Granulocytic myeloid-derived suppressor cells increase infection risk via the IDO/IL-10 pathway in patients with hepatitis B virus-related liver failure

Hepatitis B virus-related acute-on-chronic liver failure (HBV-ACLF) results in high susceptibility to infection. Although granulocytic myeloid-derived suppressor cells (gMDSC) are elevated in patients with HBV-ACLF, their role in HBV-ACLF pathogenesis is unknown. To elucidate the mechanism of gMDSC expansion and susceptibility to infection in HBV-ACLF patients, we analyzed the proportion of gMDSC in the peripheral blood and organ tissues of patients with HBV-ACLF and an ACLF mouse model established by continuous injection (eight times) of Concanavalin by flow cytometry and immunohistochemistry. We found that the proportion of gMDSC increased significantly in the blood and liver of patients with HBV-ACLF. This increase was positively correlated with disease severity, prognosis, and infection. gMDSC percentages were higher in peripheral blood, liver, spleen, and bone marrow than control levels in the ACLF mouse model. Immunofluorescence revealed that the gMDSC count increased in the liver of patients with HBV-ACLF as well as in the liver and spleen of ACLF mice. We further exposed peripheral blood monocyte cells from healthy donors to plasma from HBV-ACLF patients, recombinant cytokines, or their inhibitor, and found that TNF-α led to gMDSC expansion and significant upregulation of indoleamine 2, 3-dioxygenase (IDO), while blocking TNF-α signaling decreased gMDSC. Moreover, we detected proliferation and cytokine secretion of T lymphocytes when purified gMDSC was co-cultured with Pan T cells or IDO inhibitor and found that TNF-α-induced gMDSC inhibited T cell proliferation and interferon-γ production through the IDO signaling pathway. Lastly, the ability of gMDSC to phagocytose bacteria was low in patients with HBV-ACLF. Our findings elucidate HBV-ACLF pathogenesis and provide potential therapeutic targets.


Introduction
Acute-on-chronic liver failure (ACLF) is characterized by deep jaundice, coagulation dysfunction, and extrahepatic organ failure (1). In China, hepatitis B virus (HBV) infection accounts for approximately 87% of ACLF cases (2). Further infection from other pathogens is extremely common among ACLF patients; 81.2% develop bacterial or fungal infections during hospitalization, and 26.6% develop secondary infections. Infection significantly increases 90-d mortality in patients with ACLF. However, the exact mechanism of infection has not been fully elucidated.
Recent studies have shown that elevated immune checkpoint molecules and immunoregulatory cells may increase susceptibility to infection. For example, a marked expansion of MERTK-expressing monocytes and macrophages has been observed in ACLF patients, along with an increased number of CTLA-4 + CD4 + T cells; these changes suppress host immune response to microbes (3,4). Additionally, immunosuppressive monocytic myeloid-derived suppressor cells (mMDSC) expanded in patients with ACLF, attenuating antimicrobial innate immune responses and impairing bacterial uptake and clearance (5). These outcomes are probably related to the downregulation of TLR-3 signaling that then impairs the TLRdriven innate immune response. In vitro studies have shown that the TLR-3 agonist polyI:C significantly reduced the proportion of mMDSC while simultaneously enhancing their phagocytic capacity in ACLF (5).
MDSCs are a heterogeneous population of immature myeloid precursor cells, classified as mMDSC or granulocytic MDSCs (gMDSC) based on phenotypic and functional characteristics (6). Granulocytic MDSC suppress innate and adaptive immune responses through depleting L-arginine and generating IL-10 (7, 8). Accumulating evidence indicates that gMDSC play an important role in the development of microbial infections, autoimmune disorders, and cancer. The gMDSC population is elevated in patients with alcoholic liver disease and closely associated with disease progression (9). However, we know little regarding whether gMDSC have impaired antimicrobial response or phagocytic capacity in patients with HBV-ACLF.
This study thus analyzed gMDSC proportion and phenotypes to understand their relationship with infection complications, severity, and prognosis. We also explored the mechanisms underlying gMDSC regulation of antimicrobial response and pathogen-clearing ability in patients with HBV-ACLF.

Patients and sampling
From April 2016 to December 2017, we consecutively recruited 152 patients with CHB and 82 patients with HBV-ACLF within 24 h of admission to the Huashan Hospital and the First Hospital of Quanzhou. Healthy individuals without apparent disease (n = 40) were enrolled from Huashan Hospital as healthy controls (HC). Patients with CHB fulfilled the following established diagnostic criteria: positive hepatitis B surface antigen (HBsAg) detection for over 6 months, hepatitisrelated clinical manifestations, histological confirmation of hepatitis, and abnormal ALT levels (≥40 U/L) (10). Criteria for HBV-ACLF patients were as follows: history of CHB or liver cirrhosis, serum TBil over five times the upper limit of normal levels (5 mg/dL), INR ≥1.5, or prothrombin time activity <40% (11). Patients were excluded if they were co-infected with other viruses (e.g., HAV, HCV, HDV, and HIV), consumed alcohol, had drug-induced liver diseases, were under treatment with antivirals or immunomodulating agents, or exhibited bleeding/ infection. The study protocol was approved by the Ethics Committee of Huashan Hospital affiliated with Fudan University and the First Hospital of Quanzhou affiliated with Fujian Medical University.

Animal model
Two C57BL/6 mice (8 weeks old, weighing 20 ± 1 g) were purchased from B&K Universal Group Ltd. (Shanghai, China). Animals were bred in a specific pathogen-free barrier facility. All experiments were approved by the Ethics Committee of the Shanghai Public Health Clinical Center and Institutes of Biomedical Science, Shanghai Medical College, Fudan University. For the ACLF model, ConA (8 mg/kg, Sigma-Aldrich, USA) in 0.9% saline was injected into the retrobulbar angular vein five times every 2 days. Peripheral blood and liver tissues were sampled on the same days as the injections. Mice were euthanized after the last injection.

Immunofluorescence double-staining
Tissue samples were deparaffinized with xylene and ethyl alcohol, quickly rinsed with TBST buffer for 5 min, and blocked with 5% normal serum TBST for 1 h. Liver sections from HBV-ACLF patients were incubated with mouse monoclonal anti-CD15 (BioLengend, USA) and rabbit monoclonal anti-CD11b (Abcam, USA) antibodies at 4°C overnight. Liver and spleen sections from ACLF mice were incubated with rabbit monoclonal anti-Ly-6G (Servicebio, China), rabbit anti-CD11b (Servicebio, China), or rabbit anti-IDO (Biosynthesis Biotechnology Co., LTD, Beijing, China). Subsequently, all sections were incubated with their corresponding Alexa 488 and Alexa 555 antibodies (Jackson ImmunoResearch, West Grove, PA, USA), and then incubated again with DAPI (Sigma, CA, USA). The results were analyzed using an inverted Eclipse Ti-S microscope (Nikon, Japan).

Chromatin immunoprecipitation
The analysis was performed using the ChIP Assay Kit (Upstate, Lake Placid, NY, USA). Briefly, 293t cells were sonicated until DNA fragments averaged between 200 and 1000 bp. Chromatin was subsequently immunoprecipitated with antibodies (2 mg) against IDO, and an equal amount of IgG was used as a negative control for nonspecific immunoprecipitation. Next, ChIP DNA fragments were purified, reverse-transcribed, and used as templates for PCR. The PCR products were analyzed using agarose gel electrophoresis. Antibodies used were anti-IDO antibody (Abcam) and normal rabbit IgG (Santa Cruz). Primers specific to the IDO region were as follows: F, TCTCGGGCTCAAGCAATTC; R, TTCCGTTTAT CCAGTCATCTC.

Whole-transcriptome library preparation and sequencing
Total RNA of gMDSC (CD14 − CD15 + ) from HC and HBV-ACLF patients was extracted using the miRNeasy Mini Kit (Qiagen), then purified using the RNA Clean XP Kit (Beckman Coulter, Inc. CA, USA). Strand-specific sequencing libraries were generated using Superscript II Reverse Transcriptase (Invitrogen, USA). An Illumina HiSeq 2000 platform was used for RNA-seq.

Whole blood phagocytosis assay
The phagocytosis assay was performed using the pHrodo Escherichia coli Green BioParticle Phagocytosis Kit (Invitrogen, Paisley, UK). Whole blood (100 mL) was added to 20 mL pHrodo E. coli and incubated for 15 min at 37°C under dark conditions. Next, 1 mL each of anti-human PE CD33 (eBscience), FITC CD11b (eBscience), APC HLA-DR (eBscience), PE/Cy7 CD14 (eBscience), and BV421 CD15 (eBscience) were added. The mixture was incubated again for 30 min at 4°C before 2 mL erythrocyte lysate (Invitrogen, Paisley, UK) was added. After a third incubation for 8 min at room temperature, cells were washed twice with PBS. Cells were acquired using a flow cytometer (CytoFLEX S, Beckman, USA).

Statistical analyses
Data are expressed as means ± SEM. Significant differences were tested using the Mann-Whitney U, Wilcoxon, or Kruskal-Wallis tests, or one-way ANOVA, as appropriate. Correlation analysis was performed using Spearman's correlation coefficients. Graphs were drawn in GraphPad Prism 8.0 (San Diego, California, USA).

Patient characteristics
Patients with HBV-ACLF were older than those with chronic hepatitis B (CHB) and HC (for clinical characteristics, see Table 1). Moreover, patients with HBV-ACLF showed significantly higher physiological and biochemical indicators of liver injury (total bilirubin, TBil; alanine aminotransferase, ALT; aspartate aminotransferase, AST; alkaline phosphatase, ALP; g-Glutamyl transferase, GGT; prothrombin time, PT; and international normalized ratio, INR) than patients with CHB, but markedly lower compensatory indices of liver function (ALB, Hgb, and PLT) and virological parameters (HBV DNA). In addition, 60.38% of patients with HBV-ACLF developed bacterial infections during their hospital stay, and 49.09% of patients with HBV-ACLF showed a poor prognosis.
The proportion of gMDSC was correlated with disease progression, prognosis, and infectious complications in patients with HBV-ACLF After assessing the relationship with clinical parameters, we found that the gMDSC proportion was positively correlated with liver injury (TBil, ALT, and INR, Figure 1G), prognosis (Child-Pugh and MELD scores, Figure 1H), and inflammation (PCT and CRP, Figure 1I).
Next, we tested the association between gMDSC and infectious complications. We found that the gMDSC proportion was higher in co-infected patients than in noninfected patients ( Figure 1J). Furthermore, the poor prognosis group had more gMDSC than the good prognosis group ( Figure 1K). Details of the two groups are described in one of our previous studies (12). After comprehensive treatment, gMDSC proportions in patients with HBV-ACLF gradually increased over 4 weeks, then fell rapidly and remained low for a long time. Importantly, this decrease was driven by patients with HBV-ACLF with a good prognosis, as we did not observe a significant change in patients with HBV-ACLF with a poor prognosis at 4 weeks ( Figure 1L). These findings further suggest that gMDSC are likely to reflect impaired immunity against bacterial infection and may be a new predictor for disease prognosis or co-infection.
Next, we used flow cytometry to test BTLA and LIGHT expression for further verification of gMDSC phenotypic characteristics. The results showed that gMDSC in patients with HBV-ACLF had a significantly lower BTLA expression than HC cells, but a significantly higher LIGHT expression than gMDSC in HC and patients with CHB ( Figures 2C, E, S1C, D). Additionally, gMDSC proportions among PBMCs were positively correlated with LIGHT expression (r = 0.329, P = 0.002, Figure 2F) and negatively correlated with BTLA expression (r = -0.272, P = 0.013, Figure 2D). LIGHT and BTLA are positive and negative immune checkpoint molecules, respectively. gMDSC of HBV-ACLF patients showed high expression of LIGHT molecules and low expression of BTLA molecules, suggesting that the gMDSC phenotype is immunoenhanced, rather than immunoparalytic.

Expansion of gMDSC was dependent on TNF-a
Because the inflammatory microenvironment may support MDSC expansion (13), we investigated the effect of HBV-ACLF plasma on gMDSC. Exposure of PBMCs to HBV-ACLF plasma resulted in a higher gMDSC to PBMC ratio than exposure to HC plasma ( Figure 3A).

Granulocytic MDSCs inhibit cytokine secretion and T cell proliferation via IDO/ IL-10 pathway
After incubating HC PBMCs with or without TNF-a for 3 days, we separated gMDSC using magnetic beads and cocultured them with autologous Pan T cells. The isolated gMDSC (with or without TNF-a) significantly decreased intracellular cytokine production (IFN-g, IL-2, and TNF-a), and decreased CD4 + and CD8 + T cell proliferation. TNF-ainduced gMDSC had a slightly stronger inhibitory effect than non-TNF-a-induced gMDSC (Figures 4A, B). Next, to assess whether gMDSC impair T cell functions under pathological conditions, we co-cultured HC Pan T cells with or without gMDSC from patients with HBV-ACLF, and then analyzed cytokine secretion and proliferation of Pan T cells. The results showed that gMDSC from patients with HBV-ACLF significantly inhibited intracellular IFN-g production in CD4 + / 8 + T cells (Figures 4C, D) and markedly decreased T cell proliferation in a dose-dependent manner ( Figure 4E).
Expansion of gMDSC was dependent on TNF-a. (A) Cumulative dot plots of gMDSC proportion when PBMCs were exposed to plasma from HC or HBV-ACLF patients. (B) Histogram of gMDSC proportion when PBMCs were exposed to different concentrations of recombinant human (rh) cytokines (TNF-a, IL-6, IL-1b, IL-22, IL-37, IL-10, and FGF2). (C) Histogram of gMDSC proportion after rhTNF-a stimulated HC PBMCs for 1, 3, 5, and 7 days (n = 3). (D, E) Cumulative dot plots of gMDSC proportion when HC PBMCs were co-cultured with rhTNF-a, rhIL-6, anti-TNF-a, NFkb inhibitor, and Stat3 inhibitor for 3 days. (F) TNF signaling was the most active pathway in gMDSC purified from HBV-ACLF. Error bars, mean ± SEM; Wilcoxon test (A-E); *P <.05, **P <.01, ***P <.001, and ****P <.0001. Previous studies have shown that several factors, including Arg1 and IL-10, are responsible for gMDSC suppression of T cells (6,14). Our previous research has suggested that mMDSC impair T cell function through the IDO pathway (15). Therefore, we aimed to identify whether the IDO and IL-10 pathways are responsible for gMDSC suppression of T cell function. The results of ELISA showed that patients with HBV-ACLF had higher soluble IDO (sIDO) and IL-10 levels than HC and patients with CHB ( Figures 4F, S2A, B). Additionally, sIDO levels were positively correlated with gMDSC proportion, Child-Pugh scores, MELD scores, INR, and TBil levels, but negatively correlated with PLT levels (Figures S3A, B, C). These patterns are consistent with gMDSC expression trends and correlations.
Intracellular staining then confirmed that IDO and IL-10 expression in gMDSC was significantly higher in patients with HBV-ACLF than in HC and patients with CHB ( Figures 4G,  S2C). In addition, Pan T cells co-cultured with gMDSC from patients with HBV-ACLF had higher IDO and IL-10 expression than Pan T cells alone ( Figures 4H, S2A, D). Furthermore, IDO inhibitor 1 efficiently restored CD4 + and CD8 + T cell proliferation in a co-culture of gMDSC with autologous T cells ( Figure 4I).
We analyzed JASPAR datasets and conducted a ChIP assay to investigate whether NF-kb, a downstream target of TNF-a, regulates IDO expression through binding to its promoter region. The results of ChIP-qPCR confirmed that NF-kb binds to a site in the IDO promoter region located at approximately -609 to -625 ( Figure 4J). Thus, gMDSC seem to impair T cell function in an IDO-dependent manner.

Increased gMDSC and IDO + gMDSC proportions in ACLF mouse model
Previous studies have shown that an ACLF animal model can be successfully established using continuously repeated concanavalin A (ConA) stimulation (16). Using the same method, we successfully established an ACLF mouse model with continuous low-dose ConA (8 mg/kg) exposure ( Figure 5A). On day 10, H&E staining of the liver and spleen showed sub-massive or massive tissue necrosis, accompanied by many inflammatory cell immersions, suggesting that the ACLF model was successfully established ( Figure 5B). We divided murine MDSC into CD11b + Ly-6G − Ly-6C + mMDSC, and CD11b + Ly-6G + Ly-6C -gMDSC ( Figure 5C) (17). In the ACLF model, gMDSC increased gradually, peaked on day 5, and then gradually decreased, but were still higher than those in the control group ( Figure 5D). On day 4, ACLF mice had a significantly higher gMDSC proportion in bone marrow, peripheral blood, liver, and spleen than the control group. We also found that bone marrow had the highest proportion of gMDSC, followed by peripheral blood, liver, and spleen ( Figures 5E, F). Additionally, immunofluorescence showed that the number of double-positive CD11b + Ly-6G + cells was higher in the liver and spleen of the ACLF group than that of the control group ( Figure 5G). Another immunofluorescence assay then indicated that the number of double-positive Ly-6G + IDO + cells was higher in the liver and spleen of ACLF mice than that of control mice ( Figure 5H), indicating increased IDO expression in ACLF gMDSC.
Granulocytic MDSCs from patients with HBV-ACLF display impaired phagocytosis of E. coli Previous studies have shown that mMDSC of patients with ACLF exhibit a marked and persistent deficiency in bacterial uptake and clearance (5). We thus performed a phagocytosis assay using E. coli to test whether gMDSC from patients with HBV-ACLF displayed the same problem. Patients with HBV-ACLF had significantly fewer gMDSC with phagocytic ability than HC and patients with CHB ( Figures 6A, B). Furthermore, the capacity of gMDSC in patients with HBV-ACLF had a consistently lower capacity to phagocytose E. coli than HC gMDSC ( Figure 6C).

Discussion
Hospitalized patients with HBV-ACLF are prone to secondary bacterial or fungal infections that significantly increase mortality risk. In this study, we successfully clarified aspects of the molecular mechanisms underlying susceptibility to infection associated with HBV-ACLF. Specifically, we found that gMDSC proportions in the peripheral blood and liver of patients with HBV-ACLF increased significantly. Moreover, the increase was positively correlated with disease severity, prognosis, and infection complications. Through acting on the IDO/IL-10 pathway, gMDSC inhibited T cell proliferation and IFN-g production, attenuating antimicrobial response. Furthermore, the ability of gMDSC to phagocytose bacteria was impaired in patients with HBV-ACLF. These two factors are likely the primary reasons behind the increased susceptibility to infection among this cohort.
In chronic hepatitis B, HBsAg or HBeAg maintain persistent HBV infection and inhibit T cell response through promoting monocyte differentiation into mMDSC (15,18). Additionally, hepatic stellate cell-induced gMDSC constrain nutrient supply to proliferating T cells as a means of moderating liver damage (6). Similar to our results on gMDSC from both patients and an ACLF mouse model, previous research also found that mMDSC expanded in HBV-ACLF (or ACLF) patients and were closely associated with disease severity and progression (5,19), but impairment of their phagocytic capacity increased infection risk (5). We demonstrated that immune-enhanced functional phenotypes of gMDSC resulted in their elevation, being associated with a higher risk of infection and poor prognosis. Our findings indicate that gMDSC may be useful as biomarkers of infection in patients with HBV-ACLF.
Although TNF-a signaling promotes MDSC migration and differentiation in general (20-22), it remains unknown whether TNF-a induces gMDSC in the inflammatory microenvironment of HBV-ACLF. Many previous studies and our unpublished data have confirmed that patients with HBV-ACLF display markedly increased TNF-a levels (23, 24). Here, we provide further evidence that TNF-a and other inflammatory cytokines promote gMDSC expansion. We also show that circulating and TNF-a-induced gMDSC display immunosuppressive functions. Because of its involvement in inflammation and peripheral tolerance, IDO is commonly used as a therapeutic target for cancers and autoimmune diseases (25). Our previous study showed that mMDSC act on the IDO signaling pathway to suppress T cell responses and maintain persistent HBV infection (15). Here, we found that gMDSC IDO levels and IDO expression increased synchronously in the peripheral blood and liver of HBV-ACLF patients, as well as in the liver of ACLF mice. Elevated IDOs in all these cases were positively correlated with HBV-ACLF severity and prognosis. Additionally, exposure to an IDO inhibitor restored T cell proliferation in gMDSC isolated from HBV-ACLF patients, suggesting that gMDSC act through IDO to dampen the antimicrobial response. More importantly, this study is one of the first to empirically demonstrate that TNF-a induces IDO production in addition to promoting gMDSC expansion.
Another factor contributing to increased infection risk in patients with HBV-ACLF is decreased pathogen-clearing ability. Previous studies have shown that the phagocytic capacity of mMDSC decreases continuously or is lost in ACLF (5). Experiments with RNA-seq have revealed that gMDSC are a mature type of neutrophils, potent phagocytes similar to monocytes (26). Here, we confirmed that gMDSC phagocytic ability decreased persistently in HBV-ACLF patients, likely due to the presence of multiple defective PRRs.
This study had some limitations. First, no well-established method exists for inducing ACLF, especially in the ACLF mouse model with a background of HBV infection. Second, while we found that elevated gMDSC may indicate an increased risk of infection and death, the lack of accurate biomarkers or antibodies meant that we could not target gMDSC for removal. Thus, it was impossible to obtain direct evidence of whether gMDSC clearance reduces mortality and infection rates.
In conclusion, inflammatory cytokines, especially TNF-a, induced gMDSC expansion and promoted IDO production, thereby inhibiting the antimicrobial immune response in HBV-ACLF. Although the proportion of gMDSC rose in HBV-ACLF patients, their phagocytic capacity was impaired. These two deficiencies explain why patients with HBV-ACLF are susceptible to infection and provide possible targets for immunotherapy.

Data availability statement
The data presented in the study are deposited in the CNSA repository (https://db.cngb.org/cnsa/), accession number CNP0003467.

Ethics statement
The studies involving human participants were reviewed and approved by Ethics Committee of Huashan Hospital affiliated with Fudan University and the First Hospital of Quanzhou affiliated with Fujian Medical University. The patients/participants provided their written informed consent to participate in this study. The animal study was reviewed and approved by Ethics Committee of Huashan Hospital affiliated with Fudan University. Written informed consent was obtained from the owners for the participation of their animals in this study. their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.