The Role of Macrophage Migration Inhibitory Factor in the Function of Intestinal Barrier

Macrophage migration inhibitory factor (MIF) is a multifunctional protein that is involved in the development of gut-related inflammation. To investigate the role of MIF in the function of the intestinal barrier, we have explored intestinal permeability and gut-associated immune response in MIF-deficient (MIF-KO) mice. The absence of MIF provoked impairment of tight and adherens epithelial junctions in the colon through the disturbance of E-cadherin, zonula occludens-1, occludin and claudin-2 expression, which lead to the increase of intestinal barrier permeability. In these circumstances the diversity and content of gut microbiota in MIF-KO mice was considerably different compared to wild type mice. This change in microbiota was accompanied by an increased intestinal IgA concentration and a higher production of pro-inflammatory cytokines TNF and IFN-γ in mesenteric lymph nodes of MIF-KO mice. The forced changes of microbiota executed by antibiotics prevented the “leakage” of the barrier in MIF-KO mice, probably through up-regulation of occludin expression and normalization of cellular pore diameters. In addition, cytokine secretion was normalized after the treatment with antibiotics. These results suggest that MIF participates in the maintenance of physiological microbiota diversity and immunosurveillance, which in turn enables the proper intestinal barrier function.

Further experiments confirmed the microscopy data, since MIF-KO mice had a higher concentration of FITCdextran in the serum after per os administration (Fig. 1i). These results suggest that MIF absence provoked an increase in intestinal permeability.
Although epithelial connections were disturbed in the colon of MIF-KO mice, there were no differences in any measured elements of intestinal mucosa (thickness of the layers, height of the epithelial layer, crypt morphology, number of cells in the lamina propria) between MIF-KO and WT animals (Supplemental data and Figure S1, Table S1).
Another proof of the "leaky" gut in MIF-KO mice is the activated state of macrophages within the peritoneum 11 . It is suggested that if the barrier is permeable, bacteria from the gut can migrate to distant sites in the organism and activate resident macrophages, as it is in the case of the peritoneum. Although the number of isolated peritoneal cells was the same in both strains (7.9 ± 3.8 × 10 6 vs 6.3 ± 2.9 × 10 6 , WT vs MIF-KO, respectively), macrophages from MIF-KO mice produced higher levels of nitric oxide (Fig. 1j), equal TNF levels ( Fig. 1k), higher levels of IL-1β (Fig. 1l), and lower levels of IL-10 ( Fig. 1m), suggesting that they were exposed to the bacterial species that might have migrated from the gut. Our results are in contrast with previous data obtained in dextran sodium sulfate (DSS)-treated MIF-KO mice. In this colitis model, the absence of MIF prevents the development of colon histopathological changes induced by DSS 8 . However, the permeability was not monitored in these mice and they were bred on a different genetic background.
The effect of MIF absence on the intestinal epithelial cells. Although MIF seems to be of the outmost importance for the survival of colon epithelial cells 12 , there is no data on the possible role of MIF in the maintenance of the cell-cell junctions. Since the "leakage" of the gut is often due to the irregular infrastructure of epithelial cells junctions 13 , the content of E-cadherin and expression of occludin, pore-forming and junction-forming claudins and zonula occludens-1 (ZO-1) in the colon epithelial cells was examined. E-cadherin is responsible for the proper architecture of adherens junctions 14 and its considerably lower content detected by western blot and immunohistochemically in MIF-KO epithelial cells (Fig. 2a) coincided with already observed wider adherens junctions. Formation of tight junction depends upon complex interactions between occludin, claudins and ZO-1 15 . Lower occludin mRNA expression (Fig. 2b) could be responsible for wider tight junctions in MIF-KO mice. Although the expression of ZO-1, a scaffold protein that is involved in the linkage of junction proteins to actin filaments 15 , is increased in MIF-KO epithelial cells, a higher expression of claudin-2 ( Fig. 2b), which is involved in pore formation 16 , could further attribute to wider pores seen in between epithelial cells in the absence of MIF. Finally, there was no significant difference between WT and MIF-KO in the expression of claudin-4, a member of the tight junction complex that regulates the tightness of the gap between epithelial cells 16 .
It is already known that epithelial junctions can be disturbed through the influence of IL-1β and IL-18 17,18 . Consistent with the impaired morphology of epithelial junctions, MIF-KO epithelial cells of the large intestine produced higher levels of IL-18 compared to WT (Fig. 2c). However, IL-1β production was comparable between the strains (Fig. 2d). Since IL-18 is an end product of activation of microbial receptors 19 , the observed up-regulation of IL-18 might be a consequence of microbiota migration through a permeable intestinal barrier. Also, IL-18 is an inducer of IFN-γ production in lymphocytes 20 and this up-regulation in epithelial cells could serve to potentiate an adaptive immune response in the GALT.
The influence of MIF absence on immune surveillance in GALT. As already stated, colon GALT is comprised of intraepithelial lymphocytes, lymphocytes scattered within lamina propria and cells in mesenteric lymph nodes (MLN). Although intestinal permeability was considerably increased in the absence of MIF, the number of F4/80 + macrophages or CD3 + lymphocytes in the lamina propria did not differ between MIF-KO and WT mice ( Table 2 and Fig. 3a). In addition, no differences were found in the number of MLN cells or their proliferative capacity (Table 3). Likewise, the proportions of cells of the innate immune system: macrophages (M1 -pro-inflammatory, or M2 -anti-inflammatory), dendritic cells and NK cells were equal between the strains ( Table 3). Cells of the adaptive immune system: CD4 + helper lymphocytes, CD4 + CD25 + T regulatory cells, CD8 + cytotoxic lymphocytes, B220 + B cells or CD5 + CD19 + IL-10 + B regulatory cells were also similarly distributed in MLN of WT and MIF-KO mice (Table 3). Although the immune cell distribution appeared normal, macrophage-related production of TNF was considerably increased in MIF-KO MLN cells stimulated in vitro (Fig. 3b). It has already been shown that TNF up-regulates claudin-2 expression in epithelial HT-29/B6 cells and thereby disturbs tight junction formation. This scenario could be operable in MIF-KO mice since we indeed observed increased claudin-2 expression in the epithelium 21 . In contrast, IL-1β and IL-6 remained the same as in MLN cells of WT mice (Fig. 3b). Interestingly, the inflammatory milieu of MLN in MIF-KO mice was potentiated by decreased production of anti-inflammatory cytokine IL-10 (Fig. 3b) macrophage activation 5 and MIF deficiency should interfere with macrophage functions. This is not the case in our study, most likely as a result of other cytokines taking over some of the functions of MIF and responding adequately to the bacterial stimuli from the gut. As for cytokines from the immune cells of adaptive immunity, it was determined that lymphocyte-mediated production of IFN-γ was up-regulated in the absence of MIF (Fig. 3c), suggesting that Th1 and CD8 + populations might mediate the adaptive response to the unwanted entrance of gut microbiota. This is in accordance with the previously observed high expression of IL-18 by epithelial cells, since IL-18 can drive Th1 immunity against bacterial infections 20 . Apart from an anti-bacterial effect, another outcome of elevated IFN-γ could be the observed disturbance of tight junctions. Recent studies suggest that  Table 2). Finally, we have measured the intestinal content of IgA, an immunoglobulin produced by B lymphocytes and which controls the content of the microbiota 23 and found that IgA levels were considerably increased in MIF-KO colon compared to WT (Fig. 3d).
Overall, these results suggest that the perturbation in the production of cytokines within GALT is a reaction to the observed breach in the intestinal barrier that occurred as a consequence of MIF absence. Although characterized by higher cytokine production, the reaction of MIF-deficient immune cells to the microbiota exposure seemed to be quite limited. This might be explained by the role of MIF in the sampling of antigens from the gut lumen 9 . Without proper antigen transport, dendritic cells located beneath the microfold cells that mediate the transport cannot optimally perform their function.

MIF absence affects gut microbiota composition.
There is a constant bi-directional interplay between the mucosal immune system and gut microbiota. Since there is evidence of intestinal leakage and limited perturbation in GALT-associated immune surveillance in MIF-KO mice, it is presumed that MIF absence is associated with the disturbance of gut microbiota composition. Therefore, the weight and content of caeca were measured, since this is the place where a collection of innate microbiota of rodents resides 24 . In addition to a change in the caecum weight which was lower in MIF-KO mice (Fig. 4a), a distinguished difference in the microbiota composition between WT and MIF-KO mice was observed (Fig. 4b,d), including higher diversity of Lactobacilli strains (Fig. 4c,e). Interestingly, Ruminoccocus lactaris, that is a normal constituent of mouse microbiota 25 , was completely absent in MIF-KO gut (Fig. 4e). Further, Blautia producta, an indicator of the healthy gut and accountable for digestion of complex carbohydrates, was more abundantly present in samples collected from WT than in MIF-KO mice (Fig. 4e). This was the case when a universal primer set was used (Fig. 4e), while the difference in Blautia levels between WT and MIF-KO mice was not detected with a Lactobacilli-specific primer set (Fig. 4d).
Since higher Blautia levels are associated with the decrease of obesity in high-fat fed rats 26 , a possible correlation between Blautia levels and the fact that MIF-KO mice are prone to obesity development after 6 months of age 27 needs to be further investigated. On the other hand, Akkermansia muciniphila, an intestinal symbiont considered a guardian of intestinal mucosa 28 , was only detected in MIF-KO (Fig. 4e). Although the exact bacterial species related to the increased intestinal leakage were not detected, the mere change in the ratio between certain microbiota species 29 could account for the observed MIF-KO phenotype. The Lactobacilaceae species that inhabited the intestine of MIF-KO mice and not WT mice (specific strains from line 7 until 15 in Fig. 4e) likely emerged as a consequence of improper immune surveillance in GALT in the absence of MIF. Although the immune reaction in GALT manifested through the secretion of pro-inflammatory cytokines and IgA, it seems that MIF-KO mice were unable to properly control the gut microbiota content. In the immune response, MIF is generally involved in eradicating bacterial infections through up-regulation of toll-like receptor 4 or MHC class II on macrophages 5 , so the absence of MIF might impair these macrophage functions.
How does MIF regulate the function of the intestinal barrier? As already described, a functional intestinal barrier consists of the epithelial lining, immune cells within the epithelium and certain microbiota at the luminal side of the barrier. Having in mind all the results of this study, it was still unclear which participant of the intestinal barrier is primarily affected by MIF. In order to investigate this, a change in microbiota content was introduced in MIF-KO mice by treating the mice with antibiotics (ampicillin+neomycin) through drinking water for 14 days. The antibiotic treatment increased the caecum weight (Fig. 5a) and caused a microbiota disturbance in both WT and MIF-KO groups (Fig. 4b-e) in more or less the same manner. Particularly, lactobacilli representing the core microbiota in WT and MIF-KO mice before the antibiotic treatment were replaced by Enterococcus gallinarum. However, the average diameter of tight and adherens junctions in between epithelial cells of MIF-KO mice was significantly lower (p = 0) compared to the level observed before the treatment with antibiotics (Table 1 and Fig. 5b). This was also confirmed through analysing the serum concentration of FITC-dextran after per os administration. After the antibiotic treatment, FITC-dextran serum concentrations of MIF-KO mice were more similar to those of WT mice, suggesting better control of intestinal permeability (Fig. 5c). This was accompanied by an increase in E-cadherin presence (Fig. 5d) and occludin expression (Fig. 5e) that probably restored the architecture of the epithelial junctions in MIF-KO. Interestingly, the expression of all examined tight junction proteins was considerably up-regulated in both WT and MIF-KO mice, which probably indicates the physiological response to the antibiotic insult (Fig. 5e). This is corroborated with the results of the recent study showing that a 7-day neomycin application increases mRNA expression of ZO-1, claudin-3 and claudin-4 in the colon 30 .
Although antibiotics provoked growth of pathogenic or opportunistic pathogenic bacteria in MIF-KO mice, such as Streptococcus pneumoniae 31 , Mycoplasma falconis 32 , Streptococcus oligofermentans 33 , Clostridium innocuum 34 , Klebsiella oxytoca 34 , the number of macrophages ( Table 2, Fig. 5f) and lymphocytes ( Table 2, Fig. 5g) remained the same as before the treatment, while the levels of pro-inflammatory cytokines TNF and IFN-γ were reduced to the ones observed in MLN of WT mice under the antibiotic treatment (Fig. 5h). As for IgA levels, they did not  change upon treatment with antibiotics (Fig. 5i). Seemingly, GALT of MIF-KO mice did not recognize this change in microbiota species as a danger signal, as it did before the antibiotic treatment.
The relationship between MIF on one side and gut microbiota, epithelial barrier and gut immune system on the other side is a very complex one. Having in mind that forced microbiota alteration prevented barrier "leakage" we propose the following scenario for the increased intestinal permeability in the MIF absence. The lack of MIF provokes improper immune surveillance in GALT that favours the growth of new microbiota species that may secrete metabolites which make the barrier more permeable. In conclusion, by affecting surveillance and composition of gut microbiota, MIF indirectly regulates the permeability of the intestinal barrier.

Methods
Animals. C57BL/6 (WT) and MIF-KO mice (bred on C57BL/6 background) were kept under standard conditions at the Animal Facility of the Institute for Biological Research "Sinisa Stankovic". Male 2-6 months old  mice were housed in the same room, exposed to the same environment, and used for the experiments. All experiments were approved by the Ethical Committee of the Institute for Biological Research "Sinisa Stankovic" (No: 02-07/16) and were in accordance with Directive 2010/63/EU. The mice were kept under standard conditions (12 h light/dark cycles) with free access to standard pelleted diet and tap water.

Treatment with antibiotics. Both strains (3 months of age) were treated with Pentrexyl (ampicillin) and
neomycin (Galenika A.D. Belgrade, Serbia) for 14 days. Antibiotics (1 mg/ml) were given through drinking water (5 ml a day for each mouse). Ex vivo analysis was performed one day after the treatment was seized.
Isolation of immune cells. Immune cells were isolated from MLN which were removed aseptically and passed through a sterile plastic mesh (pores of 40 μm) in PBS + 3% FCS. After centrifugation, cells were counted and resuspended in RPMI 1640 supplemented with Antibiotic/antimicotic solution and 5% FCS (all purchased from PAA, Pashing, Austria). Cells were used either for obtaining conditioned supernatants (plated in 24-well plate for 24 h with or without concanavalin A -(ConA) -10 7 cells/ml), or placed in TriReagent (Fermentas, Vilnius, Lithuania) for obtaining protein and RNA fractions (5 × 10 6 cells per sample).

Isolation of gut epithelial cells.
Small and large intestine were removed aseptically, cut into pieces (approx. 3 cm long) and cut longitudinally. After removing the intestinal content, pieces were washed thoroughly in cold PBS, transferred to dissociating solution (2.7 mM KCl, 150 mM NaCl, 1.2 mM KH 2 PO 4 , 680 mM Na 2 HPO 4 , 1.5 mM EDTA, 0.5 mM DTT) and left for 15 minutes on ice. In order to detach loosened cells, the pieces were vortexed for 15 seconds and solution with cells was collected. After repeating this procedure on remaining intestinal pieces for three times, the suspension was centrifuged at 1000 g for 10 minutes and resuspended in the TriReagent. The purity of epithelial cells (>94%) was determined indirectly by staining cell samples with immune cell-related antibodies (1.5 ± 0.3% CD3 + lymphocytes, 2.3 ± 1.3% B220 + B lymphocytes, 0.6 ± 0.3% F4/80 + macrophages and 2.0 ± 0.4% CD11c + dendritic cells).

Measurement of intestinal permeability.
After overnight fasting, mice were administered with 50 μl of FITC-dextran (44 mg/100 g body weight) (Sigma-Aldrich, St. Louis, MO, USA) by oral gavage with a needle attached to a 1 ml syringe. After 4 hours, mice were bled from the retroorbital sinus and the intensity of the FITC dye was measured from obtained serum using Cameleon (Hidex, Turku, Finland) (excitation 488 nm, emission 534 nm). The measured values obtained from the sera without FITC-dextran (background fluorescence) were subtracted from the fluorescence intensity measured in the sera of treated mice. The exact concentration of FITC-dextran in the serum was calculated according to the standard curve values. Immunohistochemistry. Immunohistochemical analysis was performed on formalin-fixed, paraffinembedded sections using following antibodies and dilution ratios: rat monoclonal anti-mouse F4/80 antibody (1:250, Abcam, Cambridge, UK), rabbit polyclonal anti-mouse CD3 antibody (1:500, Abcam), and rat monoclonal anti-mouse E-cadherin (1:200, eBioscience). Briefly, after dewaxing and rehydration, a heat-inducing antigen retrieval procedure using citrate buffer at pH 6.0 for 21 min was performed on all tissue sections, with subsequent washing in PBS and endogenous peroxidase blocking with EnVisionTM FLEX Peroxidase-Blocking Reagent (Dako, Agilent, Santa Clara, CA, USA) for 10 min. Sections were incubated with primary antibodies overnight. Sections stained with anti-F4/80 and anti-E-cadherin antibody were incubated with secondary polyclonal rabbit anti-rat immunoglobulins/HRP (1:500, Dako) for 60 min. The CD3 sections were treated by applying the commercial EnVisionTM FLEX/HRP detection reagent (Dako). Immunoreactions were developed with diaminobenzidine (DAB, Dako) diluted in EnVisionTM FLEX Substarte Buffer (Dako). The sections were counterstained with haematoxylin.

Transmission electron microscopy (TEM) examination and ultrastructural morphometry.
Western blot. Proteins were isolated from the protein layer after the samples dissolved in TriReagent were centrifuged with chloroform at 12000 g following manufacturer's instructions. Protein concentration was measured using Lowry assay as previously described 36 and equal amounts of protein were layered onto acrylamide/ bis-acrylamide gel (concentration of the gel was dependent on the molecular weight of the target protein). The immunoblot procedure was performed as previously described 37  Measurement of IgA concentration in colon. Colon content (100 mg) was dissolved in 1 ml of PBS, vortexed for 5 min at room temperature and centrifuged (3000 g, 10 min). Supernatants were collected and 1 mM PMSF and 2 mM EDTA were added to prevent proteolysis. Samples were then vortexed and stored at −20 °C. IgA was detected by Western blot analysis and the final level of IgA was calculated as the ratio between the signal on the film and the level of total proteins in the supernatants measured by Lowry assay.
Measurement of cytokine levels. Cytokine concentration in cell culture supernatants was determined by sandwich ELISA using MaxiSorp plates (Nunck, Rochild, Denmark) and anti-mouse paired antibodies according to the manufacturer's instructions. Samples were analysed in duplicate for murine TNF, IL-6, IL-1β, IFN-γ, IL-17, IL-4, and TGF-β (eBioscience) and absorbance was measured by LKB microplate reader at 450 and 570 nm. A standard curve created from the known concentrations of appropriate recombinant cytokines was used to calculate concentration values of measured cytokines.
Denaturing Gradient Gel Electrophoresis (DGGE) analysis and DNA sequencing. Extraction of bacterial DNA from frozen faecal samples was done using the QIAamp DNA stool minikit (Qiagen, Hilden, Germany). DGGE analysis and gel manipulation after electrophoresis was entirely performed as described previously 38 . Primer sets complementary to 16S rDNA, specific for Lactobacilli (Lab-0159f paired with the universal reverse primer Uni-0515-GCr 39 ) and for Eubacteria (U-968-GC-f pared with L1401-r 40 ) were used. Fragments of interest were excised from the gel and macerated, and the suspension was incubated for 10 min at 98 °C 39 . After incubation, the suspension was centrifuged to pellet gel particles. The supernatant (30 ml) was used in PCRs with Lab-0159f and Uni-0515GCr primers 39 and U-968-GC-f and L1401-r 40 . The obtained PCR products were purified using the QIAquick PCR purification kit (Qiagen) and ligated into the pBluescriptT/A vector 41 . Ligated constructs were transformed in Ca 2+ -induced competent DH5α cells 42 , and insert-containing transformants were selected as white colonies on Luria agar (LA) plates containing 100 mg/ml ampicillin and 20 mg/ml X-Gal (5-bromo-4-chloro-3-indolyl-b-D-galactoside) as recommended by Promega. For each excised DNA band, one white colony was picked and plasmids were isolated using the QIAprep spin miniprep kit (Qiagen). The sequencing of the isolated insert-containing pBluescriptT/A plasmids was done with M13F/R primers at Macrogen Europe Service, Amsterdam, Netherlands (http://dna.macrogen.com/eng/support/ces/guide/universal_primer.jsp). Sequence annotation and the database searches for sequence similarities were performed with the BLAST tool available online (https://blast.ncbi.nlm.nih.gov/Blast.cgi).
Statistical analysis. Data are presented as mean ± SD or median and interquartile range (25th to 75th percentile). Statistical analysis was performed using Statistica 6.0 (StatSoft Inc., Tulsa, USA) software. Comparisons between the groups were done by 1-way ANOVA, followed by Student-Newman-Keuls post hoc test and finally Student's t-test or Mann-Whitney U-test where appropriate. p value less than 0.05 was considered to be statistically significant.
Data availability statement. The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.