OPEN Global transcriptomic responses
Volatile organic compounds ( VOCs ) are commonly used as solvents in various industrial settings . 
Many of them present a challenge to receiving environments , due to their toxicity and low bioavailability for degradation . 
Microorganisms are capable of sensing and responding to their surroundings and this makes them ideal detectors for toxic compounds . 
This study investigates the global transcriptomic responses of Escherichia coli K-12 to selected VOCs at sub-toxic levels . 
Cells grown in the presence of VOCs were harvested during exponential growth , followed by whole transcriptome shotgun sequencing ( RNAseq ) . 
The analysis of the data revealed both shared and unique genetic responses compared to cells without exposure to VOCs . 
Results suggest that various functional gene categories , for example , those relating to Fe/S cluster biogenesis , oxidative stress responses and transport proteins , are responsive to selected VOCs in E. coli . 
The differential expression ( DE ) of genes was validated using GFP-promoter fusion assays . 
A variety of genes were differentially expressed even at non-inhibitory concentrations and when the cells are at their balanced-growth . 
Some of these genes belong to generic stress response and others could be specific to VOCs . 
Such candidate genes and their regulatory elements could be used as the basis for designing biosensors for selected VOCs . 
Volatile organic compounds ( VOCs ) are low molecular weight molecules with a vapor pressure of ≥ 10 Pa at 20 °C 1 , while compounds with a 6-months volatility between 5 and 95 % at ambient temperature can be termed semi VOCs ( sVOCs ) 2 . 
VOC such as toluene , methyl acetate , trichloroethylene , benzene , and phenol etc. , are common indoor and urban contaminants3 . 
Examples of common sVOCs include high molecular weight alkanes , polycyclic aromatic hydrocarbons ( PAH ) , organochlorine pesticides , and substitute benzenes4 ,5 . 
While there are natural VOCs ( e.g. cyclopentanone and dimethyl disulfide ) and sVOCs ( e.g. , n-Heptadecane and 1-butyl-3-methyl - imidazolium hexafluorophosphate ) produced biologically during degradation products of plant components or for biochemical signaling6 ,7 , many VOCs and sVOCs originate from fossil fuels , industrial chemicals and solvents . 
These compounds present a challenge to receiving environments and wastewater treatment processes , due to their toxicity and low bioavailability for degradation8 . 
The toxicity of VOCs and sVOCs has been evaluated in selected animal models . 
For example , cyclopentanone , N-methyl-2-pyrrolidone ( NMP ) and dimethylacetamide ( DMA ) , were found to cause developmental toxicity in rat embryos9 and rabbits10 ,11 . 
The toxicity of sVOCs commonly found in indoor environment , such as plasticizers , solvents , and flame retardant is also well studied12 . 
In industrial settings , VOC and sVOCs have been shown to concentrate in both liquid and gas phases of wastewater treatment plant13 ,14 . 
Thus , industrial VOC and sVOC discharges present serious concerns for wastewater treatment . 
Microorganisms are constantly sensing and responding to surrounding environmental conditions , including the presence of biologically toxic compounds . 
VOCs have been found to affect microbial diversity and biodeg-radation performance in activated sludge15 and in soil16 . 
Microbial tolerance to various VOCs in bacteria falls into three broad mechanisms : 1 ) alteration of membrane protein composition17 -- 19 , 2 ) export of toxic compounds through membrane transporters20 ,21 ; and 3 ) to a lesser extent , biotransformation of the compound to less toxic variant , which has been reported for soil microorganisms and a number of Pseudomonas species22 . 
Expression of 
1Singapore Centre for Environmental Life Sciences Engineering ( SCELSE ) . 
60 Nanyang Drive , SBS-01N-27 , Singapore 637551 . 
2School of Chemical and Biomedical Engineering , NanyangTechnological University , 62 Nanyang Drive , Singapore 637459 . 
3School of Biotechnology , Dublin City University , Collins Avenue , Dublin 9 , Ireland . 
4Asian School of the Environment , NanyangTechnological University , 50 Nanyang Avenue , N2-01C-45 , Singapore 639798 . 
Correspondence and requests for materials should be addressed to F.M.L. ( email : FLauro@ntu.edu.sg ) detoxifying enzymes such as reductive dehalogenases23 and oxygenases24 , have been exploited in the bioremediation of chlorinated aliphatic hydrocarbons and polycyclic aromatic contaminated soil and groundwater . 
The E. coli K-12 MG1655 strain used in this study is the primary experimental reference model with a highly curated genome sequence with annotation25 . 
It is widely considered the E. coli strain of choice and its genome was the first published sequence of a wild-type laboratory strain of E. coli K-12 because it has relatively few genetic modifications compared to most other E. coli strains . 
E. coli has also been used extensively as biosensor due to its ease of genetic manipulation and availability of information26 . 
We chose E. coli K-12 also because the GFP : fusion library is readily available27 . 
Various genetic mechanisms have been identified to contribute to VOC tolerance in E. coli . 
For example , membrane transport proteins like the acrAB-tolC pump28 , mannose transporter29 and phosphate transporters30 in E. coli were found to confer tolerance to various VOCs . 
Regulatory elements such as the FadR , MarR31 and purR regulon32 , were found to be involved in conferring tolerance to n-hexane , p-xylene and cyclohexane . 
Genes under the central metabolic processes , such as the cyo and nuo operons , responsible for energy conservation and production , and those under galactitol metabolic process ( gat genes ) were up-regulated in response to ethanol30 and butanol33 , respectively . 
Overexpression of heat shock proteins , such as the GrpE and GroESL chaperone system also resulted in increased tolerance of various forms of butanol as well as ethanol34 ,35 . 
In addition , studies has been conducted to look at tolerance of E. coli to butanol using genomic library screening36 , microarray , and at proteomic , regulatory network and metabolite levels33 ,37,38 . 
In most of these studies , genetic responses to sub-toxic VOC and sVOC concentrations have not been described . 
Studying gene activation/inactivation following exposures to sub-toxic levels will enable mechanisms of adaptation and enhanced tolerance to be decoupled from general stress responses , which would be expected at higher concentrations . 
In addition , information on the genetic responses of microorganisms to non-inhibitory levels would be relevant to understand and improve VOC and sVOC resistance in microorganisms that can be used for biocatalysis ( e.g. for the removal of VOCs and sVOCs ) applications . 
Such information would be preliminary to the development of rapid biosensing of VOC and sVOC in contaminated wastewater , offering protective measures for wastewater treatment plants and final users of reclaimed water39 . 
In this study , we used transcriptomics to investigate the global gene expression of E. coli K-12 grown in the presence of industrially relevant VOCs and sVOCs . 
All of the selected compounds are commonly used as solvents or produced as by-products during manufacturing of polymers , cleaners and industrial chemicals , with an exception of N-methylsuccinimide ( NMS ) , which is one of the metabolites commonly used as a biomarker for exposure of the solvent N-methyl-2-pyrrolidone ( NMP ) 9 . 
We aim to understand the specific and non-specific responses to the selected compounds . 
The focus in this study is to investigate genes that are responsive at non-growth inhibi-tory concentration , yet significant enough to induce a response at the transcriptome level . 
Results and Discussion
Growth and overall transcriptome profile of E. coli grown with VOCs . 
We analyzed the transcriptome of E. coli K-12 grown in the presence and absence of selected VOCs ( Supplementary Figure S1 ) using Illumina RNA-seq . 
Growth curve experiments were performed on E. coli with 0 ( as control ) , 0.02 , 0.1 and 0.5 % ( v/v ) of the selected VOCs to determine the highest non-inhibitory concentration to be used in RNAseq experiments ( Supplementary Methods and Figure S2 ) . 
The concentrations were established to be : 0.02 % for toluene ( T ) , 0.1 % for n-butanol ( B ) , N-cyclohexyl-pyrrolidone ( CHP ) , cyclopentanone ( CP ) , dimethyl sulfide ( DMS ) , N-methyl-2-pyrrolidone ( NMP ) ; 0.5 % for N , N -- Dimethylacetamide ( DMA ) and N-methyl succinimide ( NMS ) ( Table 1 ) . 
At these concentrations the cells reach optical density ( 600nm ) of 0.4 in approximately 5 -- 6h from initial O.D. of 0.02 in MOPS media ( Supplementary Figure S2 ) . 
There was a slight growth inhibition on DMS and CHP treatment during mid-log growth at concentration of 0.1 % . 
We have regarded this inhibition as non-significant and have chosen this concentration for subsequent RNA extraction . 
Previous work using E. coli to study the genes involved in tolerance ( using microarray/genomic library screening ) of selected VOCs uses a range of concentrations from 0.5 % 36 to 1.7 % butanol36 , and up to 10 % for toluene29 . 
The concentration of n-butanol that caused a 50 % growth decrease in M9 medium in E. coli DH1 was 0.8 % ( v/v ) 33 . 
Most of these studies used concentrations that are growth inhibitory to E. coli . 
We expect that the transcriptome of E. coli using non-inhibitory levels of compounds used in the current study would provide new insights compared to existing literature . 
In the present transcriptomic analysis , read mapping against the E. coli K-12 MG1655 genome was performed which allowed us to identify differentially expressed genes . 
The analysis identified the expression of 4140 coding DNA sequence ( CDS ) tags . 
The non-metric multidimensional scaling ( NMDS ) plot of global mRNA expression profiles revealed separate clustering patterns on cells grown with VOC compared to the no VOC controls , with NMS and NMP-treated cells clustering furthest from the controls on the first dimension ( Fig. 1 ) . 
Biological rep-licates for most VOC treatments clustered tightly indicating consistency between the replicates , although the clustering for treatment DMS , DMA , B are not as tight compared to the rest of the treatments . 
The differentially expressed ( DE ) genes identified ( with cut off at log fold change of greater than 1 or less than − 1 , an average logCPM value of greater than or equal to 3 , and a p-value less than 0.05 ) are distributed across a range of average logCPM values ( Supplementary Figure S3 ) . 
More DE genes were up rather than down regulated following treatment by B , DMA , DMS , and T . 
The converse was true for CHP , CP , NMP , and NMS-treated cells ( Supplementary Figure S3 , Table 2 ) . 
The percentages of genes identified as significantly differentiated over the total gene expression profile in VOC treatments compared to the controls ranged from 9.28 % ( DMA ) up to 25.94 % ( NMS ) ( Table 2 ) . 
Similar trend was found for chemical-specific gene responses ( identified based on Venn analysis of DE genes ) , with DMA having the lowest ( 1.30 % ) and NMS the highest percentage ( 24.21 % ) ( Table 2 , Fig. 2 ) . 
In addition , a total of 625 DE genes were shared by four or more VOC treatments , suggesting a subset of common genetic responses . 
The expression pattern for these DE genes appears to be divided into two major clusters for the VOCs used in this study ( Fig. 3 ) . 
Cells grown with B , T , DMS and DMA elicited more similar transcription patterns than CHP , CP , NMP and NMS . 
( Fig. 3 ) . 
These observations suggest that some VOCs might induce more cellular responses compared to others at non-growth inhibitory concentrations . 
Clustering of the overall transcriptome patterns of VOC treatments ( Fig. 1 ) had some resemblance compared to the shared DE gene profiles ( Fig. 3 ) . 
For example , the profile of treatment CHP and CP , NMP and NMS are clustering closer to each other compared to other treatment in both the MNDS and heatmap plots . 
The relationship between the chemical properties of the compounds tested and the degree of cellular response in E. coli would be an interesting investigation in the future . 
A number of genes relating to cold-shock responses were up regulated in our transcriptomic datasets ( Supplementary Table S1 ) . 
We have disregarded these genes as response to VOCs as the promoter clones for these genes failed to show an increase in GFP expression compared to the control in our bioassays at 37 °C ( Supplementary Figure S4 ) . 
These cold-shock related genes are likely to be an artifact of concentrating the bio-mass at 4 °C . 
Functional gene categories induced by multiple chemical treatments . 
Induction of iron-sulfur assembly system . 
Fe/S proteins participate in diverse biological processes such as respiration , central metabo-lism , DNA repair and gene regulation40 . 
The iron-sulfur cluster ( ISC ) and sulfur mobilization ( SUF ) systems carry out biogenesis and maturation of all Fe/S clusters in prokaryotes . 
In the ISC system , IscU and IscS are required to build the Fe/S cluster , followed by release of Fe/S cluster by HscA and HscB . 
In the SUF system , SufSE forms the Fe/S cluster , and SufBCD complex is responsible for cluster transfer and release40 . 
The compounds used in the current study had a higher expression of genes under different Fe/S cluster biogenesis system compared to the no chemical treatment control . 
Both ISC and SUF systems were activated following B and CP treatment , while only the ISC system is activated following CHP , NMP , NMS and T treatment , and only SUF system was activated when cells were grown with DMA and DMS ( Table 3 ) . 
This suggests that different chemicals induced distinctive responses in Fe/S assembly systems . 
IscR , a gene encoding the regulator responsible for Fe/S homeostasis and regulates the expression of a number of Fe/S proteins41 , was also up regulated in cells exposed to the eight compounds tested in the current study . 
The up regulation of iscR is validated through promoter : GFP fused expression assays ( Supplementary Figure S5 ) . 
IscR represses its own expression when there is suffifficient Fe/S cluster in the cell , and the isc operon is activated when cells are under Fe/S cluster-limiting and oxidative stress conditions . 
Overexpression of iscR might indicate that the chemicals tested in the current study elicited an oxidative stress or iron-limiting condition on the cells . 
This could be caused by the action of the VOCs on outer membrane proteins33 . 
In addition , the SUF system is believed to provide better resistance to iron40 ,42 and oxidative stresses compared to the ISC system43 -- 45 . 
Whether the induction of the SUF system when cells were exposed to DMA and DMS is directly linked to oxidative stress is unknown , as other regulators , like Fur , OxyR are also known to be involved in SUF-type Fe/S regulation40 . 
Oxidative stress responses . 
A number of genes known to be induced by oxidative agents were up regulated in response to at least 4 VOCs used in the current study ( Table 3 and Supplementary Figure S5 ) . 
PqiAB is a SoxRS-regulated membrane protein known to be induced by paraquat and other superoxide generators , but it is not induce by hydrogen peroxide , ethanol and heat shock46 . 
YhcN was identified as a stress protein associated with hydrogen peroxide , cadmium and acid47 . 
MntS confers resistance to hydrogen peroxide by facilitating delivery of Mn2 + to Mn2 + - dependent enzymes48 . 
A gene encoding for methionine sulfoxide reductase , msrB , was up regulated as well . 
MsrB repairs methionine residues in proteins that have been oxidized by reactive oxygen species49 . 
Collectively , the results indicate that E. coli cells exposed to the compounds tested in the study induce oxidative stress responses even at non-inhibitory concentrations . 
In addition , there might be proteins oxidized by the presence of VOCs . 
yfbU , a gene known to be involved in cell death by oxidative DNA damage50 , was down regulated in all treatments , suggesting that the cells did not go through the toxin : antitoxin response when grown with chemical tested , but instead employ alternative oxidative stress responses as described . 
Induction of various transporter proteins . 
Transporter proteins for inorganic ions , amino acids , and the PTS systems were among the top three categories to be differentially expressed in at least 4 chemical treatments compared to the control ( Fig. 4 and Supplementary Figure S5 ) . 
Genes involved in the uptake of both inorganic iron ( e.g. feoA , feoB and efeO ) , and siderophores ( exbBD , yncD and fhuF ) were up regulated . 
Genes involved in iron uptake have been shown to increase E. coli 's tolerance to environmental stresses . 
For example , over expression of feoA increases the tolerance of E. coli to butanol36 , and efeO confers resistance to mitomycin C and other stresses such as UV irradiation compared to wild type cells51 . 
ExbB and ExbD proteins are required to provide energy for the import of iron-siderophore complexes and vitamin B12 across the outer membrane via TonB52 -- 54 . 
YncD , a putative TonB-dependent outer membrane transporter for iron55 , could be one of the protein targets of TonB-ExbB-ExbD . 
The FhuF protein is required for cells to use hydroxamate-type siderophores as iron source56 . 
Collectively , up regulation of iron uptake genes implies that the cells are actively utilizing iron , possibly for the formation of Fe/S cluster containing proteins as described above . 
Transporters for other inorganic ions were also up regulated ( Fig. 4 ) , e.g. , genes for magnesium ( mgtA ) and manganese ( MntH ) uptake . 
MntH was to shown support the growth of E. coli cells encountering iron-deficiency and oxidative stress57 . 
During H2O2 stress , mutants lacking ability to import manganese and iron suffer high rates of protein oxidation , implying the role of MntH in preventing protein damage . 
Potassium efflfflux genes ( kefB and kefG ) were up regulated too . 
Efflfflux of potassium is known to play a role in protecting the cell from electrophile toxicity through acidification of the cytoplasm58 , suggesting cells grown with VOC might be undergoing electro-philic stress . 
The second largest transporter type relates to amino acids ( Fig. 4 ) . 
In particular , the dipeptide ABC transporter , encoded by the oppABCDF operon , was up regulated in most VOC treatments . 
The OppABCDF system function in oligopeptide uptake as well as recycling of cell wall peptides59 . 
Expression of opp genes was up regulated in cells treated with 1 % isobutanol as an early stage response38 , and oppD increased antibiotic resistance in E. coli during biofilm formation60 . 
Increased expression of the opp genes support previous findings that these transporters are involved in VOC resistance . 
The tnaCAB gene cluster , responsible for the uptake of tryptophan , was down regulated in response to most VOC used . 
Mutants lacking tnaCAB had increased isobutanol tolerance61 , supporting our finding that tnaCAB plays a negative role in VOC tolerance . 
The cytoplasmic putrescine transporter protein , encoded by PpotFGHI , was significantly up regulated following n-butanol , DMA , NMP and T treatment . 
The up regulation of potG stimulates cell growth in the presence of phenylpropanoids , which indicates that PotFGHI might also be involved in the import of this compound class62 . 
Cells grown with VOCs could either have an elevated concentration of putrescine inside the cell , or could also plays a role in transport of VOCs . 
The third most abundant transporter class containing DE genes identified belong to the phosphotransferase ( PTS ) system , which is an active transport system responsible for uptake of nutrients in bacteria ( Fig. 4 ) . 
The PTS system is activated when ambient nutrient level is low63 . 
In this study , most of the DE genes under the PTS systems were down regulated in most VOC treatments , including those responsible for glucose , dihydroxyacetone , fructose , galactitol , mannose and glucitol . 
Down regulation of these systems could be explained by the high nutrient media utilized in growing the cells ( 1.5 % glucose ) , hence the cells does not require active transport for nutrient uptake . 
Other transporter types with differential gene responses include multidrug efflfflux proteins and those related to osmotic response ( Fig. 4 ) . 
Three genes related to multidrug efflfflux proteins , mdtI , mdtJ and emrB , were up regulated in most chemical treatment used in the current study . 
MdtJ and I are two components of a spermidine exporter64 and emrB is known to increase tolerance to hydrophobic compounds , such as organomercurials and nalidixic acid65 and thiolactomycin66 . 
Multidrug exporters are capable of exporting compounds consisting different structural components , hence they could potentially export the compounds tested in the current study . 
Genes known to be associated with maintaining appropriate osmotic conditions in cells , for example , osmY , and ABC transporters for transport of osmoprotectants like proline , glycine betaine , and taurine ( proP , proV , proX and tauA ) were up regulated . 
The VOC used in the current study might have an effect in the osmotic condition in E. coli cells , hence inducing the expression of this gene class . 
In addition , the expression of a DNA-binding transcriptional repressor known to confer organic and inorganic acid stress , ydcI , was up regulated in all VOC treatment . 
YdcI protein is conserved across gram-negative bacteria and a S. typhimurium mutant lacking this gene had decreased resistance to acid stress67 . 
Up regulation of ydcI genes in our study imply that this gene may also be a response to VOC . 
Universal stress proteins . 
E. coli harbors six usp genes -- uspA , C , D , E , F and G. 68,69 . 
The functions of Usps overlap to some extent , e.g. both UspA and UspE are involved in oxidative stress defense68 , while UspG and UspF are associated with fimbriae-associated adhesion68 ,70 . 
From the transcriptomic results of the current study , we observed a down regulation of uspA and uspG in most VOC treatments , while uspE and uspF were up-regulated in B , DMA , DMS and T ( Table 3 , Supplementary Figure S5 ) . 
As UspA have functions that overlaps with UspE , down regulation of uspA can be compensated for by the expression in uspE . 
Similarly , down regulation of uspG expression can be compensated by up regulation of uspF . 
Flagella and cellular motility . 
Many genes relating to flagella biosynthesis and motility ( the flg , flh and fli genes ) were significantly down regulated in all VOC treatments , with the exception of treatment NMS ( Table S2 ) . 
Previous studies have found that flagellar biosynthesis was down regulated in E. coli exposed to ethanol30 as well as heat stress71 . 
Since NMS is not a VOC , it is not surprising that these genes were not repressed . 
However , a decrease in expression of flagella genes did not result in a reduction in motility in soft agar motility assays ( Supplementary Methods and Figure S6 ) . 
It is possible that the E. coli cells have already synthesized the flagellum before flagellar gene repression occurring in the assay . 
Other possible reasons include the differences in growth condition of E. coli due to the nature of the motility assay , e.g. surface-associated soft agar versus liquid , and the time of incubation . 
Functional gene categories induced by specific chemical treatments . 
Shared DE genes responsive to CHP and CP . 
A total of 96 genes responded significantly with specificity to CP and CHP , which shared the highest number of genes compared to other chemical treatment combinations ( Figs 2 and 5 ) . 
Top COG categories of the shared DE genes belong to Post-translational modification , protein turnover and chaperones ( O ) , Amino acid transport and metabolism ( E ) , Cell wall/membrane biogenesis ( M ) ( Fig. 5 ) . 
A few DE genes identified gave indications that CP and CHP might interfere with protein structure and outer membrane integrity . 
For example , the mlaD and mlaF genes , which prevent accumulation of phospholipids ( PLs ) in the outer leaflet of the outer membrane in E. coli cells , were up regulated . 
Cells accumulate PLs in the outer leaflet of the OM when exposed to harsh chemical treatments . 
This process would disrupt the LPS organization and increasing sensitivity to small toxic molecules72 . 
Up regulation of mla genes imply that the cells ' OM lipid asymmetry could be disrupted in the presence of the chemicals tested . 
In addition , a number of genes encoding for molecular chaperones were significantly up regulated in response to CHP and CP . 
These include the protein ( re ) - folding chaperones ( htpG , fkpA , dnaK-DnaJ-GrpE and the GroES ) , protein resolubilization chaperones ( clpB ) and a protease involving in clearing the defective peptides ( hslU ) . 
Up regulation of these genes imply that CHP and CP cause cellular protein misfolding in E. coli . 
Transporter-related genes specifically up regulated in response to CHP and CP include genes encoding for peptide transport proteins ( dtpD ) , and a putative drug efflfflux system protein ( mdtG ) . 
Overexpression of mdtG has been found to increase resistance to deoxycholate ( bile acid ) and the board spectrum antibiotic fosfomycin73 . 
Up regulation of such multidrug efflfflux genes could imply that cells perceive CP and CHP compounds as drugs and attempt to export them out of the cells . 
Shared DE genes responsive to NMP and NMS . 
The next chemical pair sharing the highest number of DE genes is NMP and NMS , sharing 68 genes based on Venn analysis ( Figs 2 and 6 ) . 
NMP is an organic compound consisting of a 5-membered lactam and NMS is a metabolite of NMP biodegradation9 . 
Although NMS is not considered as a VOC , it is cyclic . 
Most DE genes under energy production and conversion responding specifically to NMP and NMS were down regulated ( e.g. hyaDC , cbdAB and frdAD genes ) , except for rsxC , which is part of the rsx operon ( Fig. 6 ) . 
The rsxABCDGE gene cluster is involved in switching off the SoxR-mediated induction of SoxS transcription factor when cells are deficient of oxidizing agents74 . 
Up regulation of these genes could imply that the cells cultured with NMP and NMS were less prone to oxidative stress and require SoxR reduction to repress downstream activation of SoxS . 
Interestingly , rxsA was shown to be important for survival of cells exposed to ionizing radiation75 . 
Genes encoding for TolA-TolQ-TolR complex , were up regulated in cells treated with NMP and NMS ( Fig. 6 ) . 
The Tol-Pal cell envelope complex is known to be involved in maintaining cell envelope integrity , and mutants have greatly increase sensitivity to drugs and detergents and are prone to periplasmic leakage76 ,77 . 
Cells treated with NMP and NMS might respond differently to membrane disruption compared to that of CP and CHP . 
NMP and NMS activate the TolAQR complex whereas cells exposed to CP and CHP activate the Mla pathway . 
The molecular mechanisms behind activation of different gene clusters in response to maintenance of cell envelope integrity would be an interesting area for future investigations . 
Genes under `` Defense mechanisms '' that were up regulated include arnE and nudE which belong to the drug/metabolite transporter superfamily and the Nudix hydrolases family respectively ( Fig. 6 ) . 
Genes relating to iron-enterobactin transporter ( fepB and fepD ) and thiosulfate : cyanide ( glpE ) sulfurtransferase were up regulated specifically following NMP and NMS treatment . 
The fepBCDG complex together with the TonB-dependent outer-membrane transporter , and fepA , is responsible for the import of ferric enterobactin across the cell envelope . 
In addition to the iron-uptake system discussed in the previous sections , cells treated with NMP and NMS appear to have an additional iron-enterobactin transporter up regulated in the conditions tested in this study . 
Stress and membrane repair-related DE genes responsive to one chemical treatment . 
Genes that responded positively to one particular VOCs were identified , with a number of them related to stress ( oxyR , dinF , ydiY ) , transport pumps for metals ( nikC , rcnA and rcnB ) and transport pumps for drugs ( emrKY , mdtA , sbmA , yebQ ) ( Fig. 7 ) . 
Expression of emrK ( part of the EmrKY-TolC multidrug efflfflux transport system ) was found to increase in the presence of sub-inhibitory concentration of a number of antibiotics78 . 
As the concentration of the chemical used in this study is considered non-inhibitory , results supported the conclusion that low concentrations of compounds are suffifficient to induce a transcriptional response in various functional categories from the cell ( Supplementary Figures S7 ) . 
A number of genes relating to cell wall biogenesis were specifically up regulated when cells were exposed to NMP ( tonB , phoE , ldtB , wzzB , ugd ) . 
Induction of these genes could imply that the cells have membrane component biosynthetic pathways activated specifically when exposed to NMP , implying that the involvement of NMP damages cell wall component , thus requiring repair . 
COG category enrichment of DE genes . 
We performed COG enrichment analysis of total DE genes induced by individual chemical treatments against the E. coli 's genome copy of COG distribution ( Fig. 8 ) . 
More than half of the chemicals tested had amino acid related genes overrepresented compared to the E. coli genomic background . 
Amino acid metabolism is central to cellular survival and it is related to many parts of cellular metabolism . 
Genes under this category have been found to be differentially expressed in E. coli cells exposed to butanol33 ,36 . 
Cells exposed to n-butanol , DMS and toluene have a significant higher number of DE genes belonging to COG category of energy conversion , implying that genes under energy conversion are responsive to these VOCs . 
NMS is the only treatment that had genes relating to translation overrepresented . 
A total of 31 genes under translation and ribosomal biogenesis category were specifically responsive to NMS , many of them encode for ribosomal subunit proteins , implying that the cells were actively synthesizing proteins . 
Being a metabolic by-product of NMP , NMS is not a VOC , and appears that this metabolite does not impair cellular metabolism/growth at all . 
Both CHP and NMP had motility gene class overrepresented compared to E. coli 's background as well . 
COG categories were under represented compared to E. coli 's genomic background including those related to replication and repair for treatment NMP and T , and cell wall biogenesis were underrepresented in treatment n-butanol and toluene . 
Collectively these results could imply that different VOCs induced genes under specific COG categories . 
Catabolic pathways of VOCs and aromatic compounds . 
Little is known about the biodegradation of VOCs focused on in this study . 
The genome of E. coli K-12 contains neither the genes responsible for the degradation of DMS ( e.g. DMS monooxygenase , DMS dehydrogenase and DMS methyltransferase ) 79 , nor for toluene ( i.e. toluene-2 ,3 - dioxygenase ) 80 . 
Transcriptomic profiles of genes encoding for ring-hydroxylating oxygenases and transformation of aromatic compounds revealed very few differentially expressed genes in cells treated with the VOCs in the current study , with the exception of tnaA and entA , which were up-regulated following tolu-ene treatment , and ubiX , ubiB , which were up-regulated with CHP and NMS treatment ( Supplementary Table S3 ) . 
However , these genes are also involved in generic cellular metabolism and their direct involvement in the transformation of VOCs in this study is yet to be determined . 
A search for xenobiotics degradation pathways in KEGG ( according to which some pyrrolidones have been classified ) , revealed that most of the described xenobiotics in KEGG are structurally very different from the VOC used here . 
Hence to the best of our knowledge , this study is the first to describe transcriptomic responses of E. coli K-12 exposed to VOCs with pyrrolidone backbone . 
In conclusion , RNA-seq data in this study suggested that a variety of genes relating to Fe/S cluster biogenesis , oxidative and universal stress responses , as well as transport and membrane bound proteins are responsive to selected VOCs in E. coli . 
These genes were differentially expressed when the cells were in balanced-growth and at the highest non-inhibitory concentrations , which is well above the basal detectable environmental levels ( PUB , personal communications ) . 
By identifying the transcriptional responses occurring between the basal levels and high concentration spikes , we have set the framework for the analysis of the dose dependent response , a key element in biosensor development . 
The numerous changes in gene expression upon exposure to the different VOCs suggests that E. coli might exhibit analogous response when exposed to chemical compounds of similar nature . 
It is interesting to speculate that the clustering of DE genes in response to different VOC tested could be related to the overall physical properties ( polarity , volatility ) and to the structure of the VOCs ( i.e. linear chain vs cyclic compounds ) used in the current study . 
Further studies are necessary to uncover the specific molecular mechanisms of E. coli 's cellular responses to chemical compounds of different structures . 
In addition , a number of DE genes described in this study , for example , those related to Fe/S cluster biogenesis , and various transporter genes , are conserved in other environmentally relevant bacteria , such as Pseudomonas species20 . 
Results from the current study hence could also be applicable to future biosensor development in bacteria other than E. coli . 
However , one should note that some Pseudomonas species are known to be able to metabolize a number of VOCs and cyclic hydrocarbons via enzymatic conversions22 ,81 , hence their global genetic response to VOCs might be different from E. coli . 
Experimental Procedures
Chemicals . 
Chemicals used were as follows : n-butanol ( B ) , N-cyclohexyl-pyrrolidone ( CHP ) , Cyclopentanone ( CP ) , N , N -- Dimethylacetamide ( DMA ) , Dimethyl sulphide ( DMS ) , N-methyl-2-pyrrolidone ( NMP ) , and N-methyl-succinimide ( NMS ) and Toluene ( T ) . 
All were purchased from Sigma-Aldrich ( Taufkirchen , Germany ) and were of analytical purity . 
E. coli cultivation and RNA extraction . 
E. coli K-12 strain MG1655 was cultured in 10 mL LB5 broth within a shaking incubator at 150 rpm and at 37 °C for 16 h . 
The overnight culture was diluted ( 1:100 ) in 10 mL MOPS medium ( Neidhardt et al. 1974 ) supplemented with 1.5 % glucose . 
Based on the MIC assays ( Supplementary Methods ) , different VOC concentrations were added at the beginning of cultivation ( Table 1 ) and three biological replicates were used for each chemical treatment . 
Cells were grown in Balch-type tubes ( 18 × 150 mm ) with 20 mm butyl rubber stopper and aluminum seal to minimize leakage of VOCs during the cultivation time . 
Cells were incubated in a shaking incubator at 37 °C and were harvested for RNA extraction when OD600 reached 0.4 . 
The RNA extraction was as follows : 5 mL aliquots of the cultures were added to two volumes of RNAprotect Bacteria Reagent ( Qiagen ) . 
The mixture was incubated at room temperature for 5 min followed by centrifugation at 4,000 × g for 10 min at 4 °C . 
The supernatant was removed and the cell pellets were stored at − 80 °C until RNA extraction . 
RNA was extracted using the RNeasy ® Mini Kit ( Qiagen ) , following the manufacturer 's recommendations . 
Contaminating DNA was removed using DNsae ( Qiagen ) until DNA concentration was less than 5 % of the RNA . 
DNA and RNA concentrations were measured using Picogreen and Ribogreen assays ( Invitrogen ) , respectively . 
RNA sequencing . 
The quality of the RNA samples was determined by running the samples on a Bioanalyzer RNA 6000 Pico Chip ( Agilent ) . 
Next-generation sequencing library preparation was performed following Illumina 's TruSeq Stranded mRNA Sample Preparation protocol with the following modifications : RNA samples were added to the elute-fragment-prime step . 
The PCR amplification step , which selectively enriches for library fragments that have adapters ligated on both ends , was performed according to the manufacturer 's recommendation . 
Each library was uniquely tagged with one of Illumina 's TruSeq LT RNA barcodes to allow library pooling for sequencing . 
Library quantitation was performed using Invitrogen 's Picogreen assay and the average library size was determined by running the libraries on a Bioanalyzer DNA 1000 chip ( Agilent ) . 
Library concentrations were normalized to 2 nM and validated by qPCR on a ViiA-7 real-time thermocycler ( Applied Biosystems ) , using qPCR primers recommended in Illumina 's qPCR protocol , and Illumina 's PhiX control library as stand-ard . 
Libraries were then pooled and sequenced in one lane of an Illumina HiSeq2500 rapid sequencing run at a read-length of 101bp paired-end . 
Sequencing data have been submitted to GenBank SRA archive with the BioProject ID : PRJNA286974 and SRP accession SRP059483 . 
RNAseq data analysis . 
Quality trimming and adaptor removal were done using Cutadapt v1 .9.082 with the following parameters : -- q 20 , -- m 30 , -- overlap 10 , -- quality-base 33 . 
Sequences were mapped to the E. coli str . 
K-12 MG1655 genome ( NCBI accession : NC_00913 .3 ) by bowtie283 with end-to-end and very-sensitive modes . 
The alignments were converted to . 
bam and . 
bam-indexed files using Samtools84 . 
Sorted alignment files were imported into R to calculate overlapping reads as counts per gene using a combination of the following R packages : Rsamtools , GenomicFeatures and GenomicAlignments85 . 
Only the concordant pairs in the sorted * . 
bam files were imported using the function `` readGAlignmentPairsFromBam '' . 
Differential genes were identified from the tabular output of gene count abundance using edgeR package86 . 
The edgeR package implements a quantile-adjusted conditional maximum likelihood ( qCML ) estimator for the dispersion parameter of the negative binomial distribution86 ,87 . 
Testing for DE genes from biological replicates is based on the exact test derived based on these models . 
To calculate differentially expressed genes , all VOC treatments were compared to the control in which the cells did not have any exposure to VOCs . 
DE genes that have at least 2-fold change , p-value less than 0.05 and logCPM value greater than 3 were considered significantly different from the no VOC control . 
Genes were mapped to COG and KEGG IDs using the December 2014 release of COG database88 and June 2013 release of the KEGG database ( Kanehisa Laboratories ) . 
Principal component analysis ( PCoA ) , Venn and heatmap analysis were performed using R packages ( vegan , venn , heatmap .2 , respectively ) , and pathway maps were plotted using iPATH89 . 
GFP kinetics using fluorescent transcriptional reporter E. coli clones . 
Selected E. coli clones with transcriptional fusions of GFP to relevant promoters of the identified DE genes were used to validate the RNAseq results27 . 
Reporter strains were inoculated from frozen stocks into 2 × LB broth and incubated for 16 h at 37 °C . 
The cells were diluted ( 1:100 ) into fresh 1 × MOPS medium supplemented with 25 μg / mL kanamycin and 1.5 % glucose and grown as described previously . 
The VOC were added at the same concentration used in RNA experiments . 
When the OD600nm reaches 0.35 , an aliquot of culture was transferred to 96-well microplate . 
Optical densities were measured at OD595nm and GFP intensity were measured at 485/535nm at 15 min interval for 4 h. Triplicates were performed and cells grown without VOC were used for comparison . 
E. coli clone with the same vector backbone without any promoter was used as background noise subtraction . 
Fold-change analysis was performed and maximum fold-change was recorded . 
Acknowledgements
The authors would like to acknowledge financial support from PUB ( Award No : M4340001.C70 ) and Singapore Centre on Environmental Life Sciences Engineering ( SCELSE ) , whose research is supported by National Research Foundation Singapore , Ministry of Education , Nanyang Technological University and National University of Singapore , under its Research Centre of Excellence Programme . 
We also acknowledge Martin Tay for providing the revised mapping table of the COG 's 2014 database . 
This work is licensed under a Creative Commons Attribution 4.0 International License . 
The images or other third party material in this article are included in the article 's Creative Commons license , unless indicated otherwise in the credit line ; if the material is not included under the Creative Commons license , users will need to obtain permission from the license holder to reproduce the material . 
To view a copy of this license , visit http://creativecommons.org/licenses/by/4.0/