Beneficial bacteria activate type-I interferon production via the cytosolic sensors STING and MAVS

Type-I interferon (IFN-I) cytokines are produced by innate immune cells in response to microbial infections, cancer and autoimmune diseases. These cytokines trigger protective responses in neighbouring cells through the activation of IFN-I stimulated genes. One of the most predominant pathways associated with IFN-I production is mediated by the cytosolic sensors STING and MAVS, intracellular adaptors that become activated in the presence of microbial nucleic acids in the cytoplasm, leading to IFN-I production via TANK-binding kinase (TBK)-1 and IFN regulatory factors. However, the role of these sensors in responses induced by beneficial microbes has been relatively unexplored. Here we have screened 12 representative strains of lactic acid bacteria (LAB), a group of beneficial microbes found in fermented food and probiotic formulations worldwide, for their ability to trigger IFN-I responses. Two isolates (Lactobacillus plantarum and Pediococcus pentosaceus) induced an IFN-I production that was significantly higher that the rest, both in macrophage cell lines and human primary macrophages. This response correlated with stronger interaction with macrophages and was susceptible to phagocytosis inhibitors, suggesting bacterial internalisation. Accordingly, macrophages deficient for STING and, to a lesser extent, MAVS failed to respond to the two LAB, showing reduced TBK-1 phosphorylation and IFN-I activation. Furthermore, LAB-induced IFN-I was biologically active and resulted in expression of interferon stimulated genes, which was also STING- and MAVS-dependent. Our findings demonstrate a major role for STING in the production of IFN-I by beneficial bacteria and the existence of bacteria-specific immune signatures, which can be exploited to modulate protective responses in the host.

Human TNF-α production was detected and quantified in the supernatants of LAB-challenged PMA-2 2 6 differentiated THP-1 cells using the eBioscience Human TNF-α ELISA Ready-SET-Go kit as 2 2 7 indicated by the manufacturer's instructions. The Human IFN-β Quantikine ELISA Kit (R&D 2 2 8 systems) was used to detect and quantify IFN-β in supernatants from primary macrophages exposed to 2 2 9 LAB for 2h. Supernatants were collected every 4 h for 12h after the 2 h phagocytosis 2 3 0 Statistical analysis 2 3 1 Statistical analysis was performed using GraphPad Prism. Data are presented as means ± standard 2 3 2 deviation (SD) and are representative of one experiment of at least three independent experiments. 2 3 3 Data from experiments with human peripheral blood mononuclear cells are representative of two or 2 3 4 three healthy donors and are mean with SD from two biological replicates. Statistical significance 2 3 5 between one sample and its corresponding control was determined using the Student's t-test and 2 3 6 1 0 within a group of samples using one way ANOVA followed by Fisher's Least Significant Difference 2 3 7 (LSD) Test. The ability of 12 LAB strains (Table 1) to trigger the activation of the inflammatory transcription 2 4 1 factors NF-κB and IRF-3 was evaluated in human differentiated THP-1 cells. We employed 2 lines of 2 4 2 THP-1 monocytes expressing GLuc under the control of either the NF-κB promoter or the promoter of 2 4 3 the IRF-3-dependent gene IFIT1. The cells were differentiated for 48 h and subsequently exposed to 2 4 4 live or dead LAB before GLuc activity was measured in the media and presented as a fold increase 2 4 5 over non-stimulated macrophages (Fig. 1A). The majority of LAB species tested -S. thermophilus, P. 2 4 6 acidilactici, L. sakei, L. kunkeei, L. casei and E. faecalis-induced significant NF-κB activation in 2 4 7 THP-1 macrophages and, with the exception of E. faecalis, this activation was enhanced in response to 2 4 8 live bacteria compared to inactivated bacterial cells. By contrast, we found that these isolates were 2 4 9 poor inducers of IFIT1 activation either dead or alive. Interestingly, two bacterial species -2 5 0 Pediococcus pentosaceus (PP) and Lactobacillus plantarum (LP) -were able to induce a significant 2 5 1 IFIT1 activation. This IFIT1 response was only observed with viable bacteria and was especially high 2 5 2 with L. plantarum. Interestingly, neither PP nor LP enhanced NF-κB activation when used as viable 2 5 3 bacteria. To visualize this contrasting behavior, we then plotted the recorded NF-κB vs IFIT1 2 5 4 activation for each LAB species and drew arbitrary cut-offs for NF-κB activation (~20 fold) and IFIT1 2 5 5 (~10 fold) based on the most common response observed in all bacterial species tested (Fig. 1B). As 2 5 6 expected most LAB grouped in the top left quadrant (Q1) as powerful NF-κB agonists, but poor IFIT1 2 5 7 inducers. Strikingly, LP and PP appeared in the bottom right quadrant (Q4) as remarkably strong 2 5 8 IFIT1 inducers, especially LP. The 2 L. lactis species occupied the bottom left quadrant (Q2) as 2 5 9 intermediate NF-κB and IFIT1 inducers. We then performed dose-dependent exposures with LP and 2 6 0 PP and showed that a challenge with 1 LP and 10 PP bacterial cells per macrophage was sufficient to 2 6 1 trigger IFIT1 activation (Fig. 1C). The highest IFIT1 response was observed with a ratio of 25 bacteria 2 6 1 1 per macrophage. Our screening therefore indicated that some LAB could activate responses 2 6 3 converging on IFN-I production.

6 4
Lactobacillus plantarum (LP) and Pediococcus pentosaceus (PP) activate IFN-I production 2 6 5 To address expression of IFN-I and verify the activation observed using luciferase reporters, we used a 2 6 6 commercial ELISA against the NF-κB-dependent inflammatory cytokine TNF-α and an ISRE-based 2 6 7 bioassay to quantify the presence of TNF-α and IFN-I in the supernatants of macrophages exposed to 2 6 8 each of the LAB isolates ( Fig. 2A and 2B, respectively). Apart from LP, PP and isolates of L. lactis 2 6 9 (LL), the exposure of macrophages to all LAB isolates resulted in high amounts of TNF-α in the 2 7 0 media, but very low levels of IFN-I. Conversely, LP and PP were capable of inducing a significant 2 7 1 ISRE activation as indicative of effective production of IFN-I, especially in the case of LP (Fig. 2B).

7 2
Similarly to the luciferase-based results, when the production of TNF-α and IFN-I were plotted we 2 7 3 identified a strong IFN-I signature for LP that correlated with a very low impact on TNF-α (bottom 2 7 4 right quadrant Q4 of Fig. 2C). PP showed a more moderate IFN-I signature and a slightly higher TNF-2 7 5 α production than LP (next to the bottom left quadrant Q2 of Fig 2C), but this was still much lower 2 7 6 than that observed with most of the LAB isolates (top left quadrant Q1 of Fig. 2C). Taken together, 2 7 7 our screens in human macrophages demonstrated that each LAB has the ability to stimulate innate 2 7 8 immunity differently and that some are potent inducers of IFN-I responses. Lactobacillus plantarum (LP) and Pediococcus pentosaceus (PP) activate responses 2 8 0 associated with IFN-I production in human PBMCs.

8 1
Next we sought confirmation of this LAB-induced IFN-I activation in human primary cells.We first 2 8 2 employed an IFN-β ELISA to detect and quantify the presence of this cytokine in the supernatants of 2 8 3 human monocyte-derived macrophages collected from PBMC and exposed to LP and PP for 2h. We 2 8 4 observed a peak of IFN-β production 8 h post-challenge that decreased at 12 h (Fig. 3A). We then 2 8 5 used flow cytometry to monitor the expression of the IFN-I-related markers CD64 and CD40 (34-36) 2 8 6 in monocytes collected from PBMC from two healthy donors upon exposure to LAB. IFN-I 2 8 1 4 cells (Fig. 6B). However, only the reduction in IFN-β expression was found to be statistically 3 4 4 significant. Finally, media from cells exposed to LP (the most potent inducer of IFN-I) were subjected 3 4 5 to ISRE bioassay. In agreement with gene expression data, absence of STING and, to a lesser extent, 3 4 6 MAVS significantly reduced the amount of biologically active IFN-I (Fig. 6C). We observed similar 3 4 7 results with KO cells exposed to LP but the reduction in functional IFN-I was less clear and equally 3 4 8 significant from both KO cells (Fig. 6D).

4 9
Discussion 3 5 0 This is the first publication reporting how human macrophages produce IFN-I in response to LAB via 3 5 1 STING and MAVS. It is very well established that pathogenic bacteria are sensed by these cytosolic 3 5 2 adapters to stimulate the production of IFN-I in macrophages (37). However, the role that STING and 3 5 3 MAVS play in the recognition of beneficial bacteria such as LAB has been underexplored up to now.

5 4
Our findings were generated from an initial unbiased screen using different representative LAB 3 5 5 species. This screen showed that LAB activate IFN-I production in human macrophages in a species-3 5 6 dependent manner as only LP and PP were able to induce a significant induction of luciferase in THP-3 5 7 1-IFIT1-GLuc macrophages. This species-dependent IFTI1 activation has also been observed with 3 5 8 other strains of LP and PP that we have recently isolated from animals (38, 39). The central dogma is 3 5 9 that LAB activate NF-κB via TLR2 (40), TLR9 (41) and Nod-like receptors (42). Our macrophage 3 6 0 challenge with LAB species such as Enterococcus faecalis, Lactobacillus casei, L. sakei and 3 6 1 Pediococcus acidilactici resulted in a very high NF-κB activation, as widely reported in previous 3 6 2 publications (38,(43)(44)(45). Unlike most of the selected LAB though, neither LP nor PP significantly 3 6 3 activated the NF-κB pathway, but both were able to induce the exogenous production of IFN-I in 3 6 4 human macrophage-like cells and human primary phagocytes isolated from PBMCs. This is a 3 6 5 remarkable finding considering that only a few studies have reported that LAB are capable of inducing 3 6 6 the production of IFN-I in innate immune cells (7,8,46,47).

6 7
Here, we have observed that LP and PP up-regulate the expression of CD40 in monocytes 3 6 8 from PBMCs, whilst displaying an antagonistic effect on CD64. Other studies with PBMCs have 3 6 9 reported that the production of IFN-β is associated with the up-regulation of CD40 (48, 49); on the 3 7 0 1 5 contrary, the presence of this cytokine leads to the down-regulation of CD64 (36). Therefore, our 3 7 1 results with CD40 and CD64 suggest that the interaction of LP and PP with human phagocytes induces 3 7 2 the production of IFN-β and this IFN-β has biological impact on human immune cells. In this respect, 3 7 3 Weiss et al. (35) found that LAB that induce IFN-β activation are also able to stimulate CD40, a 3 7 4 positive convergence that we have confirmed using macrophages derived from monocytes of PBMCs. 3 7 5 These macrophages secrete IFN-β following the phagocytosis (or binding) of LP and PP, with a 3 7 6 significant increase at 8h post-challenge. This production peak is in agreement with the previous work explain why the levels of IFN-β production that they detected (> 500 pg/mL) are much higher than 3 8 0 those observed by us (25-50 pg/mL).

8 1
The IFN-I activation that we have recorded with LP and PP requires interaction with 3 8 2 paghocytes, as previously reported with dentritic cells stimulated with other LAB species (7). The 3 8 3 viability of LP and PP was crucial to stimulate the IFN-I production in macrophages, and also that 3 8 4 both LP and PP bind and/or are phagocytosed by human monocytes and macrophages. Alive cells of 3 8 5 LP and PP trigger an endogenous IFN-I production that is significantly higher than that observed with 3 8 6 inactivated cells. Based on these findings, we believe that the phagocytic intake of viable cells of LP 3 8 7 and PP is essential to observe a good IFN-I response, which suggests a predominant role of cytosolic 3 8 8 sensors such as STING and MAVS on the recognition of both bacterial species. In general, 3 8 9 macrophages exposed to heat-killed (or inactivated) bacterial cells activate TRL/NOD-like receptors-3 9 0 dependent pathways, whereas live bacteria activate other pathways that require phagocytosis, 3 9 1 proteolytic bacterial degradation and phagolysosomal membrane destruction, leading to the release of 3 9 2 bacterial nucleic acids into the cytosol (50).

9 3
The cytosolic adapters STING and MAVS sense the presence of bacterial DNA or RNA in the 3 9 4 cytosol, resulting in the production of IFN-I via a pathway dependent on the phosphorylation of TBK1 3 9 5 (21, 22). Our TBK-1 immunoblotting assays and the transcriptional analysis on the mRNA expression 3 9 6 of IFN-β have showed that the intracellular presence of PP and LP in THP-1 macrophages activate 3 9 7 1 6 TBK-1 and the subsequent overexpression of IFN-β. Moreover, this IFN-I activation is dependent on 3 9 8 the presence of STING and MAVS as the THP-1 cells KO for either STING or MAVS were less 3 9 9 capable of activating TBK-1 and IFN-β. This observation was more evident with STING KO cells 4 0 0 exposed to LP. In this respect, some studies have reported that the bacterial recognition by STING and 4 0 1 MAVS may be dependent on the species (37); while others have demonstrated that STING is the 4 0 2 central player in the crosstalk between DNA and RNA sensing (51). In addition, we have proved that influence of STING was more evident as previously observed with regards to the endogenous IFN-I 4 0 9 activation. Furthermore, the ISGs MxA and OAS1 overexpressed in macrophages exposed to LP and 4 1 0 PP, especially to LP, and this overexpression decreased in the absence of either STING or MAVS. In this study we have demonstrated that STING and MAVS induce IFN-I production by 4 1 8 sensing the presence of cytosolic LAB. Nevertheless, we cannot rule out other IFN-I activation routes 4 1 9 such as TLR2/3 recognition via endosomes, as previously reported with other LAB (7, 8). Another 4 2 0 important aspect that is worth emphasizing is the fact that the activation of IFN-I through STING and 4 2 1 MAVS occurs in a species-dependent manner. Why macrophages are more responsive to certain LAB 4 2 2 species such as LP and PP and how nucleic acids and/or CDNs of these species are recognized by 4 2 3 STING and MAVS are very important questions that remain to be elucidated. The synthesis of CDNs 4 2 4 1 7 has been described in LAB, although very little is known about their role in bacterial physiology and 4 2 5 innate immune responses (53). In consequence, it is early to speculate whether DNA or CDNs of LAB 4 2 6 are more or less important for STING activation. However, the evidence that LP and PP hardly 4 2 7 activate NF-κB suggest a major influence of DNA. RECON, a cytosolic sensor that has been 4 2 8 discovered very recently (54), antagonize STING activation by binding bacterial CDNs, resulting in 4 2 9 NF-κB activation. B. Expression of CD64 in monocytes (blue) and neutrophils (red) exposed to LP and PP. The CD expression is represented as a percentage of fold change (increase or decrease) over a non-stimulated condition using unchallenged monocytes and neutrophils. Lactobacillus casei (LC) is included as a negative control (no IFN-I inducer). Data are mean with SD from two healthy donors and the comparative analysis was carried out with two way ANOVA and Tukey multiple comparison (* p<0.05, ** p<0.01).
C. Expression of CD40 in monocytes (blue) and neutrophils (red) exposed to LP and PP. The CD expression is represented as a percentage of fold change (increase or decrease) over a non-stimulated condition using unchallenged monocytes and neutrophils. Lactobacillus casei (LC) is included as a negative control (no IFN-I inducer). Data are mean with SD from two healthy donors and the comparative analysis was carried out with two way ANOVA and Tukey multiple comparison (* p<0.05, ** p<0.01).

Fig. 4. Human phagocytes interact with Lactobacillus plantarum (LP) and Pediococcus pentosaceus (PP)
A. Uptake of LP and PP by monocytes and neutrophils from PBMCs of 2 healthy donors. Viable bacterial cells were labelled with FITC and incubated in whole human blood for 1h at a multiplicity of infection of 1:25. Blood cell populations were distinguished based on side scatter area (SSC-A) versus forward scatter area (FSC-A), and separated into lymphocytes (grey), monocytes (blue) and neutrophils (red). Phagocytic uptake was then observed in the FITC channel, where the intensity was divided into three subpopulations based on the positivity: no interaction (-), surface binding (+), and phagocytic uptake (++). Lactobacillus casei (LC) is included as a control of no phagocytic uptake.
B. Percentage of monocytes in each of the three FITC subpopulations -no interaction, surface binding and phagocytic uptake-after exposure to LC (grey), LP (green) and PP (blue). The comparative analysis was carried out with two way ANOVA and Tukey multiple comparison *, p<0.05 **, p<0.01; ****, p<0.001).
D. Confocal microscopy images showing the detection of LP and PP inside monocyte-derived macrophages (indicated with arrows). FITC and DAPI were used to label bacteria and the macrophage nucleus, respectively. E. Number of bacteria (LC, LP and PP) that remain inside (and/or attached to) THP1 macrophages after a 2h RPMI incubation in the absence (black) or presence (white) of the phagocytosis inhibitor cytochalasin D. The bacterial intake (and/or attachment) was estimated as the difference between number of 1 bacteria per macrophage (Mj) that are present in the supernatants of THP1 macrophage cultures exposed to LP (or PP) at a ratio of 1:25 for 0 and 2h.
F. Decrease in bacterial counts for LC, LP and PP after a 2h RPMI incubation in the absence (black) or presence (white) of the phagocytosis inhibitor cytochalasin D. The bacterial decrease was calculated as the difference between number of viable bacteria per mL after 0 and 2h of incubation and expressed as log 10 CFU/mL. Comparative analysis was carried out using the Student t-test (ns, no significant; ***p<0.005, ****p<0.001).