Following the addition of 24 mM ferulic acid, the transcriptional data revealed an immediate stress response (summarized in Figure 1) in the L. brevis cultures. In all, the combined functional distribution of upregulated genes during both time points, in terms of their corresponding cluster of orthologous genes (COGs), demonstrated a marked departure from that of the reference samples. Those genes involved in carbohydrate transport and metabolism (COG G), transcriptional regulation (COG K), and amino acid transport and metabolism (COG E) appear to have undergone severe pertubations as L. brevis adjusts its transcriptional program to respond to the ferulic acid challenge. The functional distribution of differentially expressed genes is shown in Figure S1 and the expression levels of genes that appear most critical to the stress response are presented in Table 1; a complete list of significantly expressed genes is given in Table S1. Many up-regulated genes encode proteins involved in the citric acid cycle or sugar utilization, such as malate dehydrogenase, fumarase, catabolite control protein A (ccpA), or sugar transporters. The expression of a NADPH-quinone oxidoreductase may be an attempt to control excess levels ^[55] generated within an aerobic environment. The up-regulated heat shock protein (LVIS-0112), conserved in many lactic acid bacteria species ^[56], may participate in folding the heat labile phenolic acid decarboxylase ^[57]–^[60]; however, this chaperone is also expressed in response to n-butanol stress so it is likely a general stress response to protein mis-folding. The phenolic acid decarboxylase (LVIS-0213) of L. brevis is by far the most overexpressed gene, as expected given the role of this enzyme in degrading phenolic acids.

Many of the up-regulated genes detected encode hypothetical proteins with poor or no functional annotation, as is expected given the relative dearth of studies that have experimentally characterized L. brevis on a genetic level. Screening these L. brevis hypothetical ORFs using the trans-membrane hidden Markov model TMHMM ^[61] revealed that 28 of the 55 up-regulated hypothetical genes in either time point possess one or more transmembrane helices. These data suggest that the proteinaceous rather than lipid content of the membrane changes to counteract ferulic acid toxicity. These proteins may act as membrane stabilizers or ion transporters to counteract the membrane disruption triggered by ferulic acid; however, n-butanol exposure did not alter expression of these genes compared to the reference culture. These genes may therefore respond specifically to ferulic acid stress in particular or phenolic compound insults in general. Two uncharacterized major facilitator superfamily permeases (LVIS-1730 and LVIS-1917) with unknown substrate specificity were also overexpressed at different time points in response to ferulic acid stress; these exporters may be involved in the active export of ferulic acid or the decarboxylated product 4-vinylguaicol. These uncharacterized membrane proteins and transporters are promising targets for additional study to unravel the precise genetic basis of ferulic acid tolerance in L. brevis, especially given their unique occurrence in the ferulic acid stress profile.

In contrast to these shifts in energy metabolism, transcriptional regulation, and xenobiotic degradation, genes involved in membrane biogenesis (COG M) surprisingly comprise only 4% of the 195 genes, indicating that L. brevis may not appreciably alter the lipid composition of cytoplasmic membrane or the thickness of the outer peptidoglycan layer in response to ferulic acid. However, the consistent upregulation of -ketoacyl-(acyl-carrier-protein) reductase (fabG, LVIS-0378), a key enzyme in type II fatty acid synthesis ^[62], suggests that membrane lipid abundance may be altered as a possible defense mechanism against ferulic acid induced membrane fluidity. This hypothesis is supported by observed changes in the membrane of the closely related Lactobacillus plantarum when challenged with caffeic and ferulic acids ^[63]. Strangely, no other genes in the fatty acid synthesis pathway are upregulated so it is possible that the FA synthesis pathway in Lactobacillus is primarily regulated post-transcriptionally. Given the similarity between the organisms, membrane composition changes of L. brevis in response to ferulic acid are very likely to reflect those observed in the phenolic acid tolerant L. plantarum.

As expected, the highly reduced growth rate of L. brevis in the presence of ferulic acid is reflected in the functional roles of the down-regulated genes over the course of the challenge. An overview of the genes affected by ferulic acid stress is given in Table 2. Those genes encoding proteins responsible for amino acid and carbohydrate metabolism, transcription, and translation are strongly repressed, indicating that a metabolic shift from normal growth to a ferulic acid stress response has occured. Repression of several H antiporters may be due to a disruption of the cell's normal proton gradient ^[64], ^[65]. A total of 35 genes encoding proteins without known functional roles were down-regulated, but only 8 of these proteins appear to contain trans-membrane segments. On the whole, genes repressed during the ferulic acid challenge are similar to those seen during the n-butanol challenges (see below), implying growth and certain types of nutrient assimilation are inhibited within stressful environments. This result is also consistent with the repression of two key proteins involved in cell division (LVIS-0848, LVIS-1402) and observed growth defects for both stress conditions.

Given the likelihood that lignocellulosic biomass would be used for bio-butanol production, the transcriptional response of L. brevis to n-butanol insults was tracked over time in hopes of of revealing unique resistance mechanisms that could be combined with the insights gained from the ferulic acid challenges to produce a superior industrial strain. Unlike the gene expression pattern observed in the case of ferulic acid challenges, the addition of 1–2% n-butanol induced profound perturbations in lipid biosynthesis, protection against excessive oxidative stress via several means, and synthesis of protein chaperones. Given the similarity between the transcriptional profiles in response to 1% and 2% n-butanol, only the latter will be discussed here. A survey of the L. brevis n-butanol transcriptional response is provided in Tables 3 and 4, and the complete lists of up and down regulated genes are given in Tables S2 and S3. The functional distribution (in terms of COGs) over time is shown in Figure S2. Putative chaperones or chaperone components such as dnaJ (LVIS-1328), chaperone LVIS-0112, and ATP-binding subunits (LVIS-762, LVIS-1554, LVIS-1700) are significantly upregulated for the entire experimental time course. This pattern of chaperone was also similar to that observed during ferulic acid exposure, suggesting that these proteins are typically expressed as a result of environmental stress in general instead of being tied to a specific chemical stressor. There are several signs of oxidative stress as well; the continually increasing upregulation of two peptide methionine sulfoxide reductases (LVIS-0809 and LVIS-0810) which repair oxidative damage to methionine residues in conjunction with thioredoxin, NADPH-quinone oxidoreductase (LVIS-0076), and one thioredoxin (LVIS-1216) following n-butanol exposure. This response comports well with previous reports of n-butanol-induced oxidative stress in E. coli^[46]. Another component of the L. brevis oxidative stress response is the accumulation of manganese due to its lack of traditional superoxide dismutases (as determined by BLAST) ^[66], ^[67]. The high expression of the Mn /Zn ion transport system (LVIS-0451 and LVIS-0452) over the time series therefore provides additional evidence that n-butanol triggers an oxidative stress response.

Amino acid, carbohydrate, and ion transport and metabolism are significantly altered during the n-butanol stress response. Several unannotated amino acid transporters are weakly downregulated only in the first initial time point following n-butanol addition (LVIS-0619, LVIS-1789) while no change in the metabolism of amino acids known to function as osmoprotectants ^[68] is seen. Other metabolic functions involved in amino acid biosynthesis (LVIS-2023-LVIS-2027) are highly repressed as well. The expression of various enzymes responsible for the uptake and metabolism of carbohydrates (particularly pentoses), including phosphopentomutase (LVIS-1594), L-ribulokinase (LVIS-1742), and -galactosidase (LVIS-2259) continually increases throughout the experiment, perhaps as part of a concerted effort to support the energy demanding stress response. The strong downregulation of a Na /H antiporter (LVIS-2211) comports well with the observed effects of n-butanol on H gradients in other organisms ^[64], ^[65]. In addition to these metabolic changes, cell wall synthesis is also downregulated immediately following the n-butanol challenge, though this repression lessens over time. While these expression patterns are consistent with a reduced growth rate during an energy-consuming stress response, repression of other proteins possibly involved in competing phenotypes is also seen. The putative homologs of several proteins involved in acid tolerance in E. coli including glutamate decarboxylase (LVIS-2213) and glutamate acid resistance system gadC (LVIS-0078) ^[69], along with cyclopropane fatty acid synthase (LVIS-2047) ^[70] are downregulated for much of the time series. Other elements of the Lactobacillus acid stress response such as the arginine deiminase pathway (LVIS-2023, LVIS-2026, LVIS-2027) ^[71] are also strongly repressed. These responses are curiously similar to the observed antagonism between ethanol and acid tolerance phenotypes in E. coli^[72], suggesting that the same effect may exist between the n-butanol and acid tolerance phenotypes.

Along with these oxidative and general stress adaptations that are common to most organisms exposed to n-butanol, an upregulation of the entire fatty acid synthesis pathway is clearly evident from the transcriptional data. The FA synthesis pathway itself is summarized in Figure 2. Increased expression of the acetyl-CoA carboxylase subunits (accABCD) shunts additional acetyl-CoA to malonyl-CoA for dedicated use in the FAS pathway ^[62], ^[73], ^[74]. Subsequently, after the initial condensation step by fabH, fabFGZI homologs identified with BLAST in L. brevis (see Table 3) continue the fatty acid elongation process. Following termination, these fatty acids are generally converted into membrane phospholipids by the glycerolphosphate acyltransferase system ^[75]. The expression of nominal equivalents for the plsX (LVIS-0954) and plsC (LVIS-1355) do not significantly change over the course of the experiment so it is possible these enzymes are not a rate-limiting step in the membrane lipid synthesis, are regulated post-transcriptionally, or that the fatty acid flux is being diverted elsewhere in the cell. To test whether increasing flux through the FAS pathway affected n-butanol tolerance, the accABCD-overexpression strain developed by Davis et al. ^[74] was subjected to a n-butanol challenge. No significant change of maximum n-butanol tolerance or growth rate was observed with this strain, suggesting the increased production of fatty acids was insufficient or actually unable to confer a protective effect.

n-Butanol and other solvents alter the ratio of saturated and unsaturated membrane lipids in many organisms ^[44], ^[45], ^[76]. Altered gene expression within the fatty acid synthesis regulon may therefore be partially responsible for the n-butanol tolerance L. brevis. Direct assessment of the L. brevis membrane fatty acid composition following 75 min of 2% (v/v) n-butanol stress revealed only a significant 21.6% decrease in the abundance of 19∶1 cyclopropane fatty acid (19∶1-cfa) compared to unchallenged control cultures (Student's one tailed t-test, ). This fatty acid comprises only a small proportion of all fatty acids found in the L. brevis membrane during exponential growth, from 5.67% for the unchallenged controls versus 4.46% for the cultures subjected to n-butanol insult. Given the known role of cycloproprane fatty acids in conferring acid tolerance ^[70], ^[77], this result supports the inference that n-butanol and acid tolerance are opposing phenotypes, albeit for mechanistic reasons that are unclear. In order to test whether this hypothesis carries over to other organisms, the growth of a cfa gene knockout E. coli strain in 0–2% n-butanol was compared to the E. coli BW25113 parent strain. Unexpectedly, the growth rate and maximum n-butanol tolerance of the strains were very similar; this result could be explained by a lack of cyclopropane fatty acids in the E. coli membrane during exponential growth so that knocking out the cfa would have little effect on the actual membrane composition. A decrease in cyclopropane fatty acid content is also observed in several E. coli strains evolved for isobutanol tolerance ^[78]. Additional work is needed to unravel the role that cyclopropane fatty acids play in modulating n-butanol resistance and to further investigate the seemingly antagonistic relationship between n-butanol and acid tolerance.

Numerous studies have examined the transcriptional programs of C. acetobutylicum and E. coli in response to n-butanol challenge to better understand possible tolerance mechanisms. We chose to compare the n-butanol-induced transcriptional responses of L. brevis and E. coli for this purpose as E. coli is a widely used common bacterial host for a variety of applications in the biotechnology industry. The recent study by Rutherford and coworkers (2010) ^[46] identified several features of the n-butanol stress response in E. coli: increased expression of genes involved with oxidative phosphorylation (nuo, cyo, and sdh operons), an oxidative stress response involving sodA (superoxide dismutase) and yqhD (alcohol dehydrogenase), perturbed amino acid and carbohydrate transport, and an extracytoplasmic stress response as indicated by cpxP, degP, spy, and rpoE expression. Though the lack of known functions for many L. brevis genes hinders a direct comparison between these organisms, the expression of oxidative stress genes and the disturbances in L. brevis metabolism (in terms of transport and energy demands) agree well with those seen in E. coli. While L. brevis may also be under significant extracytoplasmic stress, the L. brevis rpoE-like sigma factor is not significantly upregulated at any time point, though a -like protein is upregulated during late 2% n-butanol stress. No other regulators are apparent from BLAST comparisons with E. coli genes; however, the dnaJ and LVIS-0112 chaperones along with several other protein chaperones are upregulated as expected. Strikingly, no statistically significant upregulation of the E. coli FAS genes was, in contrast to the constantly increasing expression of FAS genes in L. brevis following n-butanol induction.

This study was aimed at revealing possible components of the L. brevis stress response to ferulic acid and n-butanol, two distinct microbial inhibitors of great industrial importance. The results presented here combined with previous studies on the mechanisms of phenolic acid toxicity suggest that alterations in membrane structure and fluidity may play an important role in maintaining cell integrity and physiological electrochemical gradients in L. brevis when exposed to ferulic acid. The diverse range of membrane proteins expressed by L. brevis are likely involved in ameliorating abnormal ion flux across the membrane while enhancing rigidity, though molecular characterization of the gene products is needed for confirmation. Introduction of efflux pumps to expel ferulic acid and its decarboxylation product along with ion pumps to maintain intracellular K , H , and Na at appropriate concentrations in the presence of ferulic acid may be promising steps towards improving the phenolic acid tolerance of other organisms based upon these results.

Unlike its response to ferulic acid, L. brevis responds to n-butanol stress by countering oxidative damage, increasing carbon uptake, altering ion transport, and upregulation of the entire fatty acid synthesis operon. Direct analysis of the L. brevis membrane composition also revealed a significant decrease in the abundance of 19∶1 cycloproprane fatty acid. However, the n-butanol tolerance of an E. coli cfa knockout was not affected compared to wild-type, indicating additional elements of the stress response are required for tolerance. Both E. coli and C. acetobutylicum have the same generalized stress response but lack overexpression of the FAS genes, suggesting that increased fatty acid synthesis contributes to n-butanol tolerance in L. brevis. The upregulated fatty acid synthesis may act to restore membrane homoeostasis in opposition to n-butanol-induced fluidization. Biochemical characterization of how L. brevis regulates this FAS response and the ultimate intracellular source of the fatty acids is crucial to replicating this desirable phenotype in more industrially suitable organisms. The implied competition between acid and alcohol tolerance phenotypes seen in the transcriptional data and fatty acid composition changes suggests that evaluating the n-butanol tolerance of strains with deletions of genes that confer acid resistance may reveal unexpectedly resistant phenotypes.

L. brevis ATCC367 (American Type Culture Collection) and E. coli BW25113 (CGSC) were used in this study. L. brevis cultures were all grown at 30 C and 100 rpm unless otherwise noted. E. coli BW25113 cfa::kan was obtained from the Keio collection ^[79]. Overnight cultures used to initiate the time series inhibition experiments were grown in baffled 250 ml flasks with 25 ml MRS media (Difco) to 1.5–1.8. Three 500 ml baffled flasks with 125 ml MRS media were then inoculated with 3 ml of overnight each and grown to (mid-exponential phase; see Figure S3) for the ferulic acid and n-butanol challenges. Each 125 ml culture of L. brevis was challenged with 0.6 g ferulic acid (24 mM) or 1–2% (v/v) n-butanol. Subsequently, culture samples were rapidly harvested by filtration after 15, 75, and 135 minutes to track the immediate stress response and any long-term adaptations to either inhibitor. At each time point, samples were rapidly harvested by filtration using 0.22 m analytical filters (Nalgene) and placed immediately into 10 ml solution of RNAlater (Ambion) to preserve RNA integrity for later processing. Prior to the addition of ferulic acid, a pooled culture was created, harvested, and stored for use as a reference. Cell samples were stored at −80 C for subsequent analysis. Three biological replicates each were used in this study for the ferulic acid and 1% n-butanol (v/v) stressed cultures, and two biological replicates were used for 2% (v/v) n-butanol stressed cultures.

To evaluate the effect of upregulated fatty acid synthesis on E. coli n-butanol tolerance, HB101(DE3) ^[80], kindly provided by Zhilei Chen (Texas A&M University), was transformed with the pMSD8 plasmid containing the AccABCD synthetic operon (generously provided by J. Cronan) ^[74] via electroporation in a Gene Pulser XL (Bio Rad). The growth rate of the HB101(DE3)/pMSD8 strain was then evaluated in M9 minimal media with 0%–2% n-butanol using an TECAN Infinite M200 plate reader (TECAN) over a period of 24 hours.

Extraction of total RNA was performed with the ZR Fungal/Bacterial RNA MiniPrep kit (Zymo Research Corp) as follows: For each time point, 1.5 ml samples stored in RNALater were pelleted at 16000×g for 20 minutes and the supernatant was removed by aspiration. The bacterial pellet was then processed according to the manufacturer's protocol except that one volume of ice-cold ethanol was used to assist the RNA precipitation. DNAse I treatment was performed in-column as specified by the manufacturer. The resulting RNA was quantified using the Qubit fluorometer (Invitrogen). Gel electrophoresis was also used to confirm RNA quality. If necessary, samples were concentrated using ethanol precipitation and resuspended in 14 l of molecular biology grade water ^[81].

The SuperScript indirect cDNA labeling system (Invitrogen) was used to generate cDNA incorporating amino-allyl dUTP. Cy3 and Cy5 (GE Healthcare) or Alexa Fluor (Invitrogen) dyes were used to label the cDNA samples. Custom cDNA microarrays (Agilent) containing 15,209 probes (excluding positive and negative controls) were designed using the software package Picky ^[82] to maximize probe specificity and sensitivity under hybridization conditions with the following parameters. All probes are sixty base pair oligomers with 100% similarity to their corresponding target sequence. The minimum acceptable between the probe target and other sequences was set at 15 C, with a GC content range between 30–70%. The salt concentration in the hybridization media was set to 750 mM (personal communication with Agilent). The arrays contain at least three probes per ORF over 100 bp in length (2,157 ORFs total) and 1 probe for every 1050 bp of the L. brevis genome ^[56]. Probes for the 33 protein coding genes located on the L. brevis megaplasmids were not included in this study; the majority of these genes are related to plasmid maintenance, mobilization, or partitioning and are therefore not expected to impact the transcriptional data significantly. Labeled cDNA was hybridized onto the cDNA arrays for 18 hours, washed with Agilent Wash Buffers 1 and 2, and then immediately scanned using a GenePix 4200A reader according to the manufacturer's protocol as described for Agilent two-color prokaryotic microarrays.

The local background intensity was subtracted from the recorded signal from each array spot. Arrays were then subjected to LOWESS normalization individually using the MIDAS software package (TM4) ^[83]. The arithmetic average of the biological and technical replicate sample and reference signals were used for downstream analysis ^[46]. Differentially expressed (DE) genes were identified for each time point using the rank product method with a critical p-value ^[84]. The currently known functional annotations for L. brevis genes were obtained from the US Department of Energy Joint Genome Institute ^[2], and the number of differentially expressed hypothetical membrane proteins was identified by screening all differentially expressed hypothetical genes with the trans-membrane hidden Markov model TMHMM algorithm ^[61]. If required, BLAST ^[56] was utilized to determine L. brevis-E. coli homologous gene pairs. The statistical significance of the differentially expressed gene functional distributions were assessed using a hypergeometric distribution method ^[85]. The MeV (TM4) microarray analysis software was used for clustering and other expression profile analysis. The raw microarray data is available from the Gene Expression Omnibus (accession no. GSE24944).

L. brevis cultures were grown to for fatty acid content analysis. Control cultures were pelleted and the supernatant removed immediately, followed by storage at −80 C until processing. Experimental cultures were exposed to 2% n-butanol for 75 min and then stored in an analogous manner. Two biological replicates were used for each condition. Fatty acid analysis was later performed by Microbial ID Inc. to identify statistically significant changes in membrane composition following the n-butanol challenge.