah, S., 2008. Biological treatment of indoor air for VOC removal: potential and challenges. Biotechnol. Adv. 26, 398e410. Halpern, H.I., 1995. Development and applications of light-hydrocarbon-based star diagrams. AAPG Bull. 79, 801e815. Harley, R.A., Kean, A.J., 2004. Chemical Composition of Vehicle-related Volatile Organic Compound Emissions in Central California. Department of Civil and Environmental Engineering, University of California, USA, pp. 1e43. Hun, D.E., Corsi, R.L., Morandi, M.T., Siegel, J.A., 2011. Automobile proximity and

indoor residential concentrations of BTEX and MTBE. Build. Environ. 46, 45e53. Jolliffe, I.T., Cadima, J., 2016. Principal component analysis: a review and recent developments. Philos. T. R. Soc. A. 374, 20150202. Logue, J.M., Mckone, T.E., Sherman, M.H., Singer, B.C., 2011. Hazard assessment of chemical air contaminants measured in residences. Indoor Air 21, 92e109. Nielsen, N.P.V., Carstensen, J.M., Smedsgaard, J., 1998. Aligning of single and multiple wavelength chromatographic profiles for chemometric data analysis using correlation optimised warping. J. Chromatogr. A 805, 17e35. Orwell, R.L., Wood, R.A., Burchett, M.D., Tarran, J., Torpy, F., 2006. The potted-plant microcosm substantially reduces indoor air VOC pollution: II. Laboratory study. Water Air Soil Pollut. 177, 59e80. Orwell, R.L., Wood, R.L., Tarran, J., Torpy, F., Burchett, M.D., 2004. Removal of benzene by the indoor plant/substrate microcosm and implications for air quality. Water Air Soil Pollut. 157, 193e207. Oyabu, T., Takenaka, K., Wolverton, B., Onodera, T., Nanto, H., 2003. Purification characteristics of golden pothos for atmospheric gasoline. Int. J. Phytoremediat. 5, 267e276. Porter, J.R., 1994. Toluene removal from air by Dieffenbachia in a closed environment. Adv. Space Res. 14, 99e103. Sharaf, L.M., El Nady, M.M., 2006. Application of light hydrocarbon (C7þ) and biomarker analyses in characterizing oil from wells in the North and North Central Sinai, Egypt. Petrol. Sci. Technol. 24, 607e627. Skov, T., van den Berg, F., Tomasi, G., Bro, R., 2006. Automated alignment of chromatographic data. J. Chemometr. 20, 484e497. Solano-Serena, F., Marchal, R., Blanchet, D., Vandecasteele, J.P., 1998. Intrinsic capacities of soil microflorae for gasoline degradation. Biodegradation 9, 319e326. Solano-Serena, F., Marchal, R., Ropars, M., Lebeault, J.M., Vandecasteele, J.P., 1999. Biodegradation of gasoline: kinetics, mass balance and fate of individual hydrocarbons. J. Appl. Microbiol. 86, 1008e1016. Thompson, K.F.M., 1983. Classification and thermal history of petroleum based on light hydrocarbons. Geochim. Cosmochim. Ac. 47, 303e316. Thomsen, J.D., Sønderstrup-Andersen, H.K.H., Müller, R., 2011. People-plant relationships in an office workplace: perceived benefits for the workplace and employees. Hortscience 46, 744e752. Tomasi, G., van den Berg, F., Andersson, C., 2004. Correlation optimized warping and dynamic time warping as preprocessing methods for chromatographic data. J. Chemometr. 18, 231e241. Wild, E., Dent, J., Thomas, G.O., Jones, K.C., 2005. Direct observation of organic contaminant uptake, storage, and metabolism within plant roots. Environ. Sci. Technol. 39, 3695e3702. Wolkoff, P., Wilkins, C.K., Clausen, P.A., Nielsen, G.D., 2006. Organic compounds in office environments - sensory irritation, odor, measurements and the role of reactive chemistry. Indoor Air 16, 7e19. Wood, R.A., Orwell, R.L., Tarran, J., Torpy, F., Burchett, M., 2002. Potted-plant/growth media interactions and capacities for removal of volatiles from indoor air. J. Hortic. Sci. Biotechnol. 77, 120e129. World Health Organization, 2010. WHO Guidelines for Indoor Air Quality: Selected Pollutants. World Health Organization, Regional Office for Europe, Copenhagen. Yang, D.S., Pennisi, S.V., Son, K.C., Kays, S.J., 2009. Screening indoor plants for volatile organic pollutant removal efficiency. Hortscience 44, 1377e1381. Yoo, M.H., Kwon, Y.J., Son, K.C., Kays, S.J., 2006. Efficacy of indoor plants for the removal of single and mixed volatile organic pollutants and physiological effects of the volatiles on the plants. J. Am. Soc. Hortic. Sci. 131, 452e458.