Bacillus Calmette-Guérin Induces PD-L1 Expression on Antigen-Presenting Cells via Autocrine and Paracrine Interleukin-STAT3 Circuits

Bacillus Calmette-Guérin (BCG) is the only licensed vaccine for tuberculosis (TB), and is also used as an immunotherapy for bladder cancer and other malignancies due to its immunostimulatory properties. Mycobacteria spp., however, are well known for their numerous immune evasion mechanisms that limit the true potential of their therapeutic use. One such major mechanism is the induction of programmed death ligand-1 (PD-L1), which mitigates adaptive immune responses. Here, we sought to unravel the molecular pathways behind PD-L1 up-regulation on antigen-presenting cells (APCs) by BCG. We found that infection of APCs with BCG induced PD-L1 up-regulation, but that this did not depend on direct infection, suggesting a soluble mediator for this effect. BCG induced potent quantities of IL-6 and IL-10, and the downstream transcription factor STAT3 was hyper-phosphorylated. Intracellular analyses revealed that levels of PD-L1 molecules were associated with the STAT3 phosphorylation state, suggesting a causal link. Neutralisation of the IL-6 or IL-10 cytokine receptors dampened STAT3 phosphorylation and BCG-mediated up-regulation of PD-L1 on APCs. Pharmacological inhibition of STAT3 achieved the same effect, confirming an autocrine-paracrine cytokine loop as a mechanism for BCG-mediated up-regulation of PD-L1. Finally, an in vivo immunisation model showed that BCG vaccination under PD-L1 blockade could enhance antigen-specific memory CD4 T-cell responses. These novel findings could lead to refinement of BCG as both a vaccine for infectious disease and as a cancer immunotherapy.

The correct balance of immune effector and regulatory responses depends on a number of molecular interactions between the antigen-presenting cell (APC) and T-cell. A key interaction for immunological tolerance is between the receptors programmed death-ligand 1 (PD-L1) and programmed death-1 (PD-1). APC expression of PD-L1 leads to binding of this molecule to PD-1 on T-cells, resulting in activation of the tyrosine phosphatase SHP-2 and dephosphorylation of critical kinases involved in T-cell receptor (TCR) signalling. Blockade of this interaction diminishes Treg frequencies 1 , enhances Th1 and Th17 effector cell frequencies 2 and increases cytokine production both in vitro and in vivo 3 . The PD-L1:PD-1 interaction has thus been targeted in immunological situations that feature restricted antigen presentation, T-cell anergy and immune tolerance as detrimental characteristicsnamely chronic infectious diseases and malignancies. In the latter scenario, clinical trials have demonstrated the remarkable efficacy of drugs developed to target these receptors, with up to 87% clinical response rates in some refractory cancers 4 .
Worldwide, tuberculosis (TB) is the leading cause of death due to infectious disease. The only vaccine available is Bacillus Calmette-Guérin (BCG), which shows only modest protection in adults and alarmingly low efficacy in developing countries, where TB mortality is highest. BCG (like its pathogenic relative, M. tuberculosis) can impede antigen presentation in vivo 5,6 , which is believed to contribute to the relatively low efficacy of the vaccine in humans. While mycobacteria-induced PD-L1 expression has been postulated as a major mechanism driving impaired antigen presentation 7,8 , it is currently not fully understood (i) the molecular mechanisms underpinning BCG-mediated PD-L1 up-regulation, and (ii) the immunological consequences of blocking this pathway during BCG immunisation.
For over 30 years, BCG has been employed as a front-line immunotherapy for bladder cancer 9 , and has been used since the late 1960s for malignant melanoma 10 . Although the mechanism of action remains to be completely elucidated, it is believed that the bacteria trigger an inflammatory response that leads to immune cell infiltration of the tumour site, thus facilitating immune-mediated clearance 11 . This is likely to be mediated by innate (i.e. Toll-like receptor) signalling, providing scope for improvement by concomitant engagement of the adaptive immune responses, which are known to be suppressed by PD-L1.
Here, we show for the first time that BCG can induce the up-regulation of PD-L1 on both macrophages and dendritic cells (DCs) via autocrine/paracrine secretion of STAT3-activating cytokines, chiefly IL-6 and IL-10. Blockade of the PD-L1 receptor in vivo during BCG immunisation led to superior CD4 T-cell responses to recall antigen, thus highlighting the potential utility of this pathway in clinical settings. These findings provide new targets for improving BCG as both a TB vaccine and cancer immunotherapy.

Materials and Methods
Ethics. All  Bacteria. BCG strain Pasteur was a kind gift from Professor Juraj Ivanyi (King's College, London) and was grown according to previous reports 12 , using standard microbiological techniques. BCG expressing green fluorescent protein (GFP; also from the Ivanyi laboratory) was grown in identical conditions, but under selective media and agar containing 50 μg/mL hygromycin B (Sigma-Aldrich).
To confirm receptor blockade, mice were administered 1 mg MIH5 or MAC219 via the i.p. route, followed by euthanasia at 24 h, and immediate ex vivo staining of the splenocytes. Cells were stained with a reported competing fluorochrome-conjugated α-PD-L1 clone (10F.9G2) 15 , which binds to the same epitope as MIH5, to test for successful receptor blockade (Fig. 1). As an additional control for specificity, PD-L2 was also stained after MIH5 or MAC219 treatments.
For BCG infections, bacteria were washed in complete media without antibiotics, and then APCs were inoculated at the designated MOI. Cells were cultured in a 5% CO 2 humidified incubator at 37 °C. In some experiments, E. coli-derived lipopolysaccharide (LPS; Sigma-Aldrich) was used at 100 ng/mL. Cytokines (Peprotech, UK) were www.nature.com/scientificreports www.nature.com/scientificreports/ diluted in complete RPMI before administration. Interleukin blocking antibodies (purchased from Biolegend, UK) were pre-cultured with the cells for 2 h before stimulation. Stattic (Tocris Bioscience, UK) was diluted in DMSO (Sigma-Aldrich) and cells were treated for 2 h before infection.
For lymph node analysis, inguinal lymph nodes were excised from euthanised mice on the indicated day, followed by mechanical disruption, counting and immediate flow cytometric analysis.
Flow cytometry. In most experiments, cells were first washed in PBS and then incubated with 1:1000 viability dye (eFluor780 Fixable Viability Dye; eBioscience) under Fc receptor blockade (1:500 TruStain; Biolegend) for 15-20 minutes. Cells were then washed in flow cytometry buffer (PBS (Invitrogen) containing 0.5% BSA and 0.1% sodium azide -both from Sigma-Aldrich) and stained with the appropriate pre-titrated flow cytometry antibodies for 30 m at 4 °C. Cells were sometimes fixed using Biolegend Fixative Buffer before being acquired on a BD FACSCanto II instrument and analysed using FlowJo software. For assessing phosphorylated residues, cells were instead mildly fixed with Fixative Buffer for 10 m at 37 °C, and then permeabilised with a commercial methanol buffer (True-Phos buffer -Biolegend) for 1 hour before staining, as described elsewhere 16 . For intracellular cytokine staining, cells were fixed with Fixative Buffer, followed by permeabilisation with flow cytometry buffer containing 0.5% saponin (Sigma-Aldrich). Compensation was performed using eBioscience UltraComp Enzyme-linked immunosorbent assay. Cytokine levels in supernatants were measured using eBioscience Ready-Set-Go ELISA kits, according to the manufacturer's instructions. Plates were read at 482 nm on a Tecan200 plate reader.

Statistical analysis. For all experiments, statistical tests were performed with Microsoft Excel and GraphPad
Prism software, using averaged technical replicates where possible. Each statistical test and post-test is detailed in the relevant figure legends. A p value less than 0.05 was considered significant.

BCG Induces Up-regulation of PD-L1 Expression on both Macrophages and Dendritic Cells.
DCs and macrophages are critical for the initiation of adaptive immunity. BCG can induce the up-regulation of PD-L1 expression on pulmonary DCs in mice 7 , however it is unclear whether the same holds true for macrophages. DCs and macrophages were therefore infected with BCG over a range of multiplicities of infection (MOI) and across two time-points (24 h and 48 h); surface expression of PD-L1 was assessed by flow cytometry. As shown in Fig. 2A, the positive control lipopolysaccharide (LPS) was able to significantly increase PD-L1 expression on both cell types (p < 0.0001), with a striking >10 fold increase in macrophages at 48 h. Upon infection with BCG, both APC types expressed high levels of PD-L1 compared to the unstimulated control at 24 h and 48 h (p < 0.0001), and a dose trend was observed for increasing MOI in macrophages at 48 h.
Next, a transgenic strain of BCG that expresses green fluorescent protein (GFP) was used to determine whether the up-regulation of PD-L1 depended on direct interaction between APCs and the bacteria. Cells were gated by GFP fluorescence into GFP neg (i.e. "bystander") and GFP pos (i.e. infected) populations. These were then tested for PD-L1 expression. As anticipated, GFP pos infected cells displayed increased expression of PD-L1 that was proportionate to the infectious dose (Fig. 2B). Surprisingly, however, GFP neg bystander cells also exhibited similar dose-dependent increases in PD-L1 expression to GFP pos cells, which was significantly up-regulated compared to the uninfected control at the highest MOI (p < 0.05), and approximately 80% that of directly infected cells. In support of these observations, 0.2 µm filtration of BCG infection supernatants, when applied to new cells, was able to up-regulate PD-L1 ( Supplementary Fig. 1). As expected, control supernatants from uninfected cells did not affect PD-L1 expression. These data strongly suggested that a soluble, secreted factor in culture was driving PD-L1 up-regulation by BCG.
Interestingly, BCG was able to up-regulate other members of the B7 family (CD80, CD86, PD-L2) to a certain degree, but with a pattern of skewed up-regulation of PD-L1 compared to CD86 (Supplementary Fig. 2).
BCG Induces IL-6 and IL-10 Production in tandem with STAT3 Phosphorylation. The murine PD-L1 gene (Cd274) is under the control of complex regulatory networks and can be induced by a number of inflammatory cytokines or TLR ligands 17 . Many of these control mechanisms are cell type-specific. Since mycobacteria are adept at driving STAT3 activation 18 , and since STAT3 is capable of binding to the PD-L1 promoter in tolerogenic DCs 19 , we hypothesised that this signalling pathway was mediating the observed effects of infection.
DCs and macrophages were infected for 18 h and intracellular flow cytometry was used to determine the levels of STAT3 tyrosine residue 705 phosphorylation. As can be observed in Fig. 3A, BCG was able to hyper-phosphorylate the STAT3 transcription factor compared to unstimulated cells (p < 0.001). Notably, there was a trend for increased phosphorylation when comparing BCG to the positive control LPS. We next hypothesised that APCs were producing STAT3-activating cytokines. STAT3 can be activated weakly by IL-2 and IL-12, www.nature.com/scientificreports www.nature.com/scientificreports/ but strongly by prototypical myeloid-type cytokines such as IL-6 and IL-10. APCs were therefore infected with BCG (or stimulated with LPS) for 24-48 h and levels of IL-6 and IL-10 were measured by ELISA. BCG was found to induce potent quantities of IL-6 in both DCs and macrophages at a range of MOI ( Fig. 3B; p < 0.0001 BCG vs unstimulated cells). BCG elicited IL-10 in both cell types. Strikingly, with regards to IL-10 production, BCG greatly surpassed the ability of LPS at MOI = 2, with over 4-fold concentrations of this cytokine compared to the positive control.

PD-L1 Expression Correlates with STAT3 Phosphorylation.
With the hypothesis that BCG was inducing STAT3-activating cytokines to up-regulate PD-L1, we treated DCs and macrophages with different concentrations of IL-6 and IL-10 and measured PD-L1 up-regulation by flow cytometry. In DCs (Fig. 4A), IL-6 played a dominant role in the up-regulation of PD-L1 expression compared to IL-10. When both cytokines were used in combination, there was only a marginal increase above the levels of PD-L1 expression conferred by IL-6. In macrophages, by contrast, IL-6 and IL-10 behaved similarly in terms of receptor up-regulation, and there was a combinatorial effect that was clearly evident at 500 pg/mL. For both cell types, there was a dose-dependent effect, with 500 pg/mL of any cytokine (or combination) being superior to 50 pg/mL in up-regulating PD-L1 expression.
Next, macrophages were infected with BCG for 18 h and cells were then permeabilised as before and co-stained for p-STAT3 and PD-L1. Cells were then divided into PD-L1 lo and PD-L1 hi populations and then www.nature.com/scientificreports www.nature.com/scientificreports/ levels of p-STAT3 were quantified. As shown in Fig. 4B, PD-L1 lo cells exhibited significantly lower levels of STAT3 phosphorylation compared to their PD-L1 hi counterparts (p < 0.05), with a doubling of fluorescence intensity in some experiments. Taken together, these data suggested that BCG was utilising interleukin signalling pathways in order to up-regulate expression of PD-L1.

IL-6R/IL-10R Blockade or STAT3 Inhibition Leads to Impeded PD-L1 Up-regulation by BCG.
We next questioned whether direct intervention in the interleukin-STAT3 axis could revert the up-regulation of PD-L1 by BCG. To this end, we employed a combination of monoclonal antibodies (mAbs) that are known to block the IL-6 and IL-10 cytokine receptors, alongside the pharmacological inhibitor 'Stattic': a well-characterised and highly specific small-molecule inhibitor of STAT3 transcriptional activity 20 .
To test whether neutralising the biological activities of endogenous cytokines could reduce the increase in PD-L1 expression caused by BCG, cells were first pre-incubated under IL-6R (mAb D7715A7) or IL-10R (mAb 1B1.3a) blockade, or a combination of both, for 2 h. Isotype controls served as controls for non-specific activity. Cells were then infected with BCG and after 24 h, PD-L1 expression was determined (Fig. 5A). Both IL-6 and IL-10 blockade significantly reduced PD-L1 expression (p < 0.001) compared to the isotype control baseline. As expected, the combination of blocking both receptors led to the greatest reduction in PD-L1 fluorescence (~40-45%; p < 0.0001 vs isotype control). In keeping with the hypothesised role of STAT3 in BCG-induced PD-L1 expression, blockade of both receptors also led to a large reduction in STAT3 phosphorylation, as shown in Fig. 5B (BCG + isotype MFI: 5416; BCG + dual blockade MFI: 2911).
To confirm that STAT3 was mediating the up-regulation of PD-L1 expression by BCG, cells were then pre-treated with Stattic or a vehicle control for 2 hours before infection with a low dose of bacteria (Fig. 5C). LPS served as a positive control, since it can induce PD-L1 expression via this pathway 21 . After 24 hours, the cells treated with 5 μM Stattic and infected with BCG showed a dramatically reduced expression level of PD-L1 compared to those treated with a vehicle control (p < 0.05), with an near-100% reduction in PD-L1 expression. Similar results were observed for LPS. Strikingly, resting cells treated with Stattic actually increased PD-L1 expression compared to the vehicle control, although this was a non-significant trend with high variation. Collectively, these data proved that BCG could modulate PD-L1 expression by interleukin-STAT3 signalling circuits.

In vivo PD-L1 Blockade Augments Antigen-Specific Memory CD4 T-cell Responses. A
proof-of-principle in vivo immunogenicity assay was next performed in order to establish that targeting the PD-L1 receptor during BCG immunisation could lead to increased T-cell function. A cytokine panel spanning www.nature.com/scientificreports www.nature.com/scientificreports/ several hallmark Th1/Th17 cytokines (IFN-γ, IL-2, IL-17A, TNF-α) was used to determine T-cell reactogenicity in a vaccination model with recall mycobacterial antigens (Fig. 6A).
First, the up-regulation of PD-L1 in vivo was determined by sub-cutaneous immunisation with BCG followed by characterisation of PD-L1 expression in the draining lymph nodes (Supplementary Fig. 3). It was found that BCG was able to up-regulate PD-L1 in MHC Class II high CD11c + DCs, but unable to do the same in T-cells, confirming a specific effect in APCs. Next, mice were treated with 1 mg of the PD-L1 blocking antibody MIH5 via the intraperitoneal route, followed by immunisation with subcutaneous BCG. Mice were then given follow-up booster immunisations to maintain the PD-L1 blockade. An isotype control was used at the same concentration to control for non-specific effects. On day 21, splenocytes from immunised mice were then pulsed with either immunodominant and latency-associated antigens (Ag85B and Acr, respectively) or a mixture of protein antigens (PPD). As shown in Fig. 6B, there was a general trend for an increase in antigen-specific cytokine production in CD4 T-cells. Under PD-L1 blockade, BCG induced significantly more IFN-γ in response to Ag85B/Acr and PPD (IFN-γ: p < 0.0001), compared to the isotype control. TNF-α production was also increased in response to Ag85B/Acr (p < 0.01). Regarding IL-2 and IL-17A, there were non-significant increases caused by PD-L1 blockade under both antigen recall conditions. Turning our attention to the CD8 T-cell compartment (Fig. 6C), we found that BCG was much poorer at inducing antigen-specific cytokine responses, as has been observed previously by others 22 . With the exception of IL-2 after Ag85B/Acr pulsing (p < 0.05), splenocytes from mice that received BCG + isotype control were unable to produce more cytokines than splenocytes from mock-immunised mice. Only two cytokines were found to be increased by PD-L1 blockade beyond the PBS control group in response to Ag85B/Acr but not PPD: IFN-γ (p < 0.01) and TNF-α (p < 0.01), however these effects were marginal. Thus we concluded that PD-L1 blockade can effectively boost CD4-dependent T-cell immunity, with only marginal effects on boosting CD8 T-cell responses.

Discussion
It is long-recognised that bacteria (and indeed other pathogens) belonging to genetically distinct phyla are capable of modulating the repertoire of co-stimulatory and co-inhibitory molecules expressed on the APC surface 23 , thus affirming the critical importance of said receptors in the directing of adaptive immune responses. Mycobacteria represent an example of immune evasion par excellence due to their ancient history of co-evolution with mammalian immune systems 24 ; indeed, virtually all pathogenic mycobacteria are obligate parasites, requiring intracellular www.nature.com/scientificreports www.nature.com/scientificreports/ resources in order to thrive and propagate. It is therefore not surprising that they are biologically equipped to effectively counter adaptive immune responses that would lead to their own clearance.
A centrally important molecule in the control of T-cell immunity is PD-L1. Mice infected with mycobacteria harbour pulmonary DCs that express high levels of PD-L1 and restrict antigen presentation to CD4 T-cells 25 , and PD-L1 blockade in blood and lung lavage from TB patients can enhance responses to Mtb antigens, as seen by greater cytokine production and T-cell proliferation 26,27 . Furthermore, PD-L1 blockade is able to rescue these cells from functional exhaustion, as demonstrated by the reversal of T-cell apoptosis 28 . For bladder cancer, the tumour surface is reportedly decorated with high levels of PD-L1 molecules, and the tumour-infiltrating B-and T-cells express high levels of both PD-L1 and PD-1 29,30 . The utilisation of BCG as a prophylactic (TB) or therapeutic (malignancy) treatment for these diseases, combined with a strategy to mitigate the effects of PD-L1, could provide a strong advantage for the efficacy of BCG.
In this study, for the first time, it was shown that BCG drives up-regulation of PD-L1 expression on APCs by autocrine/paracrine cytokine circuits that lead to STAT3 phosphorylation and up-regulation of this co-inhibitory receptor (illustrated in Fig. 7). It has been known for some time that BCG is a strong inducer of these cytokines via rudimentary TLR signalling, however it was not known that they were driving PD-L1 expression in a biphasic response. The observation that inhibition of STAT3 did not lead to a decrease in PD-L1 expression in resting cells-indeed, there was instead a non-significant increase in expression-is consistent with the notion that the PD-L1 gene promoter requires distinct transcriptional apparatus during the steady-state and during active infection. This is in accord with the fact that DCs and macrophages displayed moderate levels of constitutive PD-L1 expression in the absence of any appreciable cytokine levels 31 . Physiologically, this is undoubtedly to prevent spontaneous activation of the adaptive immune system.
Using the mAb MIH5, we were able to neutralise most of the PD-L1 receptor in vivo and augment the quantities of cytokine produced by CD4 T-cells in response to recall antigen in a proof-of-principle experiment. It is interesting to note that PD-L1 blockade in vivo did not greatly boost the antigen-specific CD8 T-cell responses beyond that of the mock-immunised control group. BCG is widely known to only minimally induce cytotoxic T-cell responses, with a strong bias towards CD4 responses 22 . This may be due to so-called CD8 'decoy antigens' , such as TB10.4, that serve to divert responses away from immunogenic epitopes for cytotoxic T-cells 32 . Whether the enhancement of CD4 T-cell immunogenicity in the absence of strong CD8 T-cell immunogenicity leads to better vaccine (i.e. in the case of TB) or immunotherapeutic (i.e. in the case of a carcinoma) outcomes warrants testing in disease-specific animal models.   www.nature.com/scientificreports www.nature.com/scientificreports/ STAT3-activating cytokines, (ii) IL-6/IL-10 cytokine receptors, (iii) the STAT3 transcription factor and (iv) the PD-L1 receptor itself. STAT3 presents a highly appealing therapeutic target other than its role in PD-L1 expression due to the fact that is a pleiotropic master controller of general tolerance in APCs. STAT3 can suppress autophagy 36 , nitric oxide production 37 , IL-12 production 38 , and co-stimulatory molecule expression 39 . Given our new findings, targeting this pathway during vaccination with BCG (for TB or malignancy) could reap multiple therapeutic benefits.
In conclusion, we have revealed novel molecular insights into how BCG up-regulates PD-L1 on APCs, allowing for improved immunogenicity to specific antigens, but also more intricate understanding of fundamental host-pathogen interactions. Future work will focus on exploring this pathway in specific disease models that rely on BCG as a treatment, with the aim of bolstering immunological parameters, and ultimately, treatment efficacy.