Overall, PF4 and APC concentrations were measured in serum of 22 patients with ALF and 12 healthy subjects using a human ELISA Kit (R&D Systems and Biozol, respectively) following the manufacturer’s instructions. The demographic and clinical data of these patients are given in Supplementary Table 1. The study has been approved by the local ethics committee and all participants have granted informed consent prior to inclusion.

Male C57BL/6 wild-type mice (6–8 weeks) were purchased from Charles River Laboratories. PF4^-/- animals with a target deletion of the PF4 gene were established on the C57BL/6 background as described (Sachais et al., 2007) and backcrossed for more than 10 generations. All in vivo experiments were performed after approval by the Animals Welfare Review Board and by the German legal authorities of North-Rhine Westfalia (LANUV). Acute liver injury was induced either by a single intraperitoneal injection of D-GalN (700 mg/kg body weight, Roth) and LPS (10 μg/kg body weight, Sigma) for 2 and 6 h or by a single intraperitoneal injection of CCl₄ (0.6 mL/kg body weight, Merck) for 24 h following standard protocols (Hamesch et al., 2015; Scholten et al., 2015). For macrophage depletion, liposome-encapsulated clodronate, termed clodrolip, and control liposomes were injected intraperitoneally at 100 mg per 1 kg body weight, 24 h prior to GalN/LPS administration. In a separate experiment, PF4^-/- and wild-type mice were treated with recombinant mouse PF4 (1 mg/kg body weight, Biomol) parallel to GalN/LPS (Kowalska et al., 2011). Liver injury was histologically assessed by Hematoxylin Eosin (H&E) staining. Serum AST and ALT were measured as previously described (Zaldivar et al., 2012). Plasma concentrations of APC were measured using a mouse ELISA kit (Hölzel Biotech) following the manufacturer’s instructions.

TUNEL assay (Roche) was used to detect cell death histologically. The percentage of the TUNEL⁺ area was quantified in three independent fields per liver using the ImageJ software as described before (Zaldivar et al., 2012).

Isolation of total hepatic protein and RNA from snap-frozen liver tissue samples was performed as previously described (Berres et al., 2010). After SDS gel electrophoresis and wet blotting of total protein, cleaved caspase-3 protein was detected by a rabbit anti-mouse cleaved caspase-3 antibody (Cell Signaling) followed by horseradish peroxidise-conjugated anti-rabbit IgG (DAKO) and ECL substrate (Thermo Fisher Scientific). β-Actin was used as loading control. Intrahepatic and serum PF4 concentrations were determined using a murine ELISA kit (R&D Systems) following the manufacturer’s instructions. Quantitative RT-PCR (RT-qPCR) was carried out for TGF-β, TNF-α, IL-6, IL-10, and KC with Assays-on-Demand (Applied Biosystems). β-Actin was used as reference gene.

Five-μm frozen liver tissue sections were fixed and stained with F4/80, CD11b (both obtained from AbD Serotec), Ly6G (Thermo Fisher Scientific) and CD41 (BD Biosciences) antibodies followed by a suitable secondary antibody conjugated with Alexa fluor^® 488 or Cy3 (both Thermo Fisher Scientific). Cell nuclei were counterstained with DAPI (Vector Laboratories). The percentage of the stained area was quantified in three independent magnification fields per mouse (Berres et al., 2010).

For flow cytometry analysis, leukocytes from harvested livers were isolated by enzymatic digestion. After density gradient centrifugation and washing, these cells were stained with fluorochrome-conjugated antibodies for CD45 (BD Biosciences), CD3, CD8, CD4, and NK1.1 antibodies (all from eBioscience). Stained cells were analyzed using a LSR Fortessa flow cytometry system (BD Biosciences) and data was quantified using FlowJo (TreeStar) (Berres et al., 2010; Zaldivar et al., 2010).

Murine primary Kupffer cells were isolated from livers of wild-type mice by collagenase digestion followed by density gradient centrifugation as described by Taura et al. (2008). Cells were purified by MACS magnetic bead separation for F4/80 (Miltenyi Biotec.). For chemokine stimulation, cells were starved in RPMI containing 0.5% FCS (starving medium) for 16 h and stimulated with LPS (500 ng/mL, Sigma) in the presence or absence of recombinant mouse PF4 (1 μM, R&D Systems) for 6 h.

To quantify the phagocytic response of Kupffer cells, E. coli expressing the red fluorescent protein DsRed were used (Berger et al., 2010). To this end, Kupffer cells (1 × 10⁶) were stimulated for 6 h with LPS and co-stimulated with PF4 as mentioned. Two hours before harvesting, these cells were challenged with 1 × 10⁷ DsRed-expressing E. coli. After staining with F4/80 antibody (BD Biosciences), phagocytic Kupffer cells were measured on a BD FACSCanto II system (BD Biosciences).

Upon isolation, the Kupffer cells were separated by MACS magnetic beads separation (F4/80, Miltenyi Biotec) (Wu et al., 2012) and exposed to LPS and PF4 for 6 h. The supernatant was harvested and TNF-α was measured using a mouse ELISA Kit (R&D Systems) following the manufacturer’s instructions. Nitric oxide was determined using a method with Vanadium(III) chloride reduction followed by the Griess reagent system (Promega) (Egan et al., 2002). Briefly, Vanadium(III) chloride reduced the sample nitrate (NO₃) to NO₂ which thereby reacted with the Griess reagent, sulphanilamide, and N-(1-naphthyl)-ethylenediamine. The absorbance was then measured photometrically at 540 nm.

Isolated and MACS-separated Kupffer cells were used to assess the PF4 expression. To this end, these cells were stimulated with LPS for 6 h and collected in Cytospin Collection Fluid (Thermo Fisher Scientific). 10⁵ cells were cytocentrifuged with Cytospin (Thermo Fisher Scientific) and then stained with PF4 and DAPI. RT-qPCR was performed for PF4 with Assays-on-Demand (Applied Biosystems). β-Actin was used as reference gene.

Statistical quantification if two groups were compared was performed with the Students unpaired t-test. Results were described as mean ± SEM. When comparing more groups, ANOVA testing with Turkey’s multiple comparison post-test was performed. At least five animals per group were analyzed. Outliers were excluded, when Grubb’s test calculated on graph pad homepage¹ calculated a significant outlier. All statistical tests were performed using Graph-Pad Prism^®. A p-value of < 0.05 was defined as significant (^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001).

To investigate the expression of PF4 in human ALF, we investigated the PF4 concentration in serum samples in respective patients and healthy controls. PF4 serum levels and platelet numbers were decreased in patients with ALF when compared to healthy individuals which was markedly linked to an increased accumulation of platelets within the liver (Figure 1A–D). The human data was confirmed in wild-type mice treated with GalN/LPS for 6 h to induce acute liver injury, showing decreased PF4 concentrations (Figure 1E). We next assessed whether these decreases in PF4 serum levels were also reflected by reduced numbers of platelets in blood. Indeed, platelets were significantly reduced after GalN/LPS treatment compared to controls (Figure 1F), supporting recent findings from mouse and human studies (Schiodt et al., 2003; Miyakawa et al., 2015). Notably, intrahepatic PF4 protein expression was significantly increased in GalN/LPS-treated wild-type mice compared to controls (Figure 1G). This alteration was associated with an enhanced accumulation of platelets within the liver (Figure 1H,I).

Next, we investigated the functional relevance of PF4 in experimental acute liver injury. PF4^-/- and wild-type mice were treated with GalN/LPS for 2 and 6 h. Compared to their wild-type counterparts, PF4^-/- mice displayed increased parenchymal damage associated with tissue necrosis and hemorrhage within 6 h (Figure 2A). Analysis of apoptosis via TUNEL Assay confirmed increased liver injury in PF4^-/- mice (Figure 2B). These mice showed more TUNEL⁺ cells 6 h after GalN/LPS challenge (Figure 2B and Supplementary Figure 1A). These changes were reflected by significantly increased serum AST and ALT levels after 6 h (Figure 2C). However, analysis of liver injury 2 h after treatment showed no obvious differences between untreated and treated mice and also between mouse strains (Figure 2A–C). Apoptosis is a crucial hallmark of GalN/LPS-induced hepatitis (Leist et al., 1995), we thus investigated intrahepatic caspase-3 activation, serving as an important key effector of apoptosis, in GalN/LPS-treated mice by Western blot analysis. In wild-type mice GalN/LPS treatment resulted in caspase-3 activation at early stages of liver injury (Figure 2D) which was further enhanced 6 h after treatment (Figure 2D). In contrast, caspase-3 activation was significantly stronger increased in PF4^-/ mice 2 and 6 h after GalN/LPS injection (Figure 2D and Supplementary Figure 1B), suggesting a direct link between PF4 and the degree of apoptotic cell death in the liver. Collectively, these unexpected data show that genetic deletion of PF4 enhances GalN/LPS-induced liver damage, despite its known role as a pro-inflammatory chemokine (Deuel et al., 1981; Scheuerer et al., 2000).

To further investigate the differences in disease progression between PF4^-/- and wild-type mice, we examined the composition of cellular infiltrates within the liver. Untreated (ctrl) and GalN/LPS treated PF4^-/- animals showed no differences in the frequency of F4/80⁺ and CD11b⁺ immune cells compared to their wild-type counterparts after 2 h treatment. In contrast, PF4^-/- mice treated for 6 h accumulated significantly more F4/80⁺ and CD11b⁺ macrophages within the liver (Figure 3A,B and Supplementary Figures 1C,D). Detailed characterization of F4/80⁺ cells showed a significant increase of CD11b⁺/F4/80^low infiltrating macrophages in PF4^-/- mice after 6h GalN/LPS treatment compared to wild-type. CD11b⁺/F4/80^high remained unchainged (Supplementary Figure 1E). Consistent with macrophage frequency, the intrahepatic mRNA expression of the inflammatory cytokines encoding for TGF-β, TNF-α and IL-6 were significantly increased in PF4^-/- mice compared to wild-type mice (Figure 3C–E). Conversely, the mRNA expression of IL-10, a cytokine with anti-inflammatory activities (Chaudhry et al., 2011), was reduced in PF4^-/- mice (Figure 3F). Of note, genetic deletion of PF4 in mice during steady-state led to low, but significantly increased mRNA levels of TNF-α and IL-10 when compared to wild-type mice (Figure 3D,F), suggesting that this chemokine is important for liver macrophage homeostasis. Progression of acute liver injury is characterized by an increase of mediators facilitating the infiltration of different immune cells types, including neutrophils and T cells (34). Indeed, mRNA expression of the neutrophil chemoattractant KC (murine IL-8) was significantly altered in PF4^-/- compared to wild-type mice (Supplementary Figure 2A). Functionally, increased KC mRNA expression was associated with severe neutrophil infiltration to the liver of PF4^-/- mice (Supplementary Figure 2B), arguing against a major role of PF4 in attracting neutrophils (Deuel et al., 1981; Zaldivar et al., 2010). However, analysis of intrahepatic T cell subpopulations by flow cytometry revealed no obvious differences in frequency and absolute cell numbers of CD4⁺ and CD8⁺ T cells between PF4^-/- and wild-type animals (Supplementary Figure 3). Overall, these findings suggest that liver resident F4/80⁺ macrophages play a functional role in the phenotype observed in our model.

After having characterized the significant impact of PF4 on disease progression in the GalN/LPS model, we analyzed whether this finding could be confirmed in another acute toxic model. We tested the effect of PF4 on disease progression 24 h after a single CCl₄ injection. PF4^-/- mice displayed a strong increase in liver damage after CCl₄ treatment compared to wild-type mice assessed by the analysis of H&E stained liver sections (Supplementary Figure 4A). This finding was further strengthened by a significant increase of TUNEL⁺ cells and the inflammation-associated genes TGF-β and TNF-α within PF4^-/- livers (Supplementary Figures 4B–D).

Platelet factor 4 is known to promote recruitment and pro-inflammatory differentiation of monocyte-derived macrophages (Scheuerer et al., 2000). We next wanted to test whether PF4 exerts anti-inflammatory effects on liver infiltrating macrophages in our model. Recently, Kowalska et al. (2007) could show that endogenous PF4 stimulates APC generation in vivo. Notably, APC has been shown to directly inhibit activation and migration of monocytes as well as TNF-α production of macrophages, thereby reducing endotoxemia and sepsis mortality (Stephenson et al., 2006; Kerschen et al., 2007; Cao et al., 2010). Thus, we next analyzed APC concentrations in plasma of PF4^-/- and wild-type mice by ELISA. We found a time-dependent increase of APC generation in wild-type mice 2 and 6 h after GalN/LPS challenge, which was completely blunted in PF4^-/- mice (Figure 4A). In patients with ALF, serum levels of APC were also enhanced compared to healthy controls (Figure 4B). These data suggest that PF4 has the potential to suppress the pro-inflammatory response of macrophages via APC.

The in vivo results suggested a functional association between the chemokine PF4 and a reduced pro-inflammatory response of macrophages. Among monocytes/macrophages, liver resident macrophages (i.e., Kupffer cells) are crucial mediators of GalN/LPS-induced hepatitis (Antoniades et al., 2008). Thus, we next assessed whether PF4 directly affects Kupffer cell biology in vitro. Three parameters were tested: phagocytosis, TNF-α and nitrit oxide (NO) release. Phagocytosis of DsRed-expressing E. coli was induced by LPS in primary wild-type Kupffer cell cultures (Supplementary Figure 5A). As depicted in Figure 4C,D, PF4 significantly abrogated the phagocytic response of LPS on these cells. Moreover, PF4 attenuated the LPS-induced TNF-α expression and NO release by Kupffer cells into the supernatant (Figure 4E,F). This data confirmed our in vivo results showing that PF4 inhibits activation of tissue macrophages. Since tissue resident macrophages express PF4 (de Jong et al., 2008; Yeo et al., 2016), we tested whether primary isolated Kupffer cells express PF4. We confirmed the expression of PF4 in primary wild-type Kupffer cells by RT-qPCR and immunofluorescence analysis. Upon stimulation by LPS, PF4 expression was further increased in Kupffer cells, whereas in PF4^-/- Kupffer cells PF4 expression was absent (Figure 5A and Supplementary Figure 5B). Remarkably, stimulation of PF4^-/- Kupffer cells with recombinant PF4 reduced its basal TNF-α release, indicating that PF4 might be a local autocrine signal maintaining Kupffer cell homeostasis. Parallel to wild-type Kupffer cells, PF4 also reduced the LPS-induced TNF-α release (Figure 5B).

Given these in vitro findings, we postulated that administration of recombinant PF4 might reverse the phenotype of GalN/LPS treated PF4^-/- mice. In these mice, injection of PF4 indeed abrogated the GalN/LPS-induced liver injury (Figure 5C) and hepatocellular death (Figure 5D,E). These functional effects of systemic PF4 also translated into reduced serum levels of AST and ALT after GalN/LPS administration (Figure 5F). This response was further confirmed by reduced hepatic mRNA levels of TNF-α and IL-6 (Figure 6A,B). Importantly, the administration of PF4 led to a significant decrease of enhanced basal mRNA levels of TNF-α and IL-6 in untreated PF4^-/- animals, again suggesting a direct effect of PF4 on liver macrophage homeostasis.

To test whether systemic administration of PF4 can alter the progression of acute liver disease in wild-type mice, animals treated with GalN/LPS were additionally treated with recombinant PF4 for 6 h. Mice with recombinant PF4 showed a strong reduction in liver injury compared to vehicle-treated mice. This amelioration in disease progression gets evident when analyzing H&E and TUNEL stained liver sections (Figure 6C,D and Supplementary Figure 6A). In accordance with the results obtained in PF4^-/- mice, serum levels of AST and ALT were significantly reduced by administration of PF4 compared to vehicle treatment (Figure 6E), findings that were confirmed by reduced caspase-3 activity (Figure 6F and Supplementary Figure 6B).

Importantly, together with reduced liver injury, mice treated with PF4 showed a strong decrease in the infiltration of F4/80⁺ (Figure 7A and Supplementary Figure 6C) and CD11b⁺ (Figure 7B and Supplementary Figure 6D) macrophages when compared to vehicle-treated mice. These observed differences were associated with a strong reduction of TNF-α and also elevation of IL-10 mRNA expression in PF4-treated mice (Figure 7C,D). However, PF4 treatment led to no differences in KC mRNA level and neutrophil influx (data not shown), suggesting that a main effect of PF4 in vivo is modulating macrophage activation which results in reduced apoptotic cell death in the liver. Based on our findings we next tested whether exogenous PF4 influences APC production. Notably, administration of PF4 or GalN/LPS alone led to a significant increase in APC levels in plasma of wild-type mice when compared to untreated animals (Figure 7E). However, the administration of PF4 concomitantly to GalN/LPS did not further increase this effect, suggesting that the dose-dependent high peak plasma concentration of APC was achieved in the earlier phase of liver injury (Kowalska et al., 2011).

To investigate the role of macrophages and macrophage derived PF4 in more detail we next depleted macrophages in WT and PF4^-/- mice. To work out the impact on acute liver injury we performed additional GalN/LPS administration for 6 h. Deletion of F4/80⁺ macrophages in the liver could already be achieved by administration of a single dose clodrolip (Supplementary Figure 7). Depletion of these immune cells led to reduced liver injury in PF4^-/- mice and also WT mice compared to animals injected with empty liposomes (Figure 8A). Reduced liver injury was further reflected by less intrahepatic TUNEL⁺ cells (Figure 8B,C) and decreased mRNA expression of TGF-β and TNF-α (Figure 8D,E). However, the hepato-protective effect of macrophage depletion was significantly more pronounced in PF4^-/- mice compared to equally treated controls. Thus, we hypothesize that the increased pro-inflammatory response of macrophages in PF4^-/- mice after administration of GalN/LPS is directly involved in increased liver damage observed in the PF4^-/-.