Human Oral Isolate Lactobacillus fermentum AGR1487 Reduces Intestinal Barrier Integrity by Increasing the Turnover of Microtubules in Caco-2 Cells

Lactobacillus fermentum is found in fermented foods and thought to be harmless. In vivo and clinical studies indicate that some L. fermentum strains have beneficial properties, particularly for gastrointestinal health. However, L. fermentum AGR1487 decreases trans-epithelial electrical resistance (TEER), a measure of intestinal barrier integrity. The hypothesis was that L. fermentum AGR1487 decreases the expression of intestinal cell tight junction genes and proteins, thereby reducing barrier integrity. Transcriptomic and proteomic analyses of Caco-2 cells (model of human intestinal epithelial cells) treated with L. fermentum AGR1487 were used to obtain a global view of the effect of the bacterium on intestinal epithelial cells. Specific functional characteristics by which L. fermentum AGR1487 reduces intestinal barrier integrity were examined using confocal microscopy, cell cycle progression and adherence bioassays. The effects of TEER-enhancing L. fermentum AGR1485 were investigated for comparison. L. fermentum AGR1487 did not alter the expression of Caco-2 cell tight junction genes (compared to L. fermentum AGR1485) and tight junction proteins were not able to be detected. However, L. fermentum AGR1487 increased the expression levels of seven tubulin genes and the abundance of three microtubule-associated proteins, which have been linked to tight junction disassembly. Additionally, Caco-2 cells treated with L. fermentum AGR1487 did not have defined and uniform borders of zona occludens 2 around each cell, unlike control or AGR1485 treated cells. L. fermentum AGR1487 cells were required for the negative effect on barrier integrity (bacterial supernatant did not cause a decrease in TEER), suggesting that a physical interaction may be necessary. Increased adherence of L. fermentum AGR1487 to Caco-2 cells (compared to L. fermentum AGR1485) was likely to facilitate this cell-to-cell interaction. These findings illustrate that bacterial strains of the same species can cause contrasting host responses and suggest that food-safe status should be given to individual strains not species.


Introduction
It is thought that symbiotic bacteria of the human gastrointestinal tract may be harmful under certain conditions [1]. The term 'pathobionts' is used to describe microbes with pathogenic potential associated with chronic inflammatory conditions linked to environmental and/or genetic alterations [2]. Pathobionts differ from opportunistic pathogens (which often cause acute infections and are typically acquired from the environment) because they are harmless to a healthy host.
Lactobacillus fermentum is commonly found in fermented food products and is considered a ''generally recognised as safe'' (GRAS) organism by the US Food and Drug Administration (FDA). In vivo studies indicate that some L. fermentum strains have beneficial properties, particularly in relation to gastrointestinal health. For example, L. fermentum BR11 reduces colitis in rats [3][4][5], L. fermentum CECT5716 prevents and alleviates intestinal damage induced in mice [6], and L. fermentum ACA-DC 179 protects against colitis and Salmonella infection induced in mice [7]. In addition, clinical studies have also shown the benefits of some L. fermentum strains. In particular, L. fermentum CECT5716 enhances the effect of the influenza vaccination in healthy adults [8], reduces re-infection in women with infectious mastitis [9], and reduces the incidence of upper respiratory and gastrointestinal tract infections in six month old infants [10].
In contrast to these beneficial properties, our previous research indicated that some strains of L. fermentum may have a negative effect on intestinal barrier integrity. In that research, of the five L. fermentum human oral isolates studied, three strains reduced the trans-epithelial electrical resistance (TEER) across Caco-2 cell layers (indicating they had a negative effect on the integrity of the tight junctions between the adjacent epithelial cells), one L. fermentum strain increased TEER compared to control media (indicating that it enhanced barrier integrity) and the other did not alter TEER compared to control media [11]. The strain that caused the largest decrease in TEER compared to control media was L. fermentum AGR1487 and the strain that caused an increase in TEER compared to control media was L. fermentum AGR1485. These results suggested that some L. fermentum strains exhibit detrimental effects during targeted in vitro studies and that further research was warranted to understand the mechanisms involved.
Bacteria are known to alter intestinal barrier integrity using a variety of mechanisms [12][13][14]. For example, some enteropathogenic Escherichia coli (EPEC) strains reduce tight junction integrity by attaching to epithelial cells and causing cytoskeleton rearrangement [15]. In contrast, Lactobacillus species can enhance intestinal barrier integrity in many different ways including increasing the expression levels of genes and abundance of proteins involved in the formation of tight junctions that seal the space between adjacent epithelial cells [11,[16][17][18][19], increasing the transcription of the genes encoding the adherence junction E-Cadherin/b-Catenin protein complex [20], and activating p38 and ERK signalling [21].
The hypothesis of this research was that L. fermentum AGR1487 decreases the expression levels of intestinal epithelial cell tight junction genes and proteins, thereby reducing barrier integrity. Transcriptomic and proteomic analyses of Caco-2 cells (a model of human intestinal epithelial cells) treated with L. fermentum AGR1487 were used to obtain a global view of the effect of the bacterium on intestinal epithelial cells. Specific functional characteristics by which L. fermentum AGR1487 reduces intestinal barrier integrity were examined using confocal microscopy, cell cycle progression and adherence bioassays. The effects of TEERenhancing L. fermentum AGR1485 were also investigated for comparison.

Ethics statement
The two bacterial strains used in this research were isolated from saliva swabs of adult human volunteers as previously described [11]. Ethical approval from the New Zealand Health and Disability Committee was not required under Section 3 of the Standard Operating Procedures of the Health and Disability Committee due to the non-invasive nature of the sampling and the healthy status of the volunteers. Written consent for collection and use of the samples for research purposes was obtained from the volunteers. L. fermentum AGR1485 was isolated from a healthy individual and L. fermentum AGR1487 was isolated from a then healthy individual who was later diagnosed with inflammatory bowel disease (IBD).

Bacterial strains and mammalian cell culture
The bacteria were grown in MRS broth (de Man, Rogosa and Sharpe Broth; Difco) overnight in 5% CO 2 at 37uC. Caco-2 cell (ATCC HTB-37) stock cultures were grown in T75 flasks in M199 with 10% foetal bovine serum (GIBCO, Invitrogen Corporation), 1% non-essential amino acids (MEM non-essential amino acids 1006 solution; Sigma-Aldrich) and 1% penicillin-streptomycin (10000 units penicillin G sodium salt and 10000 mg streptomycin sulphate in 0.85% saline; GIBCO, Invitrogen Corporation) at 37uC in 5% CO 2 . The media was replaced every 3-4 days and the cells were subcultured weekly at a ratio of 1:3.

TEER assay
The TEER assays were carried out using confluent undifferentiated Caco-2 cells (5 days old; Passage number between 30 and 35) grown on 14 mm collagen membrane inserts (Cellagen Discs CD-24; MP Biomedicals) as previously described [11,22]. It was sufficient to use monolayers of this age because at this stage in development the tight junctions between the cells have already formed, and a previous study showed that the TEER response of differentiated versus undifferentiated Caco-2 monolayers was comparable [11,22]. The bacteria were prepared by growing overnight cultures (MRS broth, 37uC, 5% CO 2 ), collecting the bacterial cells by centrifugation (20,0006g for 5 minutes) and suspending the pellet in M199 with 1% non-essential amino acids to an optical density at 600 nm (OD 600 nm ) of 0.9, unless otherwise stated. The bacterial supernatants were prepared by incubating the bacteria at OD 600 nm of 0.9 in M199 with 1% non-essential amino acids overnight in 5% CO 2 at 37uC, then removing the bacterial cells by centrifugation (20,0006g for 5 min) and filter sterilisation (Millex GP 0.22 mm syringe driven filter; Millipore). Each treatment was tested in quadruplicate per assay and each assay was repeated three times. Treatments were compared in GenStat v11.1 using residual maximum likelihood (REML) analysis with an unstructured covariance model to take account of the repeated measures.

H-mannitol flux assay
Caco-2 monolayers were grown on collagen membrane inserts as described for the TEER assays. The cells were prepared the day before the assay by removing the media, washing three times with 0.1M phosphate buffered saline (PBS, pH 7.4) and adding M199 with 1% non-essential amino acids. The bacterial solutions were prepared as described for the TEER assay and 2.5 mCi/mL of 3 Hmannitol was added to each 250 mL treatment solution. Each treatment was tested in quadruplicate. Every two hours, 100 mL samples of the basal media were removed. The samples were mixed with 2 mL Starscint Scintillation Fluid and the 3 H-mannitol was measured using a Luminescence counter (PerkinElmer 1450 MicroBeta TriLux). The percentage of 3 H-mannitol that crossed the cell layer was calculated and plotted against time. Treatments were compared in GenStat v11.1 using REML analysis.

Gene and protein expression analysis
The aim of the gene and protein expression study was to investigate the expression of all biological pathways that may directly or indirectly effect barrier integrity (not just the tight junction genes and proteins); therefore, it was necessary for the cells to be differentiated into mature epithelial cells for the data to be relevant. Caco-2 cells were grown in T75 flasks for 18 days to allow them to differentiate in polarised epithelial cells. The media was replaced every 3-4 days. The bacterial solutions were prepared as described for the TEER assay. The media was removed from the Caco-2 monolayers and replaced with the treatment solution: 1) cell culture media (M199 with 1% nonessential amino acids); 2) L. fermentum AGR1485 suspended in cell culture media; and 3) L. fermentum AGR1487 suspended in cell culture media. There were eight replicates per treatment group. After 8 hours of exposure (37uC, 5% CO 2 ) the treatment solutions were removed and the monolayers were rinsed with PBS. The total RNA and protein were extracted from the samples using the TRIzol method, and gene expression and protein abundance were quantified using whole genome microarray and LC-MS/MS analyses respectively as described in Text S1. The microarray data was validated by real-time PCR using the method also described in Text S1. The ten genes chosen for validation had a range of expression patterns across the microarray analysis and are involved in a range of biological processes.

Cell cycle analysis
To assess the effect of the bacterial strains on Caco-2 cell cycle progression, the Caco-2 cells were grown in T75 flasks and treated with the bacterial strains as described for the whole genome and proteome analysis. DNA content analysis was based on the incorporation of propidium iodide into DNA and the resulting fluorescence a measure of the relative proportion of cells in the various stages of the cell cycle [23]. Six replicates of each treatment groups were assessed. Each sample was prepared and analysed using a FACSCalibur flow cytometer (Becton Dickinson, NZ) as previously described [24,25]. The fluorescence intensity of propidium iodide was collected at 585 nm, using CELLQuest Pro Software (Becton Dickinson) and analysed for DNA content using FlowJo (TreeStar V7.6.3). Treatments were compared in GenStat v11.1 using ANOVA.

Cell adherence
To assess bacterial adherence to epithelial cells, Caco-2 cells were grown in 24-well plates 37uC with 5% CO 2 for 18 days until differentiated. Caco-2 monolayers were prepared for the experiment by removing the media, washing three times with PBS and adding M199 with 1% non-essential amino acids (without foetal bovine serum and penicillin-streptomycin). L. fermentum AGR1485 and ARG1487 were grown overnight in MRS broth and approximately 10 7 (10 mL) were added to each well with each strain being assessed in triplicate. After 3 hours incubation at 37uC in 5% CO 2 , bacteria were removed from one 24-well tissue culture plate and the Caco-2 cells were washed gently five times with prewarmed PBS. Bacterial separation from Caco-2 cells was mediated by adding 1 mL of a 1% solution of Triton X-100 to each well and stirring using a stir bar and magnetic plate for 10 minutes. Bacteria were quantified by using 20 mL dilution spots on Luria Bertani (LB) agar plates. To assess adherence over 6 hours, nonadherent bacteria were removed after 3 hours and the epithelial cells were washed gently five times with PBS as described previously. Fresh media was then added to each well and the tissue culture plate incubated for a further 3 hours prior to enumeration. Treatments were compared in GenStat v11.1 using ANOVA.

L. fermentum AGR1487 had a negative effect on in vitro intestinal barrier integrity
The TEER assay was used to measure the effects of L. fermentum AGR1485 and AGR1487 on paracellular ion permeability, a measure of the integrity of tight junctions between Caco-2 cells. L. fermentum AGR1487 had a negative effect on tight junction integrity in all assays and reduced TEER by between 51% and 92% compared to control media after 10 hours (Figures 1a, 1b, 1d  and 1e). The effect of L. fermentum AGR1485 was inconsistent; it had a positive effect on TEER compared to control media in some assays (Figures 1a and 1e); in the other assays L. fermentum AGR1485 did not cause a change in TEER compared to control media (Figures 1b and 1c). When L. fermentum AGR1485 and L. fermentum AGR1487 were added together to Caco-2 cells, the negative effect of L. fermentum AGR1487 overcame the positive/ neutral effect of L. fermentum AGR1485 (Figure 1b); at a ratio of 1:1 the TEER decreased by 56% after 10 hours. The TEER-reducing effect of L. fermentum AGR1487 was dose dependent (Figure 1d). The negative effects of L. fermentum AGR1487 on TEER remained when half the number of bacteria cells was added, but one fifth and one tenth of the number of L. fermentum AGR1487 cells did not reduce TEER compared to control media. Although L. fermentum AGR1487 has a negative effect on TEER, L. fermentum AGR1487 supernatant caused a 34% increase in TEER compared to control media after 10 hours; whereas, L. fermentum AGR1485 supernatant did not cause a change in TEER compared to control media ( Figure 1e).
As a second measure of tight junction integrity, small molecule paracellular permeability was investigated by monitoring the flux of 3 H-mannitol across Caco-2 cell monolayers from the apical compartment to the basal compartment. L. fermentum AGR1485 did not alter the flux of mannitol across the cell layer compared to control media, but L. fermentum AGR1487 caused an increase in mannitol flux across the Caco-2 cell layers, indicating it increased permeability (Figure 1f). L. fermentum AGR1487 had a larger effect on Caco-2 cell gene expression levels than L. fermentum AGR1485 Whole genome expression analysis was used to determine the effect of L. fermentum AGR1485 and AGR1487 on Caco-2 cell gene expression. The raw microarray data have been deposited in the NCBI Gene Expression Omnibus under the GEO accession number GSE46352. As shown in the Venn diagram (Figure 2a), compared to control media, L. fermentum AGR1485 altered the expression levels of three genes (Table S1), whereas, L. fermentum AGR1487 altered the expression levels of 539 genes (Table S2). Partial least squares discriminant analysis (PLS-DA) showed that the three treatment groups could be discriminated based on their gene expression profiles (Figure 2b).
Genes that were differentially expressed in Caco-2 cells treated with L. fermentum AGR1487 compared to L. fermentum AGR1485 were significantly over-represented in a number of KEGG pathways, the top 5 of which were Pathogenic Escherichia coli infection, Gap junction, Glycosaminoglycan biosynthesis -chondroitin sulphate, MAPK signaling pathway and Phagosome (Table 1). Hierarchical clustering of average expression levels of genes within these significantly overrepresented KEGG pathways showed that the Caco-2 cells exposed to L. fermentum AGR1485 were more similar to that of the control Caco-2 cells than those exposed to L. fermentum AGR1487 (Figure 2c).
To validate the microarray data, the expression levels of ten genes were also quantified using real-time PCR (qPCR) and the results were compared with the gene expression data obtained using microarray analysis (Table S3). For the main comparison of interest, Caco-2 cells treated with L. fermentum AGR1487 versus untreated Caco-2 cells, there was a significant correlation between the qPCR and microarray results (Pearson coefficient = 0.68; P = 0.04), supporting the validity of the microarray data.
L. fermentum AGR1487 did not alter the expression levels of Caco-2 cell genes in the tight junction pathway but altered other tight junction related genes Analysis of the KEGG pathways showed differentially expressed genes between the Caco-2 cells treated with L. fermentum AGR1485 compared to L. fermentum AGR1487 were not over-represented in the Tight Junction pathway (P = 0.57). However, 7 tubulin genes   (TUBA4A, TUBA1B, TUBB3, TUBB4, TUBA8, TUBA3D, TUBB8) had increased expression levels in response to L. fermentum AGR1487 compared to L. fermentum AGR1485 (fold-change 1.6 to 2.8; FDR,0.05). These tubulin genes are present in the Gap Junction and Pathogenic Escherichia coli Infection KEGG pathways (Table 1). Correspondence analysis (CA) of expression levels of genes within the Gap Junction pathway showed that expression profiles of Caco-2 cells treated with L. fermentum AGR1487 were clearly different from expression profiles of untreated cells or those treated with L. fermentum AGR1485 (Figure 3). Projection of Gap Junction pathway genes showed that the expression pattern of tubulin genes was primarily what differentiated the L. fermentum AGR1487 treated samples from those untreated or treated with L. fermentum AGR1485 (Figure 3). L. fermentum AGR1487 had less effect on Caco-2 cell protein abundance than L. fermentum AGR1485 Caco-2 cell whole protein digests were isobarically labelled and analysed using LC-MS/MS to determine the effect of L. fermentum AGR1485 and AGR1487 on protein abundance profiles. Of the 63 proteins detected (abundance significantly higher than background), 30 were differentially expressed in at least one comparison (Table 2). Compared to control media, L. fermentum AGR1485 altered the abundance of 30 proteins, whereas L. fermentum AGR1487 altered the abundance of only six proteins ( Figure 4a). As with the gene expression data, the PLS-DA plot showed that the three treatment groups could also be discriminated based on their protein expression profiles (Figure 4b).
Procrustes rotation analysis showed there was a significant correlation (correlation = 0.5; P = 0.0036) between protein and gene expression profiles (Figure 4c).
L. fermentum AGR1487 increased the expression levels of microtubule associated proteins and altered the localisation of zona occludens 2 in Caco-2 cells Three tubulin proteins were detected in the analysis (tubulin beta-5 chain, tubulin beta and tubulin beta-4), but only one of these had altered abundance in one comparison: the tubulin beta- 5 chain protein had reduced abundance in response to L. fermentum AGR1485 compared to control cells (fold change = 1.59). However, compared to Caco-2 cells treated with L. fermentum AGR1485, Caco-2 cells treated with L. fermentum AGR1487 had increased abundance of the tropomyosin 1 (alpha) and tropomyosin 2 (beta) proteins, which are involved in the formation of microtubule bundles, and the plectin protein, which acts as a link between the microtubules and the actin cytoskeleton.
No tight junction-related proteins were detected in the analysis so the effect of L. fermentum AGR1485 and L. fermentum AGR1487 on the Caco-2 cell abundance of these proteins could not be determined. Therefore, confocal microscopy was used to investigate the localisation of the tight junction proteins in Caco-2 cells when treated with the bacteria. Visualisation of the abundance and localisation of ZO-2 showed that Caco-2 cells treated with L. fermentum AGR1487 did not have defined and uniform borders of ZO-2 around each cell, unlike control or AGR1485 treated cells. ZO-2 in Caco-2 cells treated with L. fermentum AGR1487 was localised in clusters with an uneven distribution (Figure 5a). L. fermentum AGR1487 did not alter Caco-2 cell cycle progression L. fermentum AGR1487 caused a large decrease in TEER and altered the visual appearance of Caco-2 cells, so DNA content analysis was used to assess whether the bacterium affected Caco-2 cell cycle progression. There were no differences in the number of Caco-2 cells in any cell cycle phase between those treated with L. fermentum AGR1485 and AGR1487 (Figure 5b).
L. fermentum AGR1487 supernatant did not cause a decrease in TEER, unlike what was observed when Caco-2 cells were treated with the live bacterium, so it was hypothesised that physical interaction between the bacterium and Caco-2 cells was required for the negative effect. The ability for L. fermentum AGR1485 and AGR1487 to adhere to Caco-2 cells was determined. Compared to L. fermentum AGR1485, the number of L. fermentum AGR1487 cells adhered to Caco-2 cells after 3 and 6 hours was 9-fold and 15-fold higher, respectively (Figure 5c).

Discussion
The results showed that, despite being from a species considered harmless, human oral isolate L. fermentum AGR1487 increased the expression levels of genes and abundance of proteins that have been implicated in reducing the integrity of the tight junctions between intestinal epithelial cells. For example, compared to control Caco-2 cells and those treated with L. fermentum AGR1485, Caco-2 cells treated with L. fermentum AGR1487 had both a higher expression level of genes encoding for tubulins and a higher abundance of microtubule-associated proteins. A high turnover in tubulin synthesis has been linked to the disassembly of tight junctions [26], which indicates that these genes and proteins may play a role in the negative effects on barrier integrity observed for L. fermentum AGR1487. Although seven tubulin genes had increased expression in response to L. fermentum AGR1487, an increase in the abundance of the corresponding tubulin proteins was not detected. This is not surprising given that the transcriptomic analysis was able detect significant expression (higher than background) for greater than 20,000 genes, whereas the proteomic analysis was able to detect significant protein abundance for only 63 proteins due to technical limitations and the sensitivity of the method. However, changes in the abundance of other microtubule-associated proteins in response to L. fermentum AGR1487 indicates that observed changes in tubulin gene expression likely resulted in changes to microtubule synthesis.
Increased expression of tubulin genes is also a mechanism used by EPEC that form attaching-and-effacing (A/E) lesions characterised by cytoskeletal alterations, disruption of the brush border cytoskeleton and proliferation of filamentous actin beneath adherent bacteria. EPEC induced A/E lesions also lead to morphological alteration of tight junctions [15,27]. Our results show that L. fermentum AGR1487 cells were required for the negative effect on barrier integrity (bacterial supernatant did not cause a decrease in TEER), suggesting that the physical interaction may be necessary for this effect. Increased adherence of L. fermentum AGR1487 to Caco-2 cells (compared to L. fermentum AGR1485) is likely to facilitate this cell-to-cell interaction. In contrast to L. fermentum AGR1487, Lactobacillus plantarum MB452 has been previously shown to reduce Caco-2 cell expression levels of tubulin genes [22]. L. plantarum MB452 is a probiotic bacterium that increases TEER, and its reduction of tubulin turnover is thought to contribute at least in part to this effect by reducing tight junction disassembly. This increase in tight junction integrity is considered beneficial because it strengthens the epithelial barrier which is important for protecting the host from entry of unwanted antigens and pathogens. It appears that different lactobacilli strains can use this mechanism of modifying tubulin gene expression levels to alter intestinal barrier integrity in either a positive (L. plantarum MB452) or negative (L. fermentum AGR1487) way.
The gene expression results indicate that L. fermentum AGR1487 may also decrease tight junction integrity by influencing mitogenactivated protein kinase (MAPK) signalling, however further work is require to confirm this. Caco-2 cells treated with L. fermentum AGR1487 had higher expression levels of genes involved in the KEGG MAPK signaling pathway, which is known to lead to tight junction disruption [28]. In addition, MAPK is known to directly associate with the microtubule cytoskeleton and there are correlations between MAPK activities and cytoskeletal alterations [29].
As well as indicating possible mechanisms for the negative effect of L. fermentum AGR1487 on intestinal barrier integrity, the data also uncovered potential anti-tumour properties of L. fermentum AGR1485. Protein expression analysis showed that Caco-2 cells treated with L. fermentum AGR1485 had decreased abundance of numerous proteins (histones and others) involved in tumourigenesis. Anti-tumour effects have also been reported with another L. fermentum strain; L. fermentum RM28 restricted the proliferation of colon cancer cells by 22-29% [30]. Although changes in protein abundance were observed in our study, analysis of cell cycle progression showed no differences in the number of Caco-2 cells in any cell cycle phase between those treated with L. fermentum AGR1485 and AGR1487. However, this may be due to the time frame of the treatment (8 hours), which may have been too short to observe differences in cell phases between treatment groups, given that the doubling time of Caco-2 cells is approximately 36 hours [31]. Further research is needed to confirm this potential beneficial anti-tumour effect of L. fermentum AGR1485.
The detrimental properties of L. fermentum AGR1487 are specific to this particular strain and not typical of all L. fermentum strains; L. fermentum AGR1485 did not have the same negative effects. L. fermentum AGR1487 was isolated from the mouth of a person who was later diagnosed with IBD. IBD is likely due to an unrestrained immune response to commensal bacteria and has been linked to impaired intestinal barrier integrity [32]. Lactobacilli found in human faeces can originate from the mouth [33,34], indicating that this oral isolate may reside in the intestines of the individual from whom it was isolated. Therefore, it is possible that this L. fermentum strain may play a role in disease progression in this particular genetically-susceptible individual. However, based on the current data it cannot be determined whether L. fermentum AGR1487 is pathobiontic (detrimental to susceptible individuals only), pathogenic (detrimental to healthy individuals) or symbiotic (not detrimental in situ), so further research is needed before conclusions can be drawn about the effect of this strain on human hosts.
In conclusion, this research showed that L. fermentum AGR1487 may negatively impact intestinal barrier integrity by increasing the turnover of microtubule proteins triggering tight junction disassembly. Current studies are investigating the effect of L. fermentum AGR1487 in mono-associated rats to determine whether the negative effects observed in vitro correspond with in vivo phenotypic differences. Additionally, the genome sequences of the two L. fermentum strains (AGR1485 and AGR1487) are being analysed for differences that may be responsible for the contrasting effects. These findings illustrate that contrasting host cell responses occur in response to bacterial strains of the same species. Given that L. fermentum is commonly found in food products, these results imply that food-safe status should be given to individual strains not species and that genome sequence information may be required to confirm strain associated genotypes.

Supporting Information
Table S1 Genes that were differentially expressed between Caco-2 cells treated control media and those treated with L. fermentum AGR1485 for 8 hours.

(XLSX)
Table S2 Genes that were differentially expressed between Caco-2 cells treated control media and those treated with L. fermentum AGR1487 for 8 hours.
(XLSX) Table S3 Comparison between microarray and qPCR analysis of the gene expression of Caco-2 cells untreated or treated with L. fermentum AGR1485 and L. fermentum AGR1487 for 8 hours.