Human peripheral blood cells (PBMCs) were obtained after informed consent from patients or legal representatives. Patient consent guidelines, isolation and characterization of human immune cells and the use of clinical data was approved by the internal review board of the ethics committee of the Jena University Hospital. ACLF cohort: Ethics vote no. 3683-02/3. Sepsis cohorts: Ethics votes nos. 2160-11/07, 2712-12/09 and 3824-11/12. All involved procedures were in accordance with the principles of the Declaration of Helsinki (49). The clinical characteristics of sepsis patients enrolled for the experiments shown in Figures 1A–D and Figures 1E, F, H, I, J are summarized in Supplementary Tables 1 , 2 , respectively. The clinical data of patients enrolled for Figure 1G are presented separately for Gram-negative ( Supplementary Table 2 ) and Gram-positive sepsis ( Supplementary Table 3 ). All individuals with sepsis fulfilled the Sepsis-3 definition. First blood sample was drawn within 24 h after sepsis symptoms emerged. Patients for cohorts of acute decompensation of chronic liver disease and suspected bacterial infection were recruited from hospitalized patients at boards of the University Hospital, Jena, Germany. Baseline characteristics and outcome of these patients with decompensated liver disease in the absence or presence of multiple organ failure syndrome (according to the EASL CLIF-C criteria for acute-on-chronic liver failure) are given in Supplementary Table 4 . Clinical scores such as model for end-stage liver disease scores (MELD), bacterial culture count, protein analysis, blood count and serum levels of C-reactive protein and creatinine were obtained from routine laboratory analysis.

PBMCs were isolated from whole blood of patients and healthy volunteers by density gradient centrifugation using Biocoll (Cat#L6715,Merck) according to the manufacturer’s protocol. PBMCs and CD14⁺ monocytes of patients with acute decompensation of liver cirrhosis with ascites were purified using positive magnetic-activated cell sorting (Miltenyi Biotec, Germany) and stored at -80°C prior to use. Monocytes derived from whole blood of patients with sepsis ( Supplementary Table 1 ) were isolated by density gradient centrifugation using Biocoll, washed in PBS and cultured in DMEM supplemented with 10 µg/ml ciprofloxacin and 10% human serum overnight. Next day, cells were washed twice with pre-warmed PBS and stimulated or not with 1 µg/ml LPS (Cat#tlrl-3pelps; InvivoGen) for 72 h. Supernatant and cells were harvested separately at different time points as indicated. Monocytes of healthy volunteers were differentiated into macrophages by culturing in DMEM medium supplemented with 10% FCS, 10 µg/ml ciprofloxacin and 10 ng/ml recombinant human M-CSF (Cat#574806, Biolegend) for 5 days. On day 6 cells were washed and stimulated with 50 ng/ml LPS alone or in combination with 100 ng/ml IFN-γ (Cat#570206, Biolegend) from 8 to 48 hours. Cells were detached using Trypsin-EDTA 0.25% (ThermoFisher) for FACS analysis or processed for real time PCR as described below. Supernatants were stored at -80 C° upon harvest for TNF measurements. To activate the non-canonical inflammasome, MDMs treated or not with LPS, IFN-γ or LPS+IFN-γ for 48h were transfected with LPS using lipofectamine-2000 (Cat#11668, Thermo Fisher) for 18h as previously described (50).

Total RNA was extracted from samples using miRNeasy Mini Kit (Cat#217084, Qiagen). For reverse transcription of miRNAs to cDNA, the Universal cDNA Synthesis Kit (Cat#339340, Qiagen) was used. Real time PCR of miRNA was performed by the miRCURY LNA miRNA PCR Assay (Cat#339345, Qiagen) using locked nucleic acid primers (Qiagen). Baseline expression of the specific miRNA genes was normalized to small nuclear RNA U6 as an internal control gene. For mRNA-expression, approximately 0.5 μg of RNA was reverse transcribed using the RevertAid First Strand cDNA Synthesis Kit (Cat#K1621, Thermo Fisher). Genes of interest were amplified with specific primers using SybrGreen kit (Cat#331416, Biozym) according to the manufacturer’s instructions. mRNA expression profiling was performed using specific primer pairs and normalized to GAPDH as a reference gene. The primers used are given in Supplementary Table 5 .

TNF was measured using the human TNF ELISA Set (Cat#DY210-05, R&D Systems) according to the manufacturer’s instructions. For analysis of pyroptotic cell death, lactate dehydrogenase (LDH) release was measured in cell-free culture supernatants by LDH-assay (Cat#MK401, Takara Clontech) according to manufacturer´s instructions.

Protein concentration in serum and cell lysates was determined via BCA assay. Isolation of protein from supernatants and cell lysates and Western blotting was performed as previously described (50) using the following antibodies: MyD88 (E-11) Maus mAb, (Cat#sc-74532, SantaCruz), caspase-4 Rabbit pAb (Cat#4450; Cell Signaling Technology), IRF-1 (D5E4) XP^® Rabbit mAb (Cat#8478, Cell Signaling Technology), β-Actin (13E5) Rabbit mAb (HRP Conjugate) (Cat#5125, Cell Signaling Technology). GSDMD Rabbit pAb (Cat#20770-1-AP, Proteintech) was used to detect full-length GSDMD and p20-GSDMD in serum; GSDMDC1 Rabbit pAb (Cat#NBP2-33422, Novus Biologicals) was used to detect full-length GSDMD and p30-GSDMD in cell supernatants and lysates. Goat anti-mouse IgG (H+L)-HRP (Cat#074-1806, KPL), Goat anti-rabbit IgG (H+L)-HRP (Cat#074-1506, KPL). Proteins were visualized using ECL-substrate (Cat#541004, Biozym) in a GBOX-Chemi-XX6 gel documentation system (Syngene). Acquisition and densitometric analysis of relative fold differences in caspase-4 protein expression was performed using ImageJ software. Band intensity was determined in the caspase-4 band (~45 kDa) or GSDMD (~20kDa) as region of interest and expressed in arbitrary units.

For surface marker staining of MDM, cells were blocked with FcR blocking reagent (Cat#130-059-901, Miltenyi Biotec) for 20min, then incubated with FITC-anti-human HLA-DR, Mouse IgG2a antibody, clone L243 (Cat#307604, Biolegend) or FITC- mouse IgG2a isotype control clone MOPC-173 (Cat#400209, Biolegend) for 1h. Samples were washed extensively with sterile FACS buffer (PBS, 2% FCS, 2 mM EDTA) then analyzed using CytExpert (Beckman). Cells were gated on singlets. Data were analyzed using FlowJo software (BD).

400 ng genomic DNA from PBMCs was converted into bisulfite-treated bsDNA using EZ DNA Methylation-Gold™ (Cat#D5005, Zymo Research) according to the instruction manual. Identically treated unmethylated DNA (Cat#59665, Qiagen) served as negative control. Positive control sample was created by in-vitro methylation using the CpG-Methyltransferase (M.SssI) (Cat#M0226S, New England Biolabs). A serial dilution using in vitro methylated and unmethylated control bsDNA was created to generate standard samples of 0%, 3%, 6%, 12%, 24% and 100% methylation level. Samples and standards were amplified with bisulfite-specific primers (F: TTTTTTGATAATGAGTTTGGAAT; R: ACCTACCATAAAAAACAACCTC) targeting a region of the CASP4 gene (Chr11:104968517-104968678). The PCR was conducted using the SybrGreen MasterMix (Cat#4913914001, Roche) with 0.5 µM of each primer and 25 ng bsDNA in a Mastercycler Gradient (Eppendorf). After clean-up with the innuPREP DOUBLEpure kit (Cat#845-KS-5050250, Analytik Jena), PCR products were sequenced by Sanger sequencing (Microsynth). Sequences were analyzed by BLAST alignments against the unmethylated reference sequence (NCBI) and by visual inspection of sequence profiles.

Statistical analysis was performed using GraphPad prism version 6.0 (GraphPad Software, La Jolla, CA, USA) as indicated in the respective figure legends. Differences between cirrhosis patients with and without ACLF were analyzed by unpaired two-tailed t-test for parametric after Log2 transformation to reach normal distribution. Two-tailed Mann–Whitney U test was used for comparing unpaired, nonparametric data. Horizontal lines represent mean values. Correlations were analyzed using Pearson´s test representing r² or linear regression. The values obtained for each individual donor were paired and a paired two-tailed t-test was used. P values are from two-tailed Pearson´s correlation of parametric regression analysis. Each data marker represents an individual patient. Data generated from in vitro experiments are presented as mean ± SEM. For all tests, P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 was considered significant or ns as not significant.

Patients with sepsis often exhibit signs of immunosuppression, including a dampened response of monocytes to LPS, a condition termed ET (51, 52). To understand the role played by the non-canonical inflammasome in this context, we examined CASP4 and CASP5 expression in monocytes from patients with sepsis and organ damage who did or did not exhibit ET. Supplementary Table 1 shows the clinical characteristics of the enrolled patients. Consistent with previous reports (48, 51–53), some patients exhibited diminished LPS-induced tumor necrosis factor (TNF) release from monocytes, the hallmark of ET ( Figure 1A ). We refer to this state as `tolerized´ from here on. We have previously shown that LPS-dependent induction of the microRNA miR-222 discriminates between tolerized and non-tolerized patients (6). As shown in Figure 1B , the pattern of LPS-dependent miR-222 induction matched the TNF secretion profile, as monocytes from the tolerized patients exhibited higher miR-222 levels. Besides directly binding and activating CASP4/CASP5, LPS also induces expression of CASP4 and CASP5 in monocytes, a crucial priming step of the inflammasome (54). Remarkably, we observed a dichotomous response of CASP4 to chronic LPS exposure, as LPS down-regulated CASP4 in monocytes of tolerized patients while it upregulated CASP4 in the non-tolerized patients ( Figure 1C ). In contrast, LPS stimulated CASP5 expression in all patients ( Figure 1D ). In line with this transcriptional profile, caspase-4 protein levels in serum from patients with Gram-negative sepsis ( Supplementary Table 2 ) inversely correlated with disease severity SOFA score ( Figures 1E, H ). We concluded that CASP4 and CASP5 were differentially regulated during the acquisition of a state of immunosuppression and that caspase-4 down-regulation was specifically associated to disease severity and immunosuppression. Activation of inflammatory caspases-4/5 results in cleavage of GSDMD into N-terminal (p30) and C-terminal (p20) fragments to induce pyroptotic cell death. Interestingly, patients with sepsis showed a prominent accumulation of the p20-GSDMD form ( Figure 1E ). To ascertain specificity of the detected p20 band, serum samples from healthy donors were loaded next to a pool of sepsis patients. GSDMD (p20) was detected only in patients, but not in healthy donors ( Figure 1F ). Furthermore, a cut-off value of ≥ 8 SOFA score was chosen to distinguish patients with a high disease severity. As shown in Figure 1G , p20-GSDMD levels were significantly higher in patients with a higher SOFA score. In addition, p20-GSDMD was negatively associated with caspase-4 expression ( Figure 1I ), suggesting less activity/expression of caspase-4 during immunosuppression-associated inflammasome activation and pyroptosis. In contrast, levels of the pro-GSDMD precursor form correlate positively with caspase-4 ( Figure 1J ), consistent with the notion that GSDMD and CASP4 are jointly regulated by the transcription factors IRF1 and IRF2 (39, 40).

To understand if differential regulation of CASP4 versus CASP5 was specific to Gram-negative sepsis or general to severe infections with immunosuppression, we studied a cohort of patients with liver cirrhosis and acute decompensation and ascites. Patients were stratified for presence of acute-on-chronic liver failure (ACLF) according to the criteria of the European Foundation for the study of chronic liver failure (47) ( Supplementary Table 4 ). ACLF is an inflammatory syndrome of multiple organ failure (MOF) in patients with pre-existing liver disease often precipitated by inflammatory events such as bacterial infections and sharing many characteristics with sepsis (55). mRNA analysis of Peripheral Blood Mononuclear Cells (PBMCs) revealed that CASP4 gene expression was suppressed specifically in patients with ACLF ( Figure 2A ), while CASP5 expression was unaltered and independent of ACLF ( Figure 2B ). Consistently, no linear correlation was observed between CASP4 and CASP5 ( Figure 2C ). Analogous findings were obtained from purified CD14⁺ monocytes ( Supplementary Figures S1A–C ), pointing to this leukocyte type as a major involved cellular player. Indeed, we observed a significant increase in caspase-5 in the CD14-enriched population ( Figure S1B ), which was not apparent in PBMCs ( Figure 2B ), suggesting a specific regulatory pattern of caspase-5 in CD14⁺ monocytes (28). Regardless of the underlying mechanisms, the change in caspase-5 was invariably in the opposite direction to that of caspase-4. To determine whether ACLF patients did indeed suffer from immunosuppression, we measured the expression of miR-222, miR-221 (6), and HLA-DRA, which encodes the alpha subunit of HLA-DR and is routinely assessed as biomarker of immunosuppression in sepsis (2, 4, 5, 46, 48, 57) and cirrhosis (45). This analysis confirmed the presence of immunosuppression in the ACLF patient cohort ( Figures 2D–F ). HLA-DRA correlated inversely with miR-222 and miR-221 expression ( Figures 2E, F ), supporting that miR-222/221 and HLA-DRA do both reliably discriminate patients with immunosuppression. Furthermore, CASP4 and HLA-DRA were positively correlated in PBMCs ( Figure 2G ) and CD14⁺ monocytes ( Supplementary Figure S1D ), confirming the association between low CASP4 levels and immunosuppression. Interestingly, although CASP5 expression correlated with HLA-DRA in the whole cohort of patients ( Figure 2H ), this link was lost in the subgroup of patients with ACLF ( Figure 2I ). Virtually the same correlations were obtained in CD14⁺ monocytes ( Supplementary Figures S1D, G ). These data indicated that CASP4 but not CASP5 is selectively downregulated in myeloid cells during immunosuppression associated to severe disease courses with MOF.

The transcription factors IRF1 and IRF2 mediate non-canonical inflammasome activation by driving the expression of CASP4 and GSDMD genes at steady state and during infection (39, 40). To understand the mechanisms of CASP4 regulation during immunosuppression, we examined the methylation pattern of the CASP4 promoter. First, we confirmed the presence of IRF1 and IRF2 recognition sequences in the CASP4 promoter region using the Pscan algorithm (58) ( Supplementary Figure S2A ). The CASP4 promoter sequences recognized by IRF2 and IRF1 appeared to be palindromic ( Supplementary Figure S2B ) and included three putative methylation-sensitive CpG motifs: cg05618647, cg16669455 and cg16315582 ( Supplementary Figure S2A ). DNA methylation was analyzed in PBMCs of healthy donors (n = 5), patients with cirrhosis and ACLF (n = 5), and patients with cirrhosis without ACLF (n = 5). Artificial dilution series validated a detection limit of 10% methylation for Sanger sequencing of bulk PCR products. None of the analyzed clinical PBMC samples showed detectable methylation (exemplary data shown in Supplementary Figure S2 ). These data did not support a major contribution of promoter methylation to the reduced CASP4 expression in patients with ACLF. Instead, we detected downregulation of IRF1 and IRF2 in patients with ACLF ( Figures 3A, B ). Accordingly, IRF1 and IRF2 correlated across the entire patient cohort with each other ( Figure 3C ) and with HLA-DRA ( Figures 3D, E ), indicating that IRF1/IRF2 levels depended on the immune status. Of note, the expression of IRF1 and IRF2 positively correlated with CASP4 ( Figures 3F, G ) but not CASP5 ( Figures 3H, I ), supporting the notion that attenuated expression of IRF1/IRF2 mediates downregulation of CASP4 in immunosuppressed patients. In addition, the expression of GSDMD was suppressed in patients with ACLF ( Figures 3J ) correlating with the expression of IRF2, IRF1, HLA-DRA and CASP4 ( Figures 3K–N ) but not CASP5 ( Figure 3O ). Analogous expression and correlation profiles in CD14⁺ monocytes corroborated the findings obtained in PBMCs ( Supplementary Figures S1E–I, J–L ).

To determine the extent to which disease severity, as reflected by MOF, truly correlated with the degree of immunosuppression in ACLF patients, we tested for correlation between HLA-DRA, as an immunosuppression marker, and IRF2 and IRF1 as CASP4 regulators with the disease severity score MELD. HLA-DRA, IRF2 and IRF1 significantly correlated with MELD ( Supplementary Figures S3A–E ), as did the mononuclear expression profiles of CASP4 and GSDMD. However, the MELD score did not correlate with CASP5 ( Supplementary Figure S3F ). These data were consistent with the findings from Figures 2A, B , 3J and confirmed that inflammatory organ damage is accompanied by the down-regulation of CASP4 and GSDMD, but not CASP5. Since leukocytosis is a common hallmark of sepsis and severe inflammatory conditions, we also tested for correlation with white blood cell counts (WBCs) as a surrogate of systemic inflammation and found the same strong correlation with HLA-DRA, IRF2, IRF1, CASP4, and GSDMD ( Figures S3G–J, L ), but not CASP5 ( Figure S3K ). Collectively, these data demonstrate that selective suppression of human inflammatory caspase-4 is related to disease severity and associated with downregulation of the transcriptional activators IRF1 and IRF2. Finally, analysis of cleaved GSDMD in sera from patients with acute decompensation (AD) without ACLF versus with ACLF recapitulated the pattern observed in patients with sepsis (compare Figures 1E, F above), i.e. an increased activation/cleavage of GSDMD in immunosuppressed ACLF patients, reflecting inflammasome activation, in a background of overall reduced transcriptional expression of GSDMD ( Figure 3P ).

Various studies have shown that ET can be prevented by treatment with IFN-γ (6, 59–61). To investigate the role of CASP4 in IFN signaling in that context, we challenged human monocyte-derived macrophages (MDM) from healthy donors with LPS and IFN-γ in different combinations. Consistent with literature (6, 59), prolonged exposure of macrophages to LPS over days leads to a decrease in TNF production, reflecting the onset of ET ( Figure 4A ). Co-stimulation with IFN-γ boosted the LPS-induced TNF production, consistent with the notion that IFN-γ counteracts ET. Analogously, IFN-γ counteracted the LPS-induced down-regulation of HLA-DR in macrophages at the protein ( Figures 4B, C ) and transcriptional levels ( Figure 4D ). While the IFN-γ dependent rescue was consistently observed at the protein and gene level, we noticed a disparity when looking at the effect of LPS alone, as an apparent increase in HLA-DR surface protein ( Figure 4C ) was not reflected at the transcriptional level ( Figure 4D ). We hypothesize that different mechanisms of LPS-induced regulation of gene expression versus protein level may account for this discrepancy. Importantly, short-term LPS treatment induced CASP4 expression in MDMs ( Figure 4E ), in line with previous reports (54) while prolonged exposure to LPS over 24h suppressed CASP4. Interestingly and analogous to HLA-DRA regulation, the reduction in CASP4 expression was prevented by IFN-γ. The same response pattern was observed for GSDMD gene expression ( Figure 4F ), indicating that IFN-γ was able to rescue the expression of several components of the inflammasome in cells committed to tolerance. We further sought to clarify the association of IRF1/IRF2 with CASP4 and GDSMD expression in ET. Unlike IRF1, IRF2 was upregulated upon acute stimulation with LPS ( Figures 4G, H ). However, IRF2 levels saturated and started to decline 24 h post LPS challenge ( Figure 4G ), suggesting that IRF2 was sensitive to tolerization. Simultaneous treatment with IFN-γ prevented the decline in IRF2 levels and, strikingly, promoted a robust activation of IRF1 ( Figure 4H ). These data suggested that the IFN-γ dependent reversal of CASP4 tolerization in human macrophages is mediated cooperatively via IRF1 and IRF2, consistent with findings by Benaoudia et al. (39). To confirm if the observed changes in gene transcription translated to protein levels and activity, we assessed protein expression of MDM cell extracts and supernatants ( Figures 4J, K ). These results confirmed the stimulatory effect of IFN-γ when added in conjunction with LPS at the level of caspase-4, IRF1 and GSDMD activation, as monitored by the levels of cleaved p30-GSDMD polypeptide in the MDM supernatant. We were unable to detect cleaved caspase-4, probably due to technical limitations of the antibodies used. Notably, IFN-γ alone showed a marked effect on protein levels of inflammasome components. Finally, we monitored cell death in the same MDM samples, as a readout of GSDMD activity and surrogate of inflammasome-dependent pyroptosis. As shown in Figure 4I , IFN-γ enhanced the low levels of pyroptotic cell death in LPS-treated MDMs, confirming that IFN-γ counteracts LPS-induced downregulation of the non-canonical inflammasome in a setting of immunosuppression. Taking everything together, these data show that CASP4 behaves as a bona-fide ET-sensitive factor, suppressed by long-term LPS exposure and restored by IFN-γ.