Biofiltration of airborne VOCs with green wall systems—Microbial and chemical dynamics

Botanical air filtration is a promising technology for reducing indoor air contaminants, but the underlying mechanisms need better understanding. Here, we made a set of chamber fumigation experiments of up to 16 weeks of duration, to study the filtration efficiencies for seven volatile organic compounds (VOCs; decane, toluene, 2-ethylhexanol, α-pinene, octane, benzene, and xylene) and to monitor microbial dynamics in simulated green wall systems. Biofiltration functioned on sub-ppm VOC levels without concentration-dependence. Airflow through the growth medium was needed for efficient removal of chemically diverse VOCs, and the use of optimized commercial growth medium further improved the efficiency compared with soil and Leca granules. Experimental green wall simulations using these components were immediately effective, indicating that initial VOC removal was largely abiotic. Golden pothos plants had a small additional positive impact on VOC filtration and bacterial diversity in the green wall system. Proteobacteria dominated the microbiota of rhizosphere and irrigation water. Airborne VOCs shaped the microbial communities, enriching potential VOC-utilizing bacteria (especially Nevskiaceae and Patulibacteraceae) in the irrigation water, where much of the VOC degradation capacity of the biofiltration systems resided. These results clearly show the benefits of active air circulation and optimized growth media in modern green wall systems.


INTRODUCTION
Green walls (also known as plant walls, plant-based biowalls, botanical biofilters) are vertical structures in which one or several houseplant species are grown on soil or a soilless support fabric or growth medium. In active green walls the volume of indoor air exposed to the system is increased by actively drawing air through the support, which together with the roots embedded in it is kept moist by regular or constant drippling irrigation. 1 Supplementary lighting may also be provided. Besides the reported positive effects of indoor plants on staff wellbeing, productivity and job satisfaction, 1-3 such hydroculturebased systems can effectively reduce indoor air CO 2 concentration 4 and increase humidity. 5

Accepted Article
This article is protected by copyright. All rights reserved.
Much of the research on house plants and green walls has focused on their potential to remove harmful indoor air contaminants, especially volatile organic compounds (VOCs). 1,2,[6][7][8][9] VOCs are chemically diverse and detectable in indoor air as a mixture of tens to hundreds of different compounds. 1,10,11 VOCs in indoor air are often associated with sick building syndrome, and are typically released from petrochemical-derived materials such as furnishings, paints, solvents and textiles. 10,12 2-ethylhexanol, commonly detected in indoor air, can be produced upon microbial degradation of plasticizers. 13 Besides producing, 14 microbes also consume (biodegrade) organic pollutants with seemingly unlimited catabolic potential to utilize different chemicals. [15][16][17][18] Plants have been used to inoculate and support diverse bacterial degrader communities for the clean-up of soil (rhizoremediation) or water (phyto/rhizofiltration). [18][19][20] Purely microbiological systems, or biofilters, have also been developed and operated to purify contaminated air even at an industrial scale, but the science of plant-assisted remediation systems for air is newer and less developed. 1 For hydroculture-based biofilters, the proposed working mechanism for air VOC removal depends on the partitioning of pollutants from the gaseous to the aqueous phase, driven by concentration difference as microbes constantly consume pollutants from the water. [15][16][17]21 Although biological indoor air purification systems show demonstrated potential, their microbiology is understudied, 1 as is the specific effect of plants on it. 9 According to Guieysse et al. 22 , one of the major challenges of biological indoor air treatment systems is to inoculate, express and maintain a diverse microbial degrader community. Other identified knowledge gaps include the comparison of passive systems to those with active air circulation, 1 removal efficiency of VOCs in mixtures of chemically dissimilar compounds supplied in a continuous manner, and other factors that can influence the performance and stability of the process. 2,9 Accepted Article This article is protected by copyright. All rights reserved.
Despite these gaps in knowledge about their exact functional mechanisms, several biological air cleaning products were commercially available in different countries in the early 2010s, as reviewed by Torpy et al. 1 One of the commercial active green wall systems launched on the market after that is a botanical biotrickling filter (as visualized in Soreanu et al. 9 ) by Naava Ltd (Jyväskylä, Finland), which as of September 2017 is available in Finland, Sweden and the United States. 23 By utilizing or simulating the components of this commercial system, we aimed to characterize the chemical and microbial dynamics of phytotechnological VOC filtration with a set of chamber fumigation tests. Experiments were carried out to better understand the underlying mechanisms of VOC removal and the relative contributions of the different components of the system: air circulation, growth medium, and plants. Rates of removal of five to seven different VOCs were quantified by gas chromatography-mass spectrometry (GC-MS) in experiments lasting from 20 hours to 16 weeks, and bacterial succession was monitored both in the rhizosphere and irrigation water by high-throughput sequencing.

Small chamber experiment setup: short-term effect of plant species and air circulation
Short-term small-scale experiments were carried out in airtight glass desiccators (volume 22.3 l, Duran; Figure 1, S1). The test plants were golden pothos (Epipremnum pinnatum cv. aureum) and white rabbit's foot fern (Davallia fejeensis Hook), supplied by Hydro Huisman Plants were maintained in their original potting soil or transferred to soilless growing

Accepted Article
This article is protected by copyright. All rights reserved. medium, one plant per pot, after thorough washing of the root system with potable tap water, by Naava Ltd, Jyväskylä, Finland. The soilless growth medium used was Naava growth mix (Nmix) used in the commercial Naava Smart Green Walls, consisting of activated carbon and other granular constituents. 24 Each pot (height 14.5 cm, width 13.5×13.5 cm) was a green wall prototype with its own fan (Figure 1, S1 added upon the start of the experiment, with 1 µl of the mix (0.2 µl of each VOC) injected into a foil cup and dropped into each of the desiccators, which were sealed immediately after. The experiment was maintained and monitored for 20-21 h. Experiments with fern and golden pothos were repeated three and six times, respectively.

Large chamber experiment 1: long-term comparison of growth media
The long-term large chamber experiments were carried out in 7-8 airtight environmental chambers made of glass, size 60×60×100 cm (volume 0.36 m 3 ; Figure 1, S1). The test plant was golden pothos, which were transferred to growth medium, one plant per pot. Large chamber experiment 1 compared two soilless growth media: Leca ® (lightweight expanded

Accepted Article
This article is protected by copyright. All rights reserved.

Accepted Article
This article is protected by copyright. All rights reserved. and monitored for 16 weeks. Owing to the variation in the room temperature and humidity as well as the air velocity at the fan that created the ingoing air stream, the VOC concentrations delivered to the fumigation chambers fluctuated during the experimental period from a minimum total VOC input of 1.7 ppm (parts per million, volume) at week 3 to a maximum of 4.3 ppm at week 7 ( Figure S2).

Large chamber experiment 2: long-term comparison of plant effect
Large chamber experiment 2 compared the commercially used Nmix with and without golden pothos. Triplicate chambers were supplied with seven green wall prototypes with plants, and triplicate chambers with seven otherwise identical prototypes but without plants. Irrigation and other conditions were as described above for both. One control chamber was left empty, containing neither pots nor irrigation water. VOCs were supplied to the seven chambers as described above ( [>99%]), with the exception that each chemical was supplied from a separate container. In addition, one chamber with seven green wall prototypes with plants and irrigation water had a separate supply of ambient (non-fumigated) air, to monitor microbial development in the absence of VOC input. The experiment was maintained and monitored for 8 weeks. The total input VOC concentrations varied over the course of the experiment from a minimum of 1.6 ppm at week 5 to a maximum of 5.4 ppm at week 7 ( Figure S2).

Accepted Article
This article is protected by copyright. All rights reserved.

Accepted Article
This article is protected by copyright. All rights reserved.

Accepted Article
This article is protected by copyright. All rights reserved.

Bacterial 16S rRNA gene amplification and sequencing
Preliminary tests revealed that up to 80% of 16S rRNA gene sequences from golden pothos root DNA extract, when processed with a regular bacterial amplicon sequencing pipeline, were chloroplasts, whereas the presence of mitochondria was minimal. A semi-nested polymerase chain reaction (PCR) approach was thus applied, utilizing chloroplastdiscriminating bacterial reverse primer 783r (equimolar mix of variants a, b and c) 26

Accepted Article
This article is protected by copyright. All rights reserved.

following the Standard
Operating Procedure for 454 amplicon data. 28

Accepted Article
This article is protected by copyright. All rights reserved. Whitney U test for differential relative abundances of bacterial taxa between substrate sources, manually Bonferroni-corrected for three comparisons in water) were performed with OriginPro 2017 (OriginLab Corporation). Bacterial 16S rRNA gene sequences with MIMARKS details have been submitted to NCBI Sequence Read Archive under BioProject NNNN (submitted, pending).  Table S1). Reduction, compared to controls without plants or growth medium, was notably greater with the commercial Naava growth mix (Nmix) than with soil (Table S2). Such immediate VOC loss could be attributed to active uptake by plants, 2,[6][7][8]15,18 or to partitioning from the gaseous phase to moist surfaces of the plants and chambers or solid phases of the growth media. 8,15 The importance of the latter mechanism was supported by the observation that active circulation of air through growth medium and rhizosphere further improved the loss; potted plants with Nmix and a fan reduced the concentrations of all added VOCs to below the detection limit in 20-21 h ( Figure 2; Table S2). The superiority of the soilless growth medium compared with soil may

Accepted Article
This article is protected by copyright. All rights reserved.

relate to the chemical (absorption by activated carbon) or microbiological (growth support)
properties of the Nmix medium, the former being more likely at least after 1 hour of VOC exposure. In a 1-week comparison of VOC removal by plants in potting soil and hydroculture (perlite/vermiculite saturated with fertilized water) without active air circulation, Irga et al. 4 reported slightly slower benzene removal for the soilless medium. In our experiment, it is possible that the difference between the two media was partly attributable to decreased air flow through more compact potting soil compared with granular Nmix, due to which followup experiments were done to compare Nmix with another granular soilless growth medium.

Accepted Article
This article is protected by copyright. All rights reserved.

In large chamber experiment 1, plants in two granular soilless growth media -Nmix and
Leca -showed initially identical VOC removal rates, irrespective of the VOC (Figure 3, left panel; Figure S3; Table S3, S4). This, together with the observation of removal rates not improving over time due to possible biological acclimation, implies that the immediate VOC dissipation was unlikely to be due to a biological process. As no difference was seen between the two granular media in the beginning, we hypothesize that initial removal of VOCs occurred simply by their partitioning from the fumigated air to the clean aqueous phase, efficiently taking place on the moist surfaces of both growth media and the leaf surfaces of plants. An alternative hypothesis is that even the pre-experiment conditions without fumigation had sufficiently acclimatized the plants and/or microbes to process the additional VOC input, but this seems unlikely with mean background VOC levels ranging  Table S4). Airflow through the growth medium and rhizosphere -one suspected explanation for the superiority in comparison with more compact potting soil in small chamber experiments -did not explain

Accepted Article
This article is protected by copyright. All rights reserved.
the superiority of Nmix, as flow rate was higher for Leca than Nmix. However, with the flow rates applied (determined by pot and medium characteristics and fan), the airflow versus residence time may have been more optimal in the green wall prototypes with the commercial green wall growth medium. According to Torpy et al. 1 , these are the key attributes for maximizing the efficiency of any biological air purification system. Other potential explanations for the more efficient VOC transfer from air to irrigation water include lower water VOC concentrations in Nmix due to better microbial biodegradation activity, or irreversible sorption of VOCs by the activated carbon. Interestingly, the efficacy of the system was relatively independent of VOC chemical structure, removal rates for Nmix ranging from 60 to 70% with two remaining prototype units and a relative humidity of <60% in week 12. Removal efficiency per remaining prototype unit approximately doubled during the 12 weeks for Nmix, and remained relatively stable for Leca (data not shown). However, after the last units were removed upon week 12 sampling, VOC dissipation dropped to negligible in chambers with just fertilized irrigation water.

Accepted Article
This article is protected by copyright. All rights reserved. Table S3, S4). This may be attributed either to active VOC uptake by plants, 2,7,8,15,18 or to plants inoculating a more active or better acclimatized degrader community into the growth medium and/or irrigation water. 2,22 As the number of units containing plants decreased below 3 (due to plant removal for destructive root sampling), the relative effectiveness of the chambers with plants decreased below the effectiveness of the chambers without plants (but which had a constant 7 units). The most probable explanation is that with fewer green wall prototypes -fewer fans and a smaller moist growth medium surface area -VOC transfer from air to water became the limiting factor, even if removal from the aqueous phase by biodegradation and/or irreversible sorption remained high. However, even in week 6 with one prototype with a plant per chamber, removal ratios were 40-60% for different test VOCs (Figure 3, S3). In this week, the ratio "airflow through medium/airflow through chamber" roughly matched the 1/3 ratio of "hourly airflow through commercial green wall/air volume exchanged hourly by mechanical ventilation", making it the closest estimation of a real life situation reached by our chamber setup, except for the mostly exaggerated input VOC concentrations. After the last units with plants were removed, VOC dissipation was close to negligible in chambers with just irrigation water.

Effect of growth media on rhizosphere bacterial community response to VOCs
Root microbiomes analyzed from large chamber experiment 1 showed that by the start of the VOC fumigation, bacterial communities in the two different soilless growth media had diverged in composition, but their diversity was identical ( Figure 4). However, under VOC exposure Nmix sustained a more diverse rhizosphere microbial community than Leca; phylogenetic diversity of root microbiomes in Leca decreased notably upon introduction of

Accepted Article
This article is protected by copyright. All rights reserved.
VOCs, and never recovered. Community composition also remained distinct between the two media. Golden pothos growth, on the other hand, did not differ between the media, and the plants showed no visible symptoms of VOC stress. The observed decreases in bacterial diversity can be caused by selective death or by selective growth of certain community members -the latter is often seen upon introduction of readily degradable organic contaminants. 16,34 Apparently the bacterial community sustained by roots in Leca was less resilient to the effect of VOCs, or alternatively VOCs were more bioavailable and able to cause either toxic or growth-supporting substrate effects in the Leca medium. The latter could be explained by the capacity of Nmix to absorb VOCs. Sorption capacity was not exhausted during the 12-week experiment with the ppm-level VOC input used, which was indicated by no drop in diversity observed in Nmix.

Effect of VOCs and plants on irrigation water bacterial community succession
In large chamber experiment 2, analysis of bacterial communities from both roots and irrigation water revealed distinct microbiomes in the two locations, even though irrigation ensured daily interaction between the two niches (generalized discriminant analysis P=0.0001 with 98% samples correctly classified). The rhizosphere communities neither showed clear temporal dynamics, nor were significantly different between fumigated and non-fumigated plants in the last sampling week, supporting the potential of Nmix medium to buffer the VOC-impact in the rhizosphere observed in large chamber experiment 1 ( Figure S4).

Accepted Article
This article is protected by copyright. All rights reserved.
Plants had an initial positive impact on bacterial phylogenetic diversity in the irrigation water (week 1 ANOVA P=0.0006; Figure 5). This confirmed the inoculation effect of plants, which is suspected to play a role in the initially higher removal rates of some VOCs in the chambers with plants. Fumigation, on the other hand, decreased bacterial diversity during the experiment compared to the non-fumigated chambers (week 3 ANOVA P=0.0002, week 6 ANOVA P=0.01). However, at the last microbial sampling point microbial biomass was twofold higher in the fumigated chambers (with one or zero plants) compared with the nonfumigated chambers (with 3 plants) (ANOVA week 6 P=0.01; Figure 5). This confirmed the theory that VOC-associated drop in diversity was caused by selective favoring of some heterotrophic bacterial groups that could use VOCs as a source of carbon and energy. [16][17][18] Interestingly, the rather moderate ppm-level VOC fumigation had a greater impact on the aquatic microbial community than did the exudates from plant roots.
With the exception of 2-ethylhexanol in experiment 2, the concentrations applied in the large chamber experiments were unrealistically high compared to VOC concentrations in normal office air, which are typically at the ppb-level. 10,12 Such exaggerated concentrations were used to stress the chemical/absorptive potential of the system and to see differences more clearly; the functionality of the actual commercial active green wall at low input of a single VOC has been reported earlier. 24 More and more evidence is being accumulated on the potential of soil and plant-associated microbes to utilize extremely low atmospheric concentrations of substrates such as H 2 . 18,35 In fact, uptake of gaseous compounds is not necessary, as long as concentrations in the irrigation water remain low (by microbial degradation and/or sorption) to support VOC partitioning to the aqueous phase as air is

Accepted Article
This article is protected by copyright. All rights reserved. circulated through moist Nmix. However, ppb-level VOC concentrations, even if taken up by high-affinity enzymes, will not necessarily support the growth or even long-term maintenance of a diverse catabolic microbial community -a bottleneck identified by Guieysse et al. 22 in the development of technology for biological treatment of indoor air. In the active green wall systems, plants are expected to feed and maintain a diverse and abundant microbiome even at times of low VOC supply. Indeed, microbial biomass of irrigation waters of commercial Naava Smart Clean Walls in school and office locations was comparable to that observed in large chamber experiment 2, with DNA yields ranging from 1 to 3 ng ml -1 (unpublished data).

Identification of potentially VOC-utilizing bacteria
Both irrigation water and rhizosphere soil were dominated by Proteobacteria ( Figure S5). In the rhizosphere, the two most common orders were Rhizobiales (Alphaproteobacteria) and Burkholderiales (Betaproteobacteria), matching the findings of Russell et al. 21

Accepted Article
This article is protected by copyright. All rights reserved.
ability to degrade hydrocarbons such as alkanes and phenols. 36 The nearly-as-abundant Nevskia is often detected on the surface of freshwater environments as well as in soil. 37 Patulibacteraceae, on the other hand, was more abundant in the rhizosphere than irrigation water. This family was originally isolated from soil, 38 and its members have recently been shown to degrade complex contaminants such as ibuprofen and N-methyl-2-pyrrolidone in aquatic environments. 39 Xanthobacteraceae, a third VOC-favored family, consisted mostly of Ancylobacter, a facultative methylotroph capable of using organochlorines as the sole carbon and energy source. 40 Very little is known about the potential VOC-degrading bacteria in green wall settings.
Russell et al. 21 -to our knowledge the only earlier comparable (i.e. cultivation-independent) green wall study -identified Hyphomicrobium as a rhizosphere-associated genus that responded positively to both exposure to VOCs as well as to growth/maturation of the plants. In our study, VOCs caused no statistically significant increase in Hyphomicrobiaceae abundance in the irrigation water, when considering all the sampling times together, whereas in the rhizosphere a VOC-associated decrease was seen ( Figure 6). We did notice a steeper increase from week 1 to week 6 in the fumigated chambers compared to the nonfumigated one, but interestingly, the genera Devosia and especially Prosthecomicrobium were much more abundant than Hyphomicrobium. Together with the study of Russell et al., 21 our results indicate that members of the Hyphomicrobiaceae may be global green wall system inhabitants. Our ongoing study will map their abundance in commercial green walls at customer locations. However, experiments with 13 C or 14 C-labelled VOCs are still required to confirm their potential to utilize and degrade airborne VOCs.

Accepted Article
This article is protected by copyright. All rights reserved.
The observed VOC-dependent increase in planktonic irrigation water microbial biomass and potential degrader bacteria indicate that significant VOC utilization potential resides in the aqueous phase of the active green wall system. Our results thus contradict the earlier assumption that quantitative VOC biodegradation takes place in the solid matrix of green walls: the rhizosphere and plant growth medium. 1 These findings also encourage further research into the aqueous microbial communities of hydroponic green wall systems, which may respond to external conditions and substrates (such as VOCs) more sensitively than the rhizosphere microbiome.