Evasin‐displaying lactic acid bacteria bind different chemokines and neutralize CXCL8 production in Caco‐2 cells

Summary Chemokines are key signals in the immune system and play an important role as proinflammatory mediators in the pathology of inflammatory bowel disease and colorectal cancer, making them an important target for therapy. Recombinant lactic acid bacteria (LAB) were engineered to bind CC and CXC chemokines by displaying chemokine‐binding proteins evasin‐1, evasin‐3 and evasin‐4 on their surface. Evasin genes were cloned into lactococcal surface display vector and overexpressed in L. lactis NZ9000 and NZ9000ΔhtrA in fusion with secretion signal and surface anchor. Evasin‐displaying bacteria removed from 15% to 90% of 11 different chemokines from the solution as determined with ELISA and Luminex multiplexing assays, whereby L. lactis NZ9000ΔhtrA proved more efficient. Lactobacillus salivarius ATCC 11741 was coated with L. . lactis‐expressed evasin fusion protein, and its ability to bind chemokines was also confirmed. Evasin‐3‐displaying L. lactis removed 76.0% of IL‐1β‐induced CXCL8 from the supernatant of Caco‐2 epithelial cells. It also prevented secretion of CXCL8 from Caco‐2 cells in a time‐dependent manner when added before induction with IL‐1β. Evasin‐displaying LAB have the ability to bind multiple chemokines simultaneously and exert synergistic activity. This innovative treatment approach therefore has the potential for mucosal therapy of inflammatory bowel disease or colorectal cancer.


Introduction
Chemokines are small chemoattractant cytokines acting via seven transmembrane G protein-coupled receptors (GCPRs) that selectively induce recruitment and activation of immune cells to the site of infection. They are classified into four subfamilies, C, CC, CXC and CX3C, on the basis of the number and spacing of the conserved cysteine residues in the amino-terminus of the protein (Zlotnik and Yoshie, 2000). Historically, their name derived from their function (e.g. macrophage inflammatory protein: MIP-1a, b), while recently, a generic nomenclature has been applied (e.g. CCL1, CCL2). Chemokines and chemokine receptors play an important role in injury, inflammation, wound repair and cancer (Proudfoot et al., 2010) and are useful as diagnostic, prognostic and therapeutic targets (Castellani et al., 2007). Local mucosal recruitment and activation of neutrophils is a fundamental event in the pathogenesis of inflammatory bowel disease (IBD) and is mediated by interaction of chemokines with chemokine receptors on their target cells (Ajuebor and Swain, 2002;Wang et al., 2009;Atreya and Neurath, 2010). Expression of several chemokines, especially CXCL8, and also of the corresponding receptors, CXCR1 and CXCR2, is constantly increased during the active phase of IBD (Raab et al., 1993). Besides CXCL8, the levels of CXC chemokines CXCL1, CXCL5 and CXCL6 have been shown to be positively correlated with severity of inflammation in IBD patients (Ina et al., 1997). Additionally, recurrent inflammation and tissue destructive lesions that are accompanied by uncontrolled activation of effector immune cells in mucosa are associated with increased risk of colorectal cancer (CRC) (Triantafillidis et al., 2009;Sebastian et al., 2014). The levels of proinflammatory chemokines, such as CCL2, CCL3, CCL4 and CCL5, and pro-angiogenic chemokines, such as CXCL1, CXCL5 and CXCL8, are also elevated in human colon tumour tissues as compared to the matched normal tissues, indicating that these chemokines and their receptors play an important role in regulating colon tumour progression, angiogenesis and metastasis (Baier et al., 2005). The development of chemokine-binding proteins as potential therapeutic agents in IBD, or as preventive agents in inflammationassociated colorectal cancer, is therefore of great interest (Triantafillidis et al., 2011).
Many pathogenic organisms have developed chemokine and cytokine mediators with the ability to interfere with the host chemokine network and escape host detection and defence systems (Proudfoot et al., 2015). Evasins belong to a family of small chemokine-binding proteins (CKBPs). They have been identified in the salivary gland of the brown tick Rhipicephalus sanguineus and are probably used by the tick to inhibit the chemokine-mediated recruitment of leucocytes to the bite site (Frauenschuh et al., 2007). CKBPs have also been identified in several other organisms, including viruses, worms and arthropods (Seet and McFadden, 2002;Gonzalez-Motos et al., 2016). Unlike other CKBPs, evasins are smaller proteins (7-11 kDa) and display a more selective profile and a unique mechanism of chemokine binding (Deruaz et al., 2008;Bonvin et al., 2014). Evasins recognize and interact with chemokines with various degrees of selectivity and inhibit the binding of chemokines to their receptors and/or cell surface glycosaminoglycans. Evasin-1 binds CCL3, CCL4 and CCL18 (Frauenschuh et al., 2007), evasin-3 binds CXCL1 and CXCL8 (Deruaz et al., 2008), and evasin-4 is able to interact with almost 20 chemokines of the CC subfamily, particularly with CCL3, CCL5, CCL18 and CCL21 (Deruaz et al., 2013). Evasins have inhibited cellular recruitment in different murine models of disease and demonstrated potent anti-inflammatory properties (Deruaz et al., 2008;Castor et al., 2010;Russo et al., 2011;Braunersreuther et al., 2013;Copin et al., 2013;Montecucco et al., 2014;Bonvin et al., 2016).
Various lactic acid bacteria and other microorganisms have been suggested as a possible therapeutic approach for IBD treatment (Ng et al., 2009). Their therapeutic efficiency can be further improved by genetic modification of LAB (Kleerebezem and de Vos, 2011;Berlec et al., 2012;De Moreno de LeBlanc et al., 2015). Several strategies have been described including the production of antioxidant enzymes (Han et al., 2006), anti-inflammatory cytokine IL-10 (Schotte et al., 2000) or other anti-inflammatory compounds (Vandenbroucke et al., 2010). A somewhat different strategy included the display of cytokine-binding proteins on the surface of LAB (Ravnikar et al., 2010;Zadravec et al., 2015a,b;Berlec et al., 2016;Kosler et al., 2017), with the goal of preventing cytokine proinflammatory action. In the present research, novel therapeutic targets, chemokines, have been addressed in a similar fashionchemokinebinding proteins (CKBPs), evasins, were displayed on the surface of LAB and their ability to bind different chemokines was demonstrated. Engineered LAB constitute an innovative approach for the treatment of IBD or CRC.

Results
Expression and surface display of evasins on L. lactis NZ9000 and L. lactis NZ9000DhtrA Synthetic evasin genes (Frauenschuh et al., 2007;Deruaz et al., 2008) with L. lactis-optimized codon usage were cloned into the plasmid for surface display (pSDBA3b) (Ravnikar et al., 2010). Evasin gene constructs ( Fig. 1) were expressed as fusion proteins under the control of NisA promotor. Evasin_B domain fusion proteins were composed of four functional parts, including a signal sequence for secretion to the growth medium [derived from the Usp45 ( van Asseldonk et al., 1990)], the gene for chemokine-binding protein (evasin-1, evasin-3 or evasin-4), the gene for reporter protein B domain (IgGbinding B domain of staphylococcal protein A) and the gene for peptidoglycan-binding domain of AcmA (Buist et al., 1995) for surface attachment. Evasin fusion proteins ( Fig. 1)  (recognizing B domain) using Western blot (Fig. 2). When fusion proteins were expressed in L. lactis NZ9000 ( Fig. 2A), the bands with the highest molecular weight corresponded to the weight of the full-length proteins (Eva-1_B 42.2 kDa, Eva-3_B 39.8 kDa, Eva-4_B 43.2 kDa and B dom 32.9 kDa). The bands of lower molecular weight corresponded to degradation products. The expression of evasin_B domain fusion proteins in L. lactis NZ9000DhtrA (Fig. 2B) yielded bands of similar molecular weights; however, the extent of degradation was significantly lower (Fig. 2B). No expression was detected in negative control Fig 2A and B) or without nisin induction (not shown). Surface display of evasin_B domain fusion proteins on L. lactis NZ9000 and L. lactis NZ9000DhtrA was confirmed and quantified with whole-cell ELISA and flow cytometry (Fig. 2C), using antibodies against B domain. The extent of surface display on L. lactis NZ9000DhtrA is shown as it was higher than that achieved on L. lactis NZ9000. Statistically significant (P < 0.01; t-test) display of evasin fusion proteins on L. lactis cell surface as compared to the negative control was observed with both methods. Flow cytometry of evasin_B domain fusion protein-displaying cells showed a shift in mean fluorescence intensity (MFI) (Fig. S1) in comparison with the negative control (pNZ8148). The extent of displayed evasin_B domain fusion proteins was lower than that of displayed B domain fusion protein without evasins in the positive control. The highest extent of surface display was observed with evasin-4_B fusion protein with both methods.
Evaluation of chemokine binding by L. lactis NZ9000 and L. lactis NZ9000DhtrA with surface-displayed evasins Preliminary evaluation of chemokine-binding ability was performed with L. lactis NZ9000, and the results are shown in Table 1. The chemokine selectivity of evasin-1and evasin-4-displaying L. lactis NZ9000 for CC chemokines was determined by analysing the binding of CCL3, CCL4 and CCL5 by ELISA. 2 9 10 9 L. lactis NZ9000 cells with surface-displayed evasin-1 fusion protein bound 23% of CCL3 from the solution, but could not bind CCL5 (Table 1). 2 9 10 9 evasin-4-expressing L. lactis NZ9000 cells removed around 40% of CCL5 from the solution, but did not bind CCL3. None of the cells bound CCL4.
The effect of evasin-3-displaying L. lactis NZ9000 on the basal secretion of CXCL8 by Caco-2 cells (without IL-1b addition) was compared to that of control L. lactis NZ9000 (containing empty plasmid pNZ8148) or wildtype E. coli DH5a. 2 9 10 9 bacterial cells ml À1 were used, representing a multiplicity of infection (moi) of 2000:1 (bacterial: epithelial cells). L. lactis NZ9000 with surface-displayed evasin-3 decreased CXCL8 concentration in the supernatant of Caco-2 cells for 26.8% in comparison with empty plasmid control-containing L. lactis NZ9000 (Fig. 4A), while, on the contrary, incubation with E. coli cells increased the CXCL8 secretion for 53.3%. Induction of CXCL8 production in Caco-2 cells by adding 25 ng ml À1 of IL-1b for 6 h, followed by 2-h incubation with bacteria, resulted in much higher total concentrations of CXCL8. However, similar effects of bacteria were observed: control L. lactis NZ9000 (containing empty plasmid pNZ8148) had no effect on CXCL8 production, while evasin-3-displaying L. lactis NZ9000 caused 26.3% reduction in CXCL8 in the supernatant of Caco-2 cells (Fig. 4A).
Evasin-3-displaying L. lactis and Lb. salivarius prevented IL-1b-induced secretion of CXCL8 from intestinal epithelial cells The preventive effect of evasin-3-displaying bacteria on CXCL8 secretion was evaluated by pre-incubating Caco-2 cells with bacterial cells for 1 h, prior to addition of IL-1b. The concentration of CXCL8 was monitored over 6 h (Fig.  5). 1 9 10 9 evasin-3-displaying L. lactis NZ9000DhtrA cells significantly impeded the time-dependent increase in CXCL8 concentration in comparison with control cells (harbouring pNZ8148) or absence of bacteria. The final CXCL8 concentration after 6 h of monitoring was for 62% lower in the supernatant of Caco-2 cells pre-treated with evasin-3-displaying L. lactis NZ9000DhtrA cells than in the supernatant without bacterial cells. Lb. salivarius cells coated with evasin-3 fusion protein from growth medium of L. lactis NZ9000DhtrA were less effective. They reduced the final CXCL8 concentration in Caco-2 supernatant by 27% in comparison with the Caco-2 supernatant without bacteria after 6 h of monitoring.

Discussion
Engineered evasin-displaying bacteria were able to bind and remove 11 different recombinant chemokines of CC and CXC subfamilies. Synergistic targeting of different chemokines and prevention of the secretion of CXCL8 by the Caco-2 intestinal cell line suggest a significant potential of engineered bacteria in the treatment of IBD or CRC by addressing the relatively unexplored option of oral neutralization of chemokines, coupled by beneficial effects of probiotics.
Surface display of evasins on L. lactis was achieved using nisin-controlled expression (Mierau and Kleerebezem, 2005) of fusion proteins consisting of chemokine-binding proteins evasins (evasin-1, evasin-3 and evasin-4) (Deruaz et al., 2008), C-terminal peptidoglycan-binding domain of AcmA (Steen et al., 2003) and Table 2. The percentage of chemokines CCL3, CCL5 and CXCL8 removed from solution after incubation with 2 9 10 9 Lb. salivarius cells that were previously coated with evasin-1 (pSDEva1), evasin-3 (pSDEva3) and evasin-4 (pSDEva4) fusion proteins, respectively, relative to control Lb. salivarius cells. Growth media of recombinant L. lactis NZ9000DhtrA served as a source of fusion proteins. The extent of binding was normalized relative to the control Lb. salivarius cells incubated with the growth medium of L. lactis harbouring empty plasmid pNZ8148. Usp45 secretion signal (van Asseldonk et al., 1990;Ravnikar et al., 2010;Zadravec et al., 2015aZadravec et al., ,b, 2016. Genes for evasins were codon optimized for L. lactis to favour higher protein yield. A 15-amino acid peptide linker (Gly 4 Ser) 3 (Trinh et al., 2004) was included between the evasin and the peptidoglycan-binding domain in evasin-1 and evasin-3 gene constructs to enable mobility and flexibility of surface-displayed evasins. The B domain of staphylococcal protein A was included as a part of fusion proteins to serve as a reporter protein, because its effective expression and surface display in L. lactis has already been demonstrated (Ravnikar et al., 2010). Expression of evasin fusion proteins containing B domain was confirmed by Western blot. The multiple bands of lower molecular weight were a consequence of AcmA degradation (Poquet et al., 2000). Expression of fusion proteins in the L. lactis NZ9000DhtrA strain that lacks extracellular protease HtrA (Lindholm et al., 2004;Cortes-Perez et al., 2006) resulted in significant reduction of proteolytic degradation by HtrA and higher recombinant protein yields. Surface display of evasin fusion proteins was confirmed and its extent quantified with whole-cell ELISA and flow cytometry by detecting B domain reporter. The extent of surface display of evasin-B domain fusion proteins was lower than that of fusion proteins containing only B domain. This may be the consequence of lower protein yield, larger size of the fusion protein, hindered availability of B domain in the displayed fusion protein or higher proteolytic susceptibility. The level of surface display was highest with evasin-4 fusion protein, the only fusion protein in which (Gly 4 Ser) 3 linker was not included due to cloning issues. Although surprising, this is in agreement with our previous report in which the linker did not contribute to the extent of surface display (Zadravec et al., 2014). The lower level of fusion protein degradation in NZ9000DhtrA resulted in the increased extent of surface display. Although this is in accordance with A. Caco-2 cells with or without IL-1b induction were untreated (Cont.) or were exposed to 2 9 10 9 cells ml À1 of bacteria (E. coli DH5a, L. lactis NZ9000 containing empty plasmid control pNZ8148 or plasmid pSDEva3). B. Removal of CXCL8 from the supernatant of IL-1b-induced Caco-2 cells by 2 9 10 8 , 1 910 9 or 2 9 10 9 cells ml À1 of L. lactis NZ9000 DhtrA (white bars) or 2 9 10 9 cells ml À1 of Lb. salivarius cells coated with growth medium of L. lactis NZ9000DhtrA (black bars) displaying evasin-3. The results were from three independent experiments performed in triplicate and are expressed as mean AE SD. Significant differences (t-test) are marked with asterisks. (** < 0.01; ***P < 0.001).

Fig. 5.
Time-dependent secretion of CXCL8 by Caco-2 cells after the preventive 1-h treatment with evasin-3-displaying bacteria (triangles) L. lactis NZ9000DhtrA (A) or Lb. salivarius coated with evasin-3-fusion protein-containing growth medium of L. lactis NZ9000DhtrA (B) that was followed by the addition of 25 ng ml À1 IL-1b (at time 0). CXCL8 secretion by Caco-2 cells was compared to that after incubation with control bacteria (squares; L. lactis NZ9000DhtrA harbouring pNZ8148 (A) or Lb. salivarius coated with growth medium of L. lactis NZ9000DhtrA cells harbouring pNZ8148 (B) or to that without bacteria (circles). expectations, it was not observed in our previous study in which only B domain was displayed (Ravnikar et al., 2010), possibly pointing to the susceptibility of evasins or linker to degradation. Taken together, these results demonstrate that L. lactis NZ9000DhtrA strain stabilizes surface-associated proteins and is more appropriate as an expression host for evasin fusion proteins.
The functionality of all three evasins on the bacterial surface was confirmed indirectly, via the greater ability of evasin-displaying bacterial cells to remove chemokines from the solution than that of control cells. The functional expression is noteworthy because of the presence of intramolecular disulfide bonds which are less common in proteins of Gram-positive LAB (Davey et al., 2016). Among recombinant proteins, disulfide bond-containing bacteriocins have been produced in L. lactis, and evasins may share the same mechanisms of disulfide bond formation (Back et al., 2016;Davey et al., 2016).
Using different evasins-displaying bacteria, the binding of 11 different chemokines: five human CC chemokines (CCL3, CCL5, CCL11, CCL18 and CCL25), four human CXC chemokines (CXCL2, CXCL8, CXCL5 and CXCL6) and two murine CXC chemokines (CXCL1 and CXCL2) has been demonstrated. The binding specificity of evasins was in accordance with that published. The extent of chemokine removal varied from app. 15% (CXCL5) to more than 90% (CXCL8) and was largely dependent on bacterial concentration. As seen with the expression and surface display, the binding of chemokines was improved when fusion proteins were expressed in L. lactis NZ9000DhtrA. The presence of B domain in fusion proteins had minor, usually detrimental, effect on binding, although positive effects were also observed, pointing to the importance of steric positioning of evasins for binding of different chemokines. CC chemokines were removed by evasin-1-and evasin-4-displaying bacteria. Evasin-1-displaying bacteria were capable of binding CCL3, but not the closely related CCL5, although binding of CCL5 by evasin-1 was previously reported (Frauenschuh et al., 2007). This may be the consequence of insufficient concentration of cells, or incorporation of evasin-1 into the fusion protein via its C-terminus, as it was shown that C-terminus was involved in chemokine binding (in contrast to evasin-4 that binds chemokines via its N-terminus) (Deruaz et al., 2013). Evasin-4 has a broader selectivity pattern than evasin-1 and evasin-3 and was previously shown to bind almost 20 chemokines of the CC subfamily, among them CCL5 with the highest affinity (Deruaz et al., 2013). Removal of CCL5 by evasin-4displaying L. lactis cells was indeed very effective (42.3%). Apart from CCL5, evasin-4-displaying cells also bound CCL3, CCL11, CCL18 and CCL25 in a multiplex Luminex assay. CXC chemokines (human CXCL2, CXCL5, CXCL6, CXCL8 and mouse homologues CXCL1 and CXCL2) were bound by evasin-3-displaying L. lactis cells. Binding of human CXCL1 was not detected, possibly due to the multiplex Luminex assay format and the competition by five different chemokines. This could also be the reason for the generally lower percentage of binding detected with Luminex. Detailed characterization of binding with evasin-3-displaying L. lactis cells was performed with CXCL8. As observed with evasin-1-or evasin-4-displaying cells, binding of CXCL8 was more effective with higher number of evasin-3-displaying bacterial cells.
Lactobacillus salivarius ATCC 11741 has been established earlier as a suitable host for heterologous surface display of proteins containing the peptidoglycan-binding domain of AcmA (Zadravec et al., 2015a,b;Kosler et al., 2017). It was shown that not all AcmA-containing fusion protein, produced and secreted by recombinant L. lactis, binds to the producer cells. The fusion protein-containing growth medium can therefore be used for heterologous coating of non-recombinant bacterial cells (Bosma et al., 2006), representing a non-GMO alternative (Zadravec et al., 2015a,b) or a surface display option for hard to transform bacteria. Lb. salivarius was coated with evasin fusion proteins produced by L. lactis, and its ability to bind chemokines was confirmed.
Inflammation of the intestinal epithelial cells is usually driven by inflammatory cytokines from the immune cells of the lamina propria. This can be recreated in the Caco-2 epithelial cell model by stimulation with IL-1b which triggers strong secretion of CXCL8 (Luerce et al., 2014) that can accumulate in the growth medium for several hours. Evasin-3-displaying L. lactis significantly reduced CXCL8 concentration (by up to 76.0%) in Caco-2 cell supernatant in comparison with untransformed L. lactis or E. coli. Evasin-3-coated Lb. salivarius was less effective in binding Caco-2-produced CXCL8 than evasin-3displaying L. lactis, despite their similar binding abilities as observed with ELISA (Tables 1 and 2). This might be due to the intrinsic slightly proinflammatory phenotype of Lb. salivarius that was shown to be able to induce TNFa production in THP-1 cells (Diaz et al., 2013). However, Lb. salivarius coated with evasin-3 fusion protein-containing NZ9000DhtrA growth medium could still remove around 20% of CXCL8 from the Caco-2 supernatant.
Preventive addition of evasin-3-displaying L. lactis to Caco-2 cells for 1 h before exposure to IL-1b was shown to reduce CXCL8 secretion in a time-dependent manner. A less intense, but still significant decrease in CXCL8 was observed with control L. lactis, in line with the intrinsic anti-inflammatory activity of L. lactis (Berlec et al., 2016). The decrease achieved with evasin-3-displaying Lb. salivarius was again less pronounced than that achieved with evasin-3-displaying L. lactis.
The preventive addition of chemokine-binding bacteria to the epithelial cell model resembles the intended administration of bacteria in intestinal inflammations and suggests a significant potential for mucosal therapy of inflammation occurring in IBD or colorectal cancer. This potential is strengthened by the intrinsic evasin affinity for different chemokines. The chemokine system is promiscuous, with individual chemokines binding to several receptors and individual receptors recognizing multiple chemokines (Allegretti et al., 2012;Steen et al., 2014). Targeting a single chemokine or single receptor would therefore not be sufficient to exert a significant effect in vivo (Koelink et al., 2012). However, targeting multiple chemokines with three different evasins could avoid redundancy and enable synergy (Bonvin et al., 2016). The ability of evasin-displaying bacteria to bind multiple cytokines simultaneously has been confirmed in this study using multiplex assays.
To conclude, we present for the first time the expression of biologically active evasins and their display on the surface of LAB. We have shown that evasin-displaying LAB are able to bind at least 11 different chemokines, some of them at the same time. Apart from binding the recombinant chemokines, the evasin-displaying bacteria also prevented CXCL8 secretion in the epithelial cell model. The results obtained thus warrant expanding the studies on chemokine-binding bacteria to animal models of intestinal inflammation with the aim of confirming their usefulness in the treatment of IBD and CRC.

Bacterial strains, media and growth conditions
The bacterial strains used in this study are shown in Table S1. Lactococcus lactis NZ9000 (de Ruyter et al., 1996;Kuipers et al., 1998;Mierau and Kleerebezem, 2005) and Lactococcus lactis NZ9000DhtrA (Lindholm et al., 2004) were grown at 30°C in M17 medium (Sigma) supplemented with 0.5% glucose (GM-17) without agitation or in the same medium solidified with 1.5% agar. To maintain selection pressure on transformation, 10 lg ml À1 chloramphenicol or erythromycin, or both, was added to the growth medium of L. lactis NZ9000 and L. lactis NZ9000DhtrA. Lactobacillus salivarius ATCC 11741 was grown in De Man, Rogosa and Sharpe (MRS) medium (Merck) at 37°C without aeration. E. coli strain DH5a was grown at 37°C with agitation in lysogeny broth (LB) medium supplemented with 100 lg ml À1 ampicillin.

DNA manipulation and plasmid construction
Detailed cloning procedures are described in Supporting information. Primers (IDT) and plasmids are listed in Table S1. Scheme of evasin constructs pSDEva1, pSDEva3 and pSDEva4 is shown in Fig. 1.

Expression of evasin fusion proteins in L. lactis
Overnight cultures of L. lactis NZ9000 and L. lactis NZ9000DhtrA harbouring plasmids pNZ8148, pSDBA3b, pSDEva1, pSDEva1_B, pSDEva3, pSDEva3_B, pSDEva4 and pSDEva4_B were diluted (1:100) in 10 ml (or 100 ml) of fresh GM-17 medium and grown to an optical density (A 600 ) of 0.50-0.80. Fusion protein expression was induced with 25 ng ml À1 nisin (Fluka) (de Ruyter et al., 1996;Kuipers et al., 1998;Mierau and Kleerebezem, 2005). After 3 h of incubation, 1 ml of culture was stored at 4°C for flow cytometric analysis, and the remaining cell culture was centrifuged at 5000 g for 10 min. The cell pellet was resuspended in 400 ll of phosphate-buffered saline (PBS,pH 7.4) and stored at À20°C for SDS-PAGE analysis, or resuspended in different volumes of PBS for whole-cell ELISA and assay of the chemokine-binding ability. The supernatant was decanted, filtered with a 0.22 lm pore size filter (Minisart; Millipore, Darmstadt, Germany), aliquoted and stored at À20°C for testing the heterologous coating of nonrecombinant bacteria with evasin fusion proteins.

Heterologous coating of Lactobacillus salivarius with evasin fusion proteins
The coating of Lb. salivarius was performed as described previously (Zadravec et al., 2015a,b;Kosler et al., 2017). Lb. salivarius overnight culture containing 2 9 10 9 cells ml À1 was centrifuged (5000 g, 5 min, 4°C) and resuspended in different volumes (10 ml in Caco-2 cell experiments) of the conditioned growth medium of L. lactis NZ9000 or L. lactis NZ9000DhtrA that contained evasin fusion proteins. Bacterial suspension was shaken gently for 2 h at room temperature (RT). After centrifugation (5000 g, 5 min, 4°C), the cells were resuspended in PBS for assaying the chemokine-binding ability with ELISA as described below. The growth medium of empty plasmid (pNZ8148)-containing L. lactis was used as a control and was incubated with Lb. salivarius in the same manner.

SDS-PAGE and Western blot
Detailed procedure is described in Supporting information.

Flow cytometry
Details of flow cytometry are described in Supporting information.
Chemokine binding by evasin-displaying lactic acid bacteria Different volumes of L. lactis NZ9000 and L. lactis NZ9000DhtrA expressing evasin-1, evasin-3 or evasin-4 and Lactobacillus salivarius ATCC 11741 coated with evasin-1, evasin-3 and evasin-4 fusion proteins were centrifuged (5000 g, 5 min, 4°C), washed twice with 500 ll PBS and finally resuspended in 200 ll of PBS containing various concentrations of chemokine standards (from ELISA kits, see below) and incubated 2 h at room temperature (RT) with gentle shaking. Cells were then removed by centrifugation (5000 g, 10 min, 4°C) and 100 ll of the supernatant collected to examine the content of chemokines using ELISA kits (Mabtech, Nacka Strand, Sweden; R&D Systems, Minneapolis, MN, USA; and PeproTech, London, UK). Alternatively, 50 ll of the supernatant was used to examine the content of chemokines in multiplexing system (Luminex 100 or MAGPIX, 's-Hertogenbosch, The Netherlands). The chemokine-binding ability of evasin-displaying bacteria was presented as the portion of cytokine removed from the solution by evasin-displaying bacteria, in comparison with control bacteria, harbouring pNZ8148 plasmid.
Caco-2 cells were seeded at 1 9 10 5 cells per well in 24-well plates and incubated at 37°C with 5% CO 2 for 24 h before treatment, as reported (Luerce et al., 2014). Secretion of CXCL8 was induced by the addition of different concentrations (250, 25, 2.5, 0.25, 0.025, 0.0025, 0.00025 ng ml À1 ) of recombinant human IL-1b (Cell Genix, Freiburg, Germany). To examine the production of CXCL8, Caco-2 cells were incubated with 25 ng ml À1 of IL-1b for different time periods (0-48 h). 25 ng ml À1 of IL-1b for 6 h was used when incubating with bacteria. Cell cultures were centrifuged (10 min 200 g at 4°C and 10 min 16 000 g at 4°C), and the supernatant was collected and stored at À80°C until analysis. CXCL8 levels were measured using a Human IL-8 (CXCL8) ELISA development kit (Mabtech) as described above.
Cultures of E. coli DH5a, L. lactis NZ9000 or L. lactis NZ9000DhtrA, harbouring plasmids pNZ8148 (control) and pSDEva3, as well as evasin-3-coated Lb. salivarius, were centrifuged, washed two times with PBS and finally resuspended in DMEM. In CXCL8 removal experiments, bacterial cells were added to the Caco-2 cell culture (preincubated with IL-1b for 6 h) at a final concentration of 2 9 10 8 , 1 9 10 9 or 2 9 10 9 cells ml À1 . CXCL8 secretion was measured after 2 h of co-incubation. To prevent CXCL8 secretion, we incubated bacterial cells with Caco-2 cells for 1 h before inducing with IL-1b for another 6 h. Untreated Caco-2 cells secreting baseline levels of CXCL8 were used as a control. Data from three independent experiments were considered. Viability of Caco-2 cells after incubation with bacteria was tested with Trypan blue exclusion staining.