In vitro and in vivo characterization of DNA delivery using recombinant Lactococcus lactis expressing a mutated form of L. monocytogenes Internalin A

Background The use of food-grade Lactic Acid Bacteria (LAB) as DNA delivery vehicles represents an attractive strategy to deliver DNA vaccines at the mucosal surfaces as they are generally regarded as safe (GRAS). We previously showed that either native Lactococcus lactis (LL) or recombinant invasive LL expressing Fibronectin Binding Protein A of Staphylococcus aureus (LL-FnBPA+) or Internalin A of Listeria monocytogenes (LL-InlA+), were able to deliver and trigger DNA expression by epithelial cells, either in vitro or in vivo. InlA does not bind to its receptor, the murine E-cadherin, thus limiting the use of LL-InlA+ in in vivo murine models. Moreover, FnBPA binds to its receptors, integrins, via fibronectin introducing another limiting factor. In order to avoid the limitations of LL-InlA+ and LL-FnBPA+, a new L. lactis strain was engineered to produce a previously described mutated form of InlA (LL-mInlA+) allowing the binding of mInlA on murine E-cadherin. Results After showing the expression of mInLA at the surface of LL-mInlA+ strain, in vitro gentamycin survival assay in Caco-2 cells showed that LL-mInlA+ is 1000 times more invasive than LL. LL-mInlA+ invasivity was also validated by fluorescence microscopy. LL and LL-mInlA+ were transformed with pValacBLG, a plasmid containing the cDNA of bovine β-Lactoglobulin (BLG), resulting in strains LL-BLG and LL-mInlA+BLG. The plasmid transfer in vitro using LL-mInlA+BLG was increased 10 times compared to LL-BLG. Moreover, the number of mice producing BLG in isolated enterocytes after oral administration of LL-mInlA+BLG in vivo was slightly higher than after oral administration of LL-BLG. Conclusions We confirmed in this study that the production of mInlA at the surface of L. lactis is a promising strategy for plasmid transfer in vitro and in vivo.

Background DNA vaccination has gained a lot of attention since its ability to induce long-lasting humoral and cellular immune responses against an encoded antigen was discovered [1]. In addition, DNA vaccination poses no danger of integration into host cellular DNA thereby raising its safety profile [2][3][4]. DNA vaccines can be easily isolated to high purity, encode multiple antigens, and possess inherent adjuvant activity due to the presence of unmethylated CpG motifs that are recognized in mammals by TLR9 [5]. So called purified "Naked" DNA vaccination was shown to be highly efficient in rodents and mice, but not in larger animals and humans [6]. Consequently, it is very important to optimize DNA vaccine vectors and develop a delivery system to facilitate cellular uptake and enhance gene transfer efficiency and expression in situ [7].
Several strategies have been explored to protect plasmids from degradation, facilitating DNA uptake by phagocytic Antigen Presenting Cells (APCs) and thereby enhancing their immunological properties. This includes delivery technologies based on encapsulation into synthetic particles (cationic liposomes or polymers) or the use of viral vectors [7,8]. Despite their potential, some limitations and safety issues still remain which can restrict the application of gene therapy -e.g. the complexity of producing liposomes and their limited packaging capacity [9]. Additionally, it was shown that some viral vectors have the capacity to randomly integrate their genetic material into the host genome causing insertional mutagenesis of a cellular oncogene, leading to tumour formation [10].
The use of bacteria as delivery vehicles for DNA vaccination has emerged as an interesting alternative to overcome many of the problems associated with viral or liposomal delivery [11]. W. Schaffner was the first to observe genetic material transfer from bacteria to mammalian cells [12]. Since then, bacteria have been extensively exploited as vaccine delivery vehicles for vaccination against bacterial and viral pathogens as well as cancer immunotherapy [13][14][15]. The use of bacteria for mucosal delivery of DNA vaccines may be advantageous due to their potential to elicit secretory IgA responses as well as systemic immunity, when compared to conventional parenteral immunization [16]. Furthermore, bacterial carriers can increase and expand the magnitude of immune responses against the vector-encoded antigen due to the natural presence of Pathogen-Associated Molecular Patterns (PAMPs) that bind to Toll-like receptors (TLRs) and activate immune cells [5].
Presently, attenuated pathogens such as Salmonella, Shigella, Listeria, Yersinia, as well as, non-pathogenic Escherichia coli have been used as experimental live delivery systems [17,18]. An advantage of using attenuated pathogens as DNA vaccine vehicles is that they possess mechanisms to adhere or invade host cells with a negligible risk of reversion to a virulent strain via gene transfer or mutation. However, a potential concern is the risk of increased virulence in young or immunocompromised individuals.
The use of food-grade lactic acid bacteria (LAB) as DNA delivery vehicle represents an alternative and attractive strategy to deliver DNA vaccines at the mucosal surfaces (ref review by 19 and 20). The dietary group of LAB, including Lactococcus lactis and many species of Lactobacillus, is generally regarded as safe (GRAS) organisms of which some are intestinal commensals of humans. Indeed, it has been extensively demonstrated that these bacteria are able to deliver a range of vaccine and therapeutic molecules for applications in allergic, infectious or gastrointestinal diseases [19,21,22]. A relatively new development, however, is their use as a vehicle for genetic immunization [23]. Previous experiments performed by our group showed that either native L. lactis (LL) or recombinant invasive LL expressing Fibronectin Binding Protein A (LL-FnBPA+) of Staphylococcus aureus or Internalin A (InlA) of Listeria monocytogenes (LL-InlA+) [24,25], were able to deliver DNA in epithelial cells both in vitro and in vivo, demonstrating potential as gene transfer vehicles [24][25][26][27]. However InlA does not bind to its murine receptor, Ecadherin, thus limiting the use of LL-InlA+ in in vivo murine model. On the other hand, FnBPA requires an adequate local concentration of fibronectin to bind to its receptors, integrins [28,29].
In order to avoid the limitations of InlA and FnBPA and improve our knowledge on the key steps by which the DNA is transferred to mammalian cells using L. lactis, LL was engineered to express a mutated form of Internalin A (mInlA; Ser192Asn and Tyr369Ser) that increased binding affinity to murine and human Ecadherin [30,31] thus allowing for in vivo experiments in conventional mice. Herein, we describe the construction and characterization of this novel L. lactis strain as a DNA delivery vector, using cow's milk β-lactoglobulin (BLG) allergen, to measure DNA transfer to intestinal epithelial cells (IECs) in vitro and in vivo.
Overall, the production of mInLA+at the surface of Lactococcus lactis increased the invasisity of bacterium and amount of plasmid transfer by 1000 and 10 fold, respectively. In vivo, BLG production was detected in isolated enterocytes after oral administration of LL-mInlA +BLG and was slightly higher than oral administration of LL-BLG.

Results
Mutated internalin A is produced on the surface of recombinant L. lactis strain To investigate surface expression and production of mInlA, L. lactis NZ9000 and LL-mInlA+ strains were incubated with specific anti-mInlA monoclonal antibody and then with FITC-conjugated anti-Mouse IgG. Stained cells were analyzed by flow cytometry. As shown in Figure 1, LL-mInlA+ strain (blue peak) showed a significant shift in the fluorescence intensity comparing to the NZ9000 strain (black peak). No shift was observed when strains were incubated with FITC-labeled anti-Mouse IgG alone (data not shown). This experiment confirmed expression of mInlA on the surface of L. lactis.

L. lactis producing mInlA is efficiently internalized by Caco-2 cells
Non-confluent Caco-2 cells were incubated for 1 h with either NZ9000 or with LL-mInlA+. Non internalized bacteria were killed by gentamicin and intracellular bacteria enumerated after lysis of the eukaryotic cells. The LL-mInlA+ strain exhibited 1000-fold greater invasion rate than NZ9000 strain ( Figure 2).

LL-mInlA+ internalization analyzed by confocal microscopy
LL-mInlA+ and NZ9000 strains were labeled with CFSE dye and then incubated with Caco-2 cells for 1 h. Cells were fixed and confocal images were obtained. Very few cell-associated bacteria could be detected after coincubation with NZ9000 ( Figure 3A). In contrast, the LL-mInlA+ strain strongly bound to the membrane of cell clusters which is compatible with the known binding of InlA to E-cadherin, a cell-cell adhesion molecule. In addition, LL-mInlA+ was located intracellularly in some cells ( Figure 3C and B).
LL-mInlA+ can efficiently deliver in vitro a DNA vaccine containing β-lactoglobulin cDNA To test the ability of LL-mInlA+ to deliver a DNA vaccine plasmid in vitro to IECs, we transformed LL-mInlA + strain with pValac:BLG [32], a plasmid derived from pValac [23] containing the cDNA for BLG, under the control of an eukaryotic promoter to generate strain LL-mInlA+BLG (Table 1).
In order to monitor plasmid transfer and production of BLG in Caco-2 cells extracts, non-confluent Caco-2 cells were incubated with noninvasive L. lactis strains, LL and LL-BLG (see Table 1), or with LL-mInlA+BLG for three hours. After incubation with these bacteria, cell supernatant and proteins extracts from Caco-2 cells were tested for BLG expression using an EIA. BLG production was measured in Caco-2 cells protein extracts incubated with either LL-BLG or LL-mInlA+BLG. However, incubation with the LL-mInlA+BLG strain resulted in 10 fold higher levels of BLG compared to LL-BLG strain demonstrating that surface expression of mInlA enhanced intracellular delivery of the DNA vaccine DNA ( Figure 4A).
Secreted levels of BLG were increased 2 fold after coincubation with LL-mInlA+BLG compared to LL-BLG ( Figure 4B). These data shows that LL and LL-mInlA+, can mediate gene transfer of a DNA vaccine to Caco-2 cells in vitro and that invasiveness significantly increases the efficiency of DNA delivery.

DNA delivery efficiency in vivo is slightly improved by the production of mInlA
Mice were intragastrically administrated with LL, LL-BLG or LL-mInlA+BLG for three consecutive days, and the small intestine removed for isolation of IECs. BLG production was detected in protein extracts from IECs of mice administered with LL-BLG and LL-mInlA+BLG but not with control mice ( Figure 5). In both of the LL-BLG and LL-mInlA+BLG treated groups, some mice did not show production of BLG suggesting that DNA delivery may be a stochastic event depending on environmental factors. Even if this trend was not statistically significant, the number of mice producing BLG (in each of the three individual experiments) was systematically higher (11 mice) in the group administered with invasive bacteria than with noninvasive bacteria (8 mice producing BLG) suggesting that the LL-mInlA+strain is a Count Figure 1 Characterization of mInlA production at the surface of L. lactis. Black peak corresponds to the negative control, the wild type strain (LL) and the blue peak corresponds to L. lactis strain producing mInlA (LL-mInlA+). slightly better DNA delivery vehicle than non-invasive strain.

Discussion
There is a large body of research demonstrating that the use of L. lactis is able to elicit humoral and cellular immune responses to an antigen produced in rodents (for reviews see [19][20][21][22]).
Recently, we showed the ability of either native or recombinant invasive L. lactis as both in vitro and in vivo DNA delivery vehicle [24][25][26][27]. Recombinant invasive L. lactis strains were obtained by producing heterologous invasins which are proteins expressed at the surface of pathogens responsible for their invasivity. We first built lactococci expressing Internalin A (InlA) from Listeria monocytogenes (LL-InlA+) and showed that LL-InlA+ were able to 1) deliver a plasmid in vitro and 2) be invasive in vitro and in vivo in guinea pigs [24].
Nevertheless, the use of LL-InlA+ is restricted because InlA does not bind efficiently to its murine receptor, the E-cadherin [33]. Subsequently, we produced another invasin, the Fibronectin Binding Protein A (FnBPA) from Staphylococcus aureus and demonstrated that LL-FnBPA + were invasive and able to transfer a plasmid in vitro more efficiently than non-invasive L. lactis [25]. However, FnBPA requires an adequate local concentration of fibronectin in order to bind to its receptors, integrins [28,29], and this limitation could be a problem in vivo. So, in this study we produced a mutated Internalin A (mInlA) at the surface of L. lactis. The two mutations introduced were demonstrated to allow the binding of mInlA to murine E-cadherin thus permitting in vivo experiments with conventional mice [30,31].
We first checked that mInlA was expressed and properly directed to the surface of L. lactis. The shift of fluorescence peak obtained for LL-mInlA+ in FACS LL-mInlA+ internalization in Caco-2 cells analyzed by confocal microscopy. NZ9000 and L. lactis producing mutated internalin A (LL-mInlA+) were stained with CFSE dye (in green) and co-incubated with Caco-2 cells. Cell membranes were stained with DiI cell-labeling solution (in red) and the fluorescent samples were analyzed by confocal microscopy as described in the methods. 3A. Non-internalization of NZ9000 strain in Caco-2 cells. 3B. Intracellular localization of LL-mInlA+ in some cells. 3C. LL-mInlA+ bind to the membrane of cell clusters where mInlA receptor, E-cadherin, is exposed. analysis was significantly higher as compared to NZ9000 strain thus confirming successful surface expression of mInlA on L. lactis. Other invasins, from Gram-positive bacteria, such as InlA or FnBPA, have already been successfully expressed in L. lactis confirming that the signal peptide for secretion and the anchoring signal are well recognized by the L. lactis machinery. Production of invasins from Gram-negative bacteria, such as Yersinia pseudotuberculosis invasin at the surface of L. lactis has never been successful (Denis Mariat, personal communication). The invasivity was assessed by gentamicin assay in nondifferentiated E-cadherin expressing human epithelial cell line Caco-2 cells. This experiment showed that LL-mInlA +strain is 1000-fold more invasive than NZ9000 strain. Wollert and collaborators (2007) observed a 2-foldincrease in the adhesion and invasion efficiency of L.
monocytogenes strain producing mInlA compared to wildtype listeria expressing native InlA by using gentamicinprotection-invasion assays in Caco-2 cells [30]. A confocal image taken after gentamicin assay showed clearly that LL-mInlA+ is capable of adhering to and entering in nondifferentiated Caco-2 cells. The preferential distribution of recombinant bacteria at the periphery of the Caco-2 cell islets can be explained by the fact that E-cadherin is accessible only at the periphery. A similar type of bacterial distribution, around the Caco-2 cell islets, was previously observed when Caco-2 cells were co-incubated with LL-FnBPA+ [25].
LL-mInlA+ and LL strains were then transformed with pValac: BLG plasmid, co-incubated with Caco-2 cells and BLG expression was followed 72 h later by ELISA. BLG was detected in the cytoplasmic fraction of Caco-2 cells which were co-incubated with noninvasive and invasive strains carrying pValac: BLG. This data confirms prior observations that even noninvasive L. lactis can transfer functional plasmids to Caco-2 cells [23]. Cells were also capable of secreting the allergen, which is an interesting characteristic facilitating antigen uptake and presentation by professional APCs through crosspriming pathways [1]. The use of LL-mInlA+ improved BLG expression around ten times compared to noninvasive strain. Our hypothesis is that invasive lactococci can enter in higher numbers inside epithelial cells and thus deliver more plasmids.
Noninvasive and invasive L. lactis, carrying pValac: BLG or not, were orally administered for 3 consecutive days in BALB/c mice. On the fourth day, enterocytes from the small intestine were isolated and BLG production was measured by enzyme immunoassay (EIA). Isolated enterocytes from mice administered with invasive LL-mInlA+BLG produced the same amount of BLG as compared to mice immunized with noninvasive LL-BLG. Thus, we confirmed that noninvasive lactococci are able to transfer a functional plasmid in vivo in mice [27]. The OD405nm/mg total protein Figure 5 β-Lactoglobulin detection in mice isolated enterocytes after oral administration of noninvasive and invasive lactococci strains. Mice were orally administered 3 consecutive days with LL, LL-BLG or LL-mInlA+BLG. Seventy two hours after the last gavage, mice were sacrificed and BLG was assayed in protein extracts from isolated small intestine enterocytes. Results showed the sum of two independent experiments. use of LL-mInlA+BLG enhanced slightly the number of mice positive for plasmid transfer. Surprisingly, BLG production was not increased.
These results partly confirmed what we published recently with LL-FnBPA+BLG in vitro and in vivo [32]. Oral administration in mice of LL-FnBPA+BLG or LL-FnBPA+GFP elicited a GFP or BLG production in enterocytes. As with LL-mInlA+ the BLG production was not increased with LL-FnBPA+. However the number of mice producing BLG was significantly higher after oral administration of LL-FnBPA+BLG compared to non invasive LL-BLG. Considering these results it seems that LL-FnBPA+strain is a better DNA delivery vehicle than LL-mInLA+.
As no significant advantages were observed by using LL-mInlA+BLG compared to LL-BLG, we hypothesize that interactions of recombinant mInlA with their receptors were impeded in mouse intestinal epithelium. This lack of invasion in vivo was also observed by another group working with E. coli strain expressing invasin from Yersinia pseudotuberculosis as an oral vaccine for cancer immunotherapy. They showed that invasive E. coli was unable to enter gut epithelial cells due to a basolateral localization of the receptor, B1-integrin [34]. They demonstrated that invasive E. coli expressing Y. pseudotuberculosis invasin were selectively uptaken from the intestinal lumen into Peyer's patches using an ex vivo model. Similarly, E-cadherin, the mInlA receptor, is also expressed on the basolateral membrane of IECs which are strongly linked to each other in the gut making E-cadherin less available. It has been shown recently that L. monocytogenes could enter the epithelial membrane through extruding epithelial cells at the top of the villi but mainly through goblet cells which are located deeper in the crypt [35]. It is thus possible that LL-mInlA+BLG strain is not able to reach its receptor deeply buried in the crypt. The pathway whereby bacteria could penetrate gut epithelial monolayers could be through Microfold (M) cells in Peyer's patches. These cells are able to take up particles/bacteria from the lumen [36]. Nevertheless, we cannot exclude any possibility that lactococci can also interact with other cells from the epithelial membrane such as dendritic cells. Some subset of dendritic cells is now well known to produce dendrites, able to reach the lumen in order to sample its content [37].
The other hypothesis is that the plasmid would be released in the lumen by lysed lactococci and then captured by the enterocytes. It has been shown that lactococci do not persist in the gut and are very sensitive to its physico-chemical condition [38]. Most likely, plasmid transfer in vivo is a combination of both mechanisms, bacteria and released plasmid captures. Considering these data, the use of lactobacilli which persist longer in the gut than lactococci could be a better option for DNA delivery.

Conclusions
Mutated Internalin A protein was successfully expressed at the surface of L. lactis NZ9000, as demonstrated by FACS analysis. LL-mInlA+ strain was demonstrated to be 1000 times more invasive as compared to NZ9000 strain. This invasiveness capacity was confirmed by confocal microscopy experiments wherein LL-mInlA+ was found to be attached to Caco-2 cells and intracellularly located. Assays of BLG detection after BLG expression by eukaryotic cells revealed that the invasive status improved plasmid transfer in vitro. In vivo, the number of mice expressing BLG was higher (n = 11) in the group immunized with invasive bacteria than with noninvasive bacteria (n = 8). Even though this difference was not statistically significant, these study suggests that LL-mInlA + strain can be used as a DNA delivery vehicle for in vitro or in vivo experiments. The use of other LAB species which can persist longer in the gastrointestinal tract, such as lactobacilli, to mediate DNA transfer is currently being evaluated.

DNA manipulation and plasmids construction
Procedures for DNA manipulation were carried out as described by Sambrook et al. (1989) [39], with a few modifications. Plasmids were purified by the alkaline lysis method after bacterial incubation for 30 min at 37°C in TES solution (25% sucrose, 1 mM EDTA, 50 mM Tris-HCl pH 8) containing lysozyme (10 mg/ ml). The quality of the DNA, including its concentration and purity, was estimated by measuring the absorbance at 260 nm and 280 nm in spectrophotometer (Spectra-Fluor Plus, Tecan). Restriction and modification endonucleases were used according to recommendations of the suppliers. Details concerning the plasmids used in this study are found in Table 1.
Finally, pOri253:mInlA was transformed in E. coli DH5α and in L. lactis NZ9000 strain as described in the next section.

Bacterial strains, media and growth conditions
Bacterial strains are listed in Table 1. Briefly, L. lactis NZ9000 strain were grown in M17 medium containing 0.5% glucose (GM17) at 30°C without agitation and 10 μg/ml of erythromycin (Ery) or 5 μg/ml of chloramphenicol (Cm) were added, when required. Electroporation of L. lactis NZ9000 with pOri253:mInlA and/or pValac: BLG [32] plasmids was performed as described by Langella et al. (1993) [41]. Transformants were plated on GM17 agar plates containing Ery or Cm at the same concentration mentioned above and incubated at 30°C for two days before subsequent freezing or colony forming unit (CFU) counting. Positive clones were confirmed by colony PCR using specific oligos.

Mice handling
Specific pathogen-free BALB/c mice (females, 6 weeks of age; Janvier, France) were maintained under normal husbandry conditions in the animal facilities of the National Institute of Agricultural Research (UEAR, INRA, Jouy-en-Josas, France). All animal experiments began after allowing the animals 1 week for acclimation and were performed according to European Community rules of animal care and with authorization 78-149 of the French Veterinary Services.
Detection of mInlA expression by L. lactis using flow cytometry analysis L. lactis NZ9000 and recombinant L. lactis expressing mInlA were centrifuged (5000 rpm), washed with phosphate buffered saline (PBS) and then resuspended at a concentration of approximately 1x10 9 CFU/ml in 500 μl of PBS containing 0.5% of bovine serum albumin (BSA) and 10 μg/mL of monoclonal antibody anti-InlA kindly provided by Dr. Pascale Cossart (Cell Biology and Infection Department/Unité des Interactions Bactéries-Cellules, Pasteur Institute, Paris). After one hour incubation at 4°C, the bacteria were pelleted by centrifugation washed with PBS and then resuspended in 500 μl of PBS plus 0.5% of BSA containing fluorescein isothiocyanate (FITC)-conjugated AffiniPure Fab fragment Goat Anti-Mouse IgG (H+L) (Jackson Immuno Research). After 1 h incubation at 4°C, bacteria were washed once more with PBS and fixed in 2% paraformaldehyde for 30 min at 4°C. FITC labeled antibody binding to InlA was assessed by flow cytometry (Accuri C6 Flow Cytometer W ) using excitation at 494 nm and emission in the range of 510-530 nm (FL1-A channel). Data analysis was performed using CFlow Software (Accuri Cytometers, Inc.). The result was expressed as the average of three independent experiments performed in triplicate.

Invasion assay of bacteria into intestinal epithelial cells
The human intestinal epithelial cell line Caco-2 (ATCC number HTB37) derived from a colon carcinoma was used to measure invasion capacity of each strain. Caco-2 cells were cultured in RPMI medium containing 2 mM L-glutamine (BioWhittaker, Cambrex Bio Science, Verviers, Belgium) and 10% fetal calf serum in p-24 plates (Corning Glass Works) until they reached 70-80% confluence. In the assays on non-confluent Caco-2 cells, approximately 4x10 5 cells were present in each p-24 well. Bacterial strains were grown to an OD 600 of 0.9-1.0, pelleted and washed in PBS, then added to the Caco-2 cell cultures at a multiplicity of infection (MOI) of approximately 1000 bacteria per eukaryotic cell. The gentamicin survival assay was used to evaluate bacteria survival. In summary, recombinant or wild type L. lactis were applied in the apical side of eukaryotic cells and co-incubated during one hour at 37°C, in 5% CO 2 . After this period, cells were washed in order to remove bacteria in excess and then 150 μg/mL of gentamicin was added for 2 h to kill the extracellular bacteria. Cells were then lysed with 0.2% triton-X 100 diluted in water. Finally, serial dilutions of the cell lysate were plated for bacterial counting. CFU of intracellular bacteria were expressed as the average of three independent gentamicin assays performed in triplicate. Invasion rate was calculated as the ratio of CFU counts.

Confocal laser scanning
Bacteria were stained as described by Lee et al. (2004) [42]. Stationary phase culture of recombinant or wild type L. lactis, were washed twice in PBS and stained with 50 μM of green fluorescent dye carboxyfluorescein succinimidyl ester (CFSE) at 37°C for 20 min under constant shaking in the dark. CFSE labeled bacteria were used to perform the invasion assay as described above in non-differentiated Caco-2 cells grown on filter inserts. After 1 h of infection, cells were washed three times with PBS and fixed using 4% paraformaldehyde. Cell membranes were stained with 1 μM Vybrant W CM-DiI cell-labeling solution (Invitrogen) for 1 h at room temperature. Cells were mounted in Vectashield solution (Vector Labs, Burlingame, USA) to minimize photobleaching. Confocal images were obtained using a Zeiss LSM 510 system consisting of a Zeiss Axioskop with a Zeiss Plan Neofluar 63x NA 1.3 oil objectives. Stacks of images were reconstructed using Zeiss LSM software.

β-Lactoglobulin (BLG) expression by human intestinal epithelial cells after incubation with bacteria
In order to measure BLG expression and secretion by human epithelial cells the gentamicin survival assay was performed with Caco-2 cells as described above, however, bacteria and Caco-2 cells were incubated for three hours. After gentamicin treatment, plates were maintained for 72 h at 37°C, in 5% CO 2 . Supernatant was collected by centrifugation at 78.2 g (800 rpm) for 10 min and stored at -80°C. One mL of PBS supplemented with a cocktail of protease inhibitors (Roche) was then homogenized by sonication (3 times 10 s). Samples were kept at -80°C and used to measure BLG production using an Enzyme ImmunoAssay (EIA) described in the next section.
Enzyme immunometric assay (EIA) for quantification of bovine β lactoglobulin in human epithelial cells The method used for BLG quantification is described elsewhere [43]. In summary, 96 microtitre plates were coated with 3.5 μg/ml of anti-BLG monoclonal antibody, diluted in 50 mM phosphate buffer (PB) pH 7.4, and incubated overnight at room temperature. After washing, plates were blocked with EIA buffer (0.1 M PB pH 7.4; 1 g/1 L BSA; 0.15 M NaCl; 0.001 M Na 2 EDTA; 0.1 g/1 L sodium azide) and stored sealed at 4°C until use. Standard (recombinant BLG) and samples diluted in EIA buffer were added and kept at 4°C for 18 h. After this time, plates were extensively washed and then acetylcholinesterase conjugated monoclonal anti-BLG antibody (1 Ellman Unit/ml) was added for 18 h at 4°C. After washing, Ellman reagent was added and enzymatic reaction was measured at 405 nm in a spectrophotometer (SpectraFluor Plus, Tecan).

Oral administration of mice
Conventional BALB/c mice, 3 to 6 weeks of age were purchased from INRA animal care facilities (Jouy-en-Josas, France), acclimatized for 1 week before immunization under standard animal husbandry conditions in the animal facility (Unité d'Expérimentation Animale, Jouy-en-Josas, France). Mice (n = 8) were intragastrically administered with 1x10 9 (CFU) of strains, LL, LL-BLG or LLmInlA-BLG on 3 consecutive days using a gavage tube feeding. On the fourth day, the small intestine was collected for subsequent BLG quantification in isolated IECs.

Intestinal epithelial cells isolation
Mice were euthanized, and their small intestines were removed, rinsed with complete DMEM medium (containing 2 mM L-glutamine and 10% fetal calf serum). The length of intestine was opened and submerged in buffer A (in mM: 120 NaCl, 4.7 KCl, 2.4 KCl, 1.2 KH 2 PO 4 , 1.2 Na 2 HP0 4 , 25 NaHCO 3 , 10 HEPES, 5 EDTA, 0.5 DTT, 0.25% BSA; at pH 7.4 warmed to 37°C) for 20 min with agitation at 240 rpm [44]. Cells were collected by centrifugation (415.73 g -2000 rpmfor 5 min) at room temperature, washed once with PBS and lysed by sonication (3 times, 10 s). The cell lysate was centrifuged for 10 min at 3143.98 g (5500 rpm), then the supernatant was recovered and stored at -80°C. The EIA to detect BLG was performed as described above.

Statistical analyses
The results are expressed as mean ± standard error (SE) values. Statistical significance between the groups was calculated using the One Way ANOVA (and nonparametric) test, followed by the "Bonferroni" post-test. Values of p < 0.05 were considered significant.