Connexin-43-dependent ATP release mediates macrophage activation during sepsis

Bacterial spillage into a sterile environment following intestinal hollow-organ perforation leads to peritonitis and fulminant sepsis. Outcome of sepsis critically depends on macrophage activation by extracellular ATP-release and associated autocrine signalling via purinergic receptors. ATP-release mechanisms, however, are poorly understood. Here, we show that TLR-2 and −4 agonists trigger ATP-release via Connexin-43 hemichannels in macrophages leading to poor sepsis survival. In humans, Connexin-43 was upregulated on macrophages isolated from the peritoneal cavity in patients with peritonitis but not in healthy controls. Using a murine peritonitis/sepsis model, we identified increased Connexin-43 expression in peritoneal and hepatic macrophages. Conditional Lyz2cre/creGja1flox/flox mice were developed to specifically assess Connexin-43 impact in macrophages. Both macrophage-specific Connexin-43 deletion and pharmacological Connexin-43 blockade were associated with reduced cytokine secretion by macrophages in response to LPS and CLP, ultimately resulting in increased survival. In conclusion, inhibition of autocrine Connexin-43-dependent ATP signalling on macrophages improves sepsis outcome.


Introduction
Sepsis is associated with high mortality and was now recognised as a Global Health Priority by the World Health Organization (Reinhart et al., 2017). Mechanistically sepsis is characterised by a dysregulation of the host immune response to bacteria resulting in local and systemic exacerbated proinflammatory response and a concomitant protracted anti-inflammatory response (Delano and Ward, 2016;Hotchkiss et al., 2013). Accordingly, immuno-regulatory drugs have been considered as a potential treatment, however, clinical results have been unsatisfactory so far (Remick, 2003). Despite the urgency to improve patient outcomes and major advances in the understanding of sepsis, a specific therapy is still lacking.
Peritoneal sepsis, as a result of a hollow-organ perforation, is typically associated with invasion of the peritoneal cavity by gut contents containing intestinal bacteria. Intestinal commensals and pathogens proliferate then in the peritoneal cavity and release pathogen associated molecular patterns that interact with pathogen recognition receptors on local macrophages and neutrophils to initiate strong inflammatory reactions (Iwasaki and Medzhitov, 2004).
Macrophage activation is critically modulated by extracellular ATP and associated purinergic signalling via autocrine/paracrine mechanisms (Idzko et al., 2014). During sepsis, purinergic signalling regulates core inflammatory cell functions such as migration and inflammatory mediators production (Idzko et al., 2014;Ledderose et al., 2018;Ledderose et al., 2016). ATP depletion or blockade of purinergic receptors were shown to increase survival and improve outcome of sepsis in rodent models (Cauwels et al., 2014;Csó ka et al., 2015). The initiating event triggering purinergic signalling is the release of extracellular nucleotides and includes active vesicular and channel-mediated release in addition to the passive release from necrotic cells (Dosch et al., 2018). Active ATP release from inflammatory cells can occur both via vesicular exocytosis or via connexin or pannexin hemichannels, mainly Connexin-43 (CX43) and pannexin-1 (Dosch et al., 2018;Junger, 2011). In sepsis, it is still unknown which mechanisms mediate ATP release and if targeting these mechanisms would impact outcomes. In addition, it remains to be described which inflammatory cells are involved in ATP release and to what extend ATP determines autocrine/paracrine inflammatory cell regulation during sepsis.
By screening release mechanisms of ATP using small molecule blockers, we have identified CX43 to be critical in peritoneal sepsis. CX43 is a gap junction protein forming either gap junctions or hemichannels and has been shown to mediate ATP release from inflammatory cells including macrophages (Csó ka et al., 2015;Kang et al., 2008). CX43 full knockout is perinatally lethal in mice due to cardiac malformation (Eckardt et al., 2004;Liao et al., 2001). Thus, in order to specifically explore the function of CX43 on inflammatory cells, we developed a conditional MAC-CX43 KO mouse (Lyz2 cre/cre Gja1 flox/flox ), in which CX43 is specifically deleted on macrophages and neutrophils. We identified that peritoneal macrophages express high CX43 levels in mice and humans and that CX43 expression on macrophages critically controls ATP release, local and systemic inflammation, and ultimately survival in a model of abdominal peritonitis via the purinergic receptor P2Y1. Our findings suggest that CX43 represents a new potential target for sepsis therapy.

Small peritoneal macrophages release extracellular ATP via Connexin-43 during peritonitis
Inflammatory cell populations present in the peritoneal cavity during peritonitis were characterised using CLP as a murine model of peritoneal sepsis and include infiltrating small peritoneal macrophages (SPM) and resident large peritoneal macrophages (LPM) (Ghosn et al., 2010;Wang and Kubes, 2016). Ten hours after CLP, SPM populated the peritoneal cavity while the fraction of LPM disappeared ( Figure 1A-B, Figure 1-figure supplement 1A-B). Additionally, two large populations of neutrophils invaded the peritoneal cavity upon sepsis while dendritic cells disappeared. Given the observation that SPM specifically populate the peritoneal cavity after CLP, we tested the ability of macrophages to secrete ATP. Isolated C57BL/6 wild-type (WT) peritoneal macrophages actively released ATP in response to TLR-4 agonist LPS (1 mg/ml) as well as the TLR-2 agonist Pam3CSK4 in a dose-dependent manner ( Figure 1C-D). ATP released in the extracellular space was promptly degraded by ecto-nucleotidases, as lower levels of extracellular ATP were observed in absence of treatment with ARL67156 trisodium salt, an ecto-ATPase inhibitor (Figure 1-figure supplement 1C).
To understand the importance of CX43 in ATP release from macrophages and its impact on macrophage activation, we developed conditional MAC-CX43 KO (Lyz2 cre/cre , Gja1 flox/flox ) mice in which CX43 is specifically deleted in activated macrophages (Figure 1-figure supplement 2). As expected from the previous results using a specific CX43 blocker, ATP was released from WT peritoneal macrophages upon LPS stimulation but not from MAC-CX43 KO peritoneal macrophages, while intracellular ATP levels remained constant ( Figure 1H Populations were defined based on F4/80 (BUV395), CD11b (APC/Cy7), Ly6G (PE-Cy7) and CD11c (Pacific blue). In response to CLP, elevated levels of neutrophils (1)/(2) (F4/80 low , CD11b high , Ly6G high ) and small (infiltrating) peritoneal macrophages (3) (F4/80 int , CD11b int ) were observed but not large In mice, analysis of cellular fractions present in the peritoneal cavity during peritonitis revealed that CX43-dependent ATP release is macrophage-specific, since both SPM and LPM are CX43-positive, while neutrophils are CX43-negative ( Figure 2D). CX43 protein expression in WT murine peritoneal macrophages is induced by LPS ( Figure 2E Mechanistically, MyD88 and TRIF are required for the expression of CX43 after LPS stimulation ( Figure 2G-H), while it is independent of caspase-1 and caspase-11 ( Figure 2I). Thus, CX43 in SPM is responsible for LPS-induced ATP release in peritoneal sepsis in a MyD88/TRIF dependent manner.

Connexin-43 expressing macrophages are recruited systemically during peritonitis
Sepsis is a systemic inflammatory reaction that leads to extensive organ damage remote from the primary infection site, typically in the liver and the lungs. To understand whether CX43 positive macrophages would be present locally at the site of infection only or in remote organs as well, we analysed hepatic and pulmonary tissues in our CLP model. In the liver of controls, CX43 was expressed mainly on parenchymal cells including cholangiocytes, while after CLP it was highly upregulated on infiltrating cells in a time-dependent manner ( Figure 3A-B, Figure 3-figure supplement 1A). CX43 protein expression in the liver was dependent on MyD88/TRIF activation ( Figure 3C). Following CLP, infiltrating macrophages and neutrophils increased in the liver while resident macrophages (Kupffer cells) remained constant ( Figure 3D-E). CX43 expression in the liver was restricted to infiltrating macrophages, while neutrophils and resident macrophages were CX43 low ( Figure 3F). Hepatic lympho-cellular populations, such as T cells, B cells, and NK cells did not express CX43 ( In the lungs, CX43 is constitutively expressed on muscle cells and resident macrophages populations ( Figure 3G). In the acute respiratory distress syndrome that is associated with severe peritonitis, alveolar infiltration of CX43 inflammatory cells was observed ( Figure 3G-H). These results reveal that CX43 expression is increased in the liver and lungs as remote target organs during sepsis. (resident) peritoneal macrophages (4) (F4/80 high , CD11b high ) or dendritic cells (5) (F4/80 low , CD11b low , Ly6G low , CD11c high ) (N = 5 animals per group). (B) Relative frequency of large (resident) peritoneal macrophages (LPM) and small (infiltrating) peritoneal macrophages (SPM) in the peritoneal cavity of controls (green) and 10 hr after CLP (orange) (N = 5 animals per group, one-way ANOVA). (C) Extracellular ATP levels in the supernatant of peritoneal macrophages after 30 min stimulation with LPS (TLR-4 agonist) or Pam3CSK4 (PAM, TLR-2 agonist). Cells were isolated from C57 Bl/6 WT mice (each dot is representative of an independent biological replicate, representative of more than five independent experiments, unpaired t-test). (D) Dosedependent ATP release by peritoneal macrophages in response to LPS as quantified by a luciferin-luciferase assay (N = 12 (control) and 6 (LPS))*. (E) LPS-induced ATP release by macrophages can be blocked by carbenoxolone ((Carbeno) combined pannexin/connexin blockade), 18-alpha-GA (18-a-GA) global connexin blockade), Gap27 (specific blockade of Connexin 43 (CX43)) but not by probenecid ((Prob) pannexin channel blocker) and N-Ethylmaleimide ((NEM) blockade of vesicular exocytosis) (N = 10 (negative and positive controls) and 5 (blockers))*. (F-G) Non-specific (18-a-GA) and specific (Gap27) blocking of connexin hemichannels decreases LPS-induced ATP release from WT peritoneal macrophages in a dose-dependent manner (N = 12 (negative control) and 6 (treated groups), negative and positive controls were the same for both experiments)*. (H) LPS-induced ATP release from WT compared to MAC-CX43 KO peritoneal macrophages with and without specific CX43 blocking (N = 8)*. (I) LPS-induced ATP release has no impact on intracellular ATP levels (N = 8)*. *Data represent independent biological replicates, compared by unpaired t-test and are representative of three or more independent experiments. DOI: https://doi.org/10.7554/eLife.42670.002 The following figure supplements are available for figure 1:  Connexin-43-mediated local and systemic ATP release contributes to macrophage over-activation via P2Y1 Levels of extracellular ATP were increased in the peritoneal cavity as well as in the systemic circulation 10 hr after CLP ( Figure 4A-B). Following CLP, we observed lower systemic ATP levels in MAC-CX43 KO mice compared to the WT mice, suggesting a role for CX43 in systemic ATP release, supporting the results in Figure 1 showing that local ATP release from macrophages is CX43-dependent ( Figure 4B). In a next step, we assessed autocrine responses of CX43-dependent ATP release in peritoneal macrophages. Pharmacological blocking (Gap27) or genetic deletion of CX43 decreased pro-inflammatory cytokines levels (TNF-alpha and IL-6) after 3 and 6 hr of LPS stimulation in peritoneal macrophages in vitro ( Figure 4C-D). Gap27 decreased TNF-alpha levels after 6 hr LPS-stimulation and IL-6 levels after 3 hr LPS-stimulation. Next, we aimed to understand whether CX43 blocking or deletion, and consequent decrease in ATP release, regulate macrophage differentiation. Functional markers of M1 differentiation, such as Inos and Il12rb, were decreased in peritoneal macrophages under CX43 blocking or deletion ( . Therefore, autocrine CX43-dependent ATP signalling alters cytokine secretion and drives M1 macrophage differentiation but not phagocytosis. To determine if cytokine secretion in response to abrogated ATP release after CX43 deletion can be reverted, ATPgammaS, a non-hydrolyzable form of ATP, was administered in vitro. Thereby the inhibited secretion of IL-6 by peritoneal macrophages was restored to control levels ( Figure 4E, Figure 4-figure supplement 1G). Administration of apyrase, a soluble ecto-ATPase consuming extracellular ATP, decreased pro-inflammatory cytokines levels. Thus, the downstream effects of LPSinduced CX43-dependent ATP release is mediated via purinergic receptors. To identify specific P2 receptors responsible for this effect, a screen using different P2 receptors blockers was performed by measuring the effects of LPS-dependent TNF-alpha and IL-6 release from peritoneal macrophages (Source data 1). The observed reversal of TNF-alpha and IL-6 secretion following unspecific P2 receptor blockade (suramin) and specific P2Y1 blockade (MRS 2279) indicate a crucial role of P2Y1 in macrophage activation following LPS stimulation ( Figure  . Extracellular ATP is hydrolysed by the ecto-nucleotidase CD39 to ADP/AMP and by ecto-5'nucleotidase CD73 to adenosine. CD39 mRNA levels were reduced in response to CX43 pharmacological blockade or genetic deletion in MAC-CX43 KO peritoneal macrophages compared to WT controls, while no difference between WT and MAC-CX43 KO was observed for ecto-5'nucleotidase CD73 (  patients without peritoneal inflammation (green) compared to patients with peritonitis (orange) (N = 5 patients in control group and N = 6 patients in peritonitis group, Mann-Whitney test). (D) Specific CX43 expression (Alexa Fluor 488) in macrophages (small and large peritoneal macrophages, SPM and LPM) among peritoneal viable (AmCyan low ) CD45 high cells (N = 5 animals per group, tSNE analysis as described in statistical methods). (E-F) Increased expression of CX43 in peritoneal macrophages following stimulation with LPS (1 mg/ml) for the indicated time-points (Western blot: N = 2 independent biological replicates per time-point, quantification of immunoblots using ImageJ, unpaired t-test. Immunofluorescence: PE-Texas red = CX43; FITC = F4/80; blue = DAPI). (G) Abrogated CX43 protein expression in peritoneal macrophages from MyD88/TRIF double KO mice upon stimulation with LPS 1 mg/ml (N = 2 mice per time-point, quantification of immunoblots using ImageJ, unpaired t-test). (H) LPS-induced ATP release from murine WT compared to MyD88/TRIF KO peritoneal macrophages (N = 7 independent biological replicates, one-way ANOVA). (I) CX43 protein expression in peritoneal macrophages from caspase 1/11 double KO mice upon stimulation with LPS 1 mg/ml (N = 2 mice per time-point, quantification of immunoblots using ImageJ, unpaired t-test). Full uncut western blot membranes are available in Source data 1. DOI: https://doi.org/10.7554/eLife.42670.005 The following figure supplement is available for figure 2:  To test the relevance of CX43 for sepsis outcome, we compared our MAC-CX43 KO mice (Lyz2 cre/ cre , Gja1 flox/flox ) to mice heterozygous for the lyzozyme two cre sequence (Lyz2 cre/wt , Gja1 flox/flox ) with or without pharmacological inhibition with Gap27 as well as sham operated controls. Murine sepsis score (Supplementary file 1C), weight loss, survival time, bacterial load, and markers of inflammation were monitored. The extent of sepsis, as assessed by the murine sepsis score, was significantly reduced in MAC-CX43 KO compared to WT mice without a difference in weight loss ( Figure 5A-B).
CLP is a lethal model with a survival time expected to be reached between 18 and 24 hr following surgery. CX43 blocking or specific genetic deletion in CLP operated mice was associated with significantly prolonged survival time compared to control WT C57BL/6 mice ( Figure 5C). Bacterial load was not different in both peritoneal fluid and blood excluding altered phagocytosis by macrophages to be relevant ( Inflammatory cytokines were assessed systemically (serum), locally (peritoneal fluid), and in key organs typically injured during sepsis (lungs, liver, kidney, intestine). In the serum, we observed decreased levels of pro-inflammatory cytokines (TNF-alpha and IL-33 but not IL-6 and IL-1 beta) in MAC-CX43 KO compared to WT mice 10 hr after CLP ( Figure 5D-F, Figure 5-figure supplement 3C). In addition, IL-10 levels were lower, indicating a decrease of both pro-and anti-inflammatory cytokines in MAC-CX43 KO mice ( Figure 5G). In the peritoneal fluid and in the peripheral effector organs, we observed decreased IL-6 levels (with the exception of the intestine) of mice with CX43 deletion or blockade compared to non-treated WT mice ( Figure 5H-K, Figure 5-figure supplement 3D). A relevant role of neutrophils can be excluded, first, because of limited expression of CX43 in response to sepsis (Figures 2 and 3) and because no alterations of myeloperoxidase activity in the liver of mice 10 hr after CLP was observed in response to CX43 blockade (Figure 5-figure supplement 3E-F).
Taken together these data suggest that CX43 regulates the secretion of inflammatory cytokines in an autocrine manner during sepsis by the release of ATP and activation of P2Y1.  (1) and infiltrating macrophages (2) increased in the liver, while resident macrophages (3) remained constant compared to controls (green). (F) Infiltrating macrophages highly expressed CX43 levels while neutrophils (1) and resident macrophages (3) were CX43 low (N = 3 animals in control group and N = 6 animals in CLP group, unpaired t-test). (G, H) Elevated expression of CX43 on non-parenchymal cells in the lungs in response to CLP compared to control mice (N = 4 animals in control group and N = 7 animals in CLP group, unpaired t-test). Full uncut western blot membranes are available in Source data 1. DOI: https://doi.org/10.7554/eLife.42670.007 The following figure supplement is available for figure 3:  (cps)). Luminescence was significantly higher 10 hr after CLP compared to sham-operated controls. (B) Direct measurement of extracellular ATP in the plasma collected from vena cava inferior is elevated in response to CLP and higher in WT mice compared to MAC-CX43 KO mice (each dot is representative for a single animal, Figure 4 continued on next page

Discussion
Following hollow organ perforation, intestinal bacteria penetrate the peritoneal cavity and generate inflammation eventually leading to sepsis ( Figure 6). Our data have shown that ATP release from peritoneal macrophages is an active process that occurs in response to TLR-2 and TLR-4 agonists in a CX43-dependent manner. Increased expression of CX43 on macrophages following TLR-4 pathway activation has been previously shown (Qin et al., 2016). Our study has extended this observation by describing the functional role of CX43 for autocrine signalling in macrophages. This is mainly relevant in infiltrating SPM, which are the main CX43-positive population during peritonitis, replacing resident LPM. Once released in the extracellular space via CX43, ATP typically signals through P2 receptors in an autocrine manner. Among the P2 receptors that have been shown to play a critical role for macrophage activation (Csó ka et al., 2015;Greve et al., 2017;Lecut et al., 2012;Maître et al., 2015), we identified the P2Y1 to be critical in a model of abdominal peritonitis. This effect mainly depends on alterations in cytokine secretion while phagocytic capacity was not changed, adding evidence to previous conflicting results on putative purine mediated phagocytosis (Anand et al., 2008;Glass et al., 2013).
We have shown that both pharmacologic blockade and genetic deletion of CX43 in activated macrophages led to increased survival compared to controls. This was associated with lower levels of key inflammatory cytokines in the acute phase of sepsis including IL-6, TNF-alpha in response to CX43 blockade or deletion. Decreased levels of IL-10 and IL-33 indicate that blocking CX43 might improve the long-term outcomes, as these cytokines have been shown to be involved in immunosuppression following sepsis (Nascimento et al., 2017). This supports previous data showing that PAMPs and DAMPs interacting with PRRs induce ATP release and finally alterations in cytokine secretion (Cauwels et al., 2014;Sakaki et al., 2013). This effect was inflammasome independent for TNF-alpha and IL-6. Furthermore, the finding that IL-1-beta secretion is not altered via the CX43/ P2Y1 axis supports inflammasome independent signalling (Gicquel et al., 2015;Sakaki et al., 2013). P2Y1 purinergic receptors are G aq -protein coupled receptors signalling via phospholipase Cb and increased cytosolic calcium levels ( Barańska et al., 2017). The function and downstream signalling of P2Y1 has been well described in platelets or neurons but not in macrophages . It is likely that downstream signalling of P2Y1 mediated TNF-alpha and IL-6 secretion in macrophages to be dependent on the same mechanisms. CX43/P2Y1 dependent modulation of IL-6 secretion may be partially dependent on TNF-alpha since IL-6-promoter is also activated by TNF-alpha among other stimuli explaining the sequence of appearance during peritonitis (Hunter and Jones, 2015;Jones and Jenkins, 2018).
In this study, plasma and peritoneal ATP levels might be underestimated in vivo, since ATP is rapidly degraded by different membrane-bound ecto-nucleotidases (NTPDases 1, 2 and 8) and soluble ecto-nucleotidases (Fausther et al., 2012;Feldbrügge et al., 2018;Yegutkin et al., 2003). This is supported by the finding that by blocking ecto-nucleotidases, we observed higher ATP levels in the extracellular space of peritoneal macrophages in vitro (Figure 1-figure supplement 1C). Altered  (Gap27 (1 mM)) or genetic deletion of CX43 decreased TNF-alpha (C) und IL-6 (D) secretion from peritoneal macrophages in response to stimulation with LPS (1 mg/ml) for 3 and 6 hr (removal of 7 outliers after ROUT test in Figure 4D)*. (E) IL-6 release was restored by activation of purinergic receptors by exogenous administration of ATPуS (10 nM) and abrogated in response to apyrase (100 IU/ml) (each dot is representative of an independent biological replicate, two-way ANOVA, representative of three experiments). (F-G) IL-6 release from peritoneal macrophages in response to LPS (1 mg/ml) and by blocking major P2X (G) and P2Y (H) purinergic receptors (N = 4)*. *Data represent independent biological replicates, compared by unpaired t-test and are representative of three or more independent experiments. DOI: https://doi.org/10.7554/eLife.42670.009 The following figure supplements are available for figure 4: purinergic signalling by MAC-CX43 KO peritoneal macrophages resulting in decreased cytokine secretion was reversed by the administration of ATPgammaS or apyrase. Therefore, CX43 may impact on ATP dependent autocrine signalling via activation or desensitisation of purinergic receptors that has been shown for P2Y1 (Gomes et al., 2009;Rodríguez-Rodríguez et al., 2009). The expression of CD39 was disturbed in response to CX43 deletion as seen in other models of inflammation (Deaglio and Robson, 2011). However, altered expression had no impact on the kinetics of ATP hydrolysis in cultured peritoneal macrophages.
The effects of CX43 observed in this study are specific for macrophages given the exclusive expression of CX43 on macrophages conversely to neutrophils or lymphocytes. Therefore, we present a novel pharmacological target to address macrophage function specifically. Clinical applicability is substantiated by human data where we identified CX43 positive macrophages in the peritoneal cavity of patients with peritonitis but not in control patients, allowing us to hypothesise that CX43 plays a comparable role in human macrophages and in mouse macrophages.
As CX43 proteins form pores that are permeable for other small molecules, typically lower than 1 kDa, the effect on the secretion of other nucleotides, including UDP (Qin et al., 2016), as well as other metabolites could influence sepsis outcome as well.
For in vitro experiments, thioglycollate was used to induce recruitment of macrophages to the peritoneal cavity. Thioglycollate-induced exudate macrophages differ from resident macrophages in metabolic activity and phagocytic activity (Pavlou et al., 2017;Shaw and Griffin, 1982). However, given the robust differences between the groups we expect that the findings are applicable to SPM that are the recruited and highly active macrophage population in peritonitis.
In conclusion, we identified that CX43 is upregulated specifically on macrophages locally and systemically during sepsis and that the elevated levels of CX43 mediate cytokine secretion but not phagocytosis of macrophages via extracellular ATP and the purinergic receptor P2Y1. Therefore, CX43 is critical for the pathophysiology of sepsis and may be a potential new target for improving the outcome of this critical clinical condition.

Experimental animals
Animals were housed in specific pathogen free (SPF) conditions. Animals used in our experiments were aged between 8 to 12 weeks and were females. WT mice were on C57Bl/6(J) background and were purchased from Harlan, Netherlands. MyD88/TRIF -/mice and caspase 1/11 -/mice were provided by Andrew MacPherson (Mucosal Immunology, University of Bern) and were bred germ-free at the Clean Mouse Facility, University of Bern, Switzerland. These mice were colonised with SPF flora for experimental purposes.

Caecal ligation and puncture
We used caecal ligation and puncture (CLP) to induce peritonitis leading to high-grade sepsis (Rittirsch et al., 2009). In brief, mice were anaesthetised with isoflurane and a longitudinal 1 cm midline laparotomy was performed to expose the caecum. The caecum was tightly ligated using a 6.0 silk suture (6-0 PROLENE, Ethicon) with 2/3 of the caecum included below the ligature and was perforated twice using an 22 gauge needle. The abdomen was closed and mice were subcutaneously injected with 1 ml of 0.9% saline. Sham-operated mice received the same surgical operation without caecum ligation and puncture. No antibiotics were administered to mice that underwent the operation. The mice were sacrificed at 5 and 10 hr for experimental purposes. A separated set of animals, including WT and MAC-CX43 KO, were used in survival studies and were sacrificed latest 72 hr after CLP. All the animals operated with CLP were monitored using a Murine Sepsis Score that was adapted from a previous publication (Shrum et al., 2014) (Supplementary file 1C). Animals were treated with analgesics (buprenorphin) or sacrificed according to their score.

Collection of blood, peritoneal fluid and organs
Prior to harvesting, mice were anaesthetised using 2 ml/kg mouse body weight of a mixed triple combination of Fentanyl (0.05 mg/ml), Midazolam (5 mg/ml) and Medetomidin (1 mg/ml). Animals were placed on a surgical tray, abdominal skin was removed and laparotomy was performed. Blood was collected from inferior vena cava using a 22 gauge catheter (BD Insyte, #381223) and a 2 ml syringe into 1.5 ml Eppendorf tubes. Blood was incubated at room temperature for 60 min and centrifuged at 1,000 g for 10 min. Supernatant was transferred in a new tube and further centrifuged at 10'000 g for 10 min. Serum was finally collected in a new tube and snap frozen in liquid nitrogen. In order to collect peritoneal fluid, 5 ml ice cold PBS was injected i.p. using a 10 ml syringe and a 23 gauge needle. The abdomen was massaged for 15 s and at least 1 ml peritoneal fluid was collected in Eppendorf tubes. Tubes were centrifuged at 10,000 g for 10 min to pellet cells present in the peritoneal fluid and supernatant was pipetted into a new tube. Both pellets and supernatants were snap frozen in liquid nitrogen for further experiments. Various organs, including liver, lungs, kidney and small intestine were collected and either further processed for flow cytometry analyses or snap frozen in liquid nitrogen for future experiments.

Isolation and culture of mouse peritoneal macrophages
Mice were injected i.p. with 2 ml 3% Brewer thioglycollate (Sigma-Aldrich, #B2551) 3 days prior macrophage isolation. In order to collect macrophages, the peritoneal cavity was washed with ice cold harvest medium (Dulbecco's phosphate-buffered saline without calcium and magnesium supplemented with 3% fetal calf serum, Sigma-Aldrich, #D8537). Collected fluids from at least five different mice were pooled and filtered through 100 mm filter and cells were centrifuged at 800 g and 4˚C for 5 min. Erythrocytes were lysed using erythrocyte lysis buffer (Qiagen, #160011922). Cells were counted using a counting chamber and were plated after resuspension in culture medium (Dulbecco's Modified Eagle Medium/F12 GlutaMAX (ThermoFisher Scientific, Gibco, #10565018) supplemented with 50 IU/ml penicillin G and 50 mg/ml streptomycin and 10% fetal calf serum). HEK293-pmeLUC cells were provided by Francesco Di Virgilio (Department of Morphology, Surgery and Experimental Medicine, University of Ferrara). HEK293-pmeLUC cells were cultured in DMEM/F12 supplemented with 10% fetal calf serum, 50 IU/ml penicillin G, 50 mg/ml streptomycin and 0.2 mg/ ml G418 sulphate (Geneticin, Calbiochem). Cells were cultured in 75 cm 2 tissue culture flasks (VWR, #734-2313) and kept in the incubator at 37˚C and 5% CO 2 . The viability of cultured cells was checked under the microscope before every cell passage. After detachment of cells using a 0.05% trypsin/EDTA solution (ThermoFisher Scientific, #15400054), cells were counted using a counting chamber (Brand, #717805). 40 Â 10 3 cells were plated on 96-wells transparent flat bottomed cell culture plates (Sarstedt, Tissue culture plate 96-wells, sterile, #83.3924.300) and 1 Â 10 6 cells were plated on 6-wells transparent flat bottomed cell culture plates (Sarstedt, Tissue culture plate 6-wells, # 83.3920.005) for experimental purposes.

Quantification of extracellular and intracellular ATP
Extracellular ATP was measured using a luciferin-luciferase assay as described in manufacturer protocol (Biothema, #144-041) and bioluminescence was quantified using a Tecan Infinite 200 plate reader (Tecan). In vitro, the supernatant was collected from 96-wells cell culture plate. The ecto-ATPase inhibitor ARL 67156 (Sigma-Aldrich, #67156) was used to reduce extracellular ATP degradation. In vivo, extracellular ATP was quantified in the peritoneal fluid and in the plasma using the same kit and 1:20 dilution in PBS. Plasma was collected in Eppendorf tubes with CTAD (citrate-theophylline-adenosine-dipyridamole) (BD Vacutainer, #367599) to avoid platelet activation and ATP release via degranulation. Intracellular ATP was quantified in peritoneal macrophages using a luminescent cell viability assay following manufacturer's protocol (CellTiter-Glo 2.0 Assay, Promega, # G9241).
In vivo ATP imaging using plasma-membrane targeted luciferase Measurements of extracellular ATP using a chimera plasma-membrane targeted luciferase on HEK293 cells were performed as previously described and by adapting the protocol for our own purposes (Pellegatti et al., 2008;Pellegatti et al., 2005). Cells were cultured as described above. 5 Â 10 6 cells were injected i.p. right after CLP or sham and before suturing the abdominal wall. 6 hr after CLP, 3 mg D-luciferin diluted in 100 ml PBS were injected i.p. Bioluminescence was measured using a gamma camera (NightOWL LB 983, Berthold Technologies) 8 min after D-luciferin injection and is proportional to the amount of ATP present in the peritoneal cavity.

Quantification of Colony forming units (CFUs) in blood and peritoneal lavage
As previously described (Csó ka et al., 2015), blood and peritoneal lavage fluid were diluted serially in sterile physiologic saline. Fifty microliters of each dilution was aseptically plated and cultured on LB agar or trypticase blood agar plates (BD Biosciences) at 37˚C. After 12-16 hr of incubation, the number of bacterial colonies was counted. The number of cultures is expressed as CFUs per millilitre of blood or peritoneal lavage fluid.
Serum cytokines were measured using electrochemiluminescence assays (Meso Scale Discovery (MSD), Rockville, Maryland, USA). Serum TNF-alpha, IL-6, IL-10 and IL-33 were measured using a multiplex assay using the V-PLEX Mouse Cytokine 19-Plex kit (MSD). Samples were diluted 1:3 in proprietary buffer (MSD) and measured in duplicate according to manufacture's protocol. Plates were run on MSD plate reader model 1250.

Neutrophil detection by myeloperoxidase assay
Myeloperoxidase assay was carried out as previously described (Fahrner et al., 2014;Mota et al., 2007). Briefly, liver samples were thawed, 50 mg of tissue was weighed out and homogenised in 1 ml of 20 mM phosphate buffer using a TissueLyser (Qiagen) 2 min at 20 Hz, centrifuged at 10,000 g for 10 min, and the pellet was resuspended in 50 mM phosphate buffer (pH 6.0) containing 0.5% hexadecyltrimethyl ammonium bromide. The samples were then freeze-thawed four times using liquid nitrogen followed by a 40 s sonication, centrifuged at 10,000 g for 10 min, and the supernatant was used for the MPO assay. The assay mixture contained 10 ml of the sample, 25 ml 3,3'5,5-tetramethylbenzidine (final concentration 1.6 mM), 25 ml H 2 O 2 in 80 mM phosphate buffer (pH 5.4, final concentration H 2 O 2 0.3 mM), and 40 ml of 50 mM phosphate buffer (pH 6.0) containing 0.5% hexadecyltrimethyl ammonium bromide. The mixture was incubated at 37˚C for 2 min, stopped with the addition of 1 M HCl, and the absorbance was measured at 450 nm.

Flow cytometry
To perform flow cytometry, leukocytes were isolated from the peritoneal cavity and from the liver. Isolation of peritoneal leukocytes was performed as described above by washing the peritoneal cavity with ice cold PBS. Leukocytes isolation from the liver was performed as previously described (Kudira et al., 2016). The liver was cut in small pieces and placed in 10 ml RPMI containing 3% FCS 1 mg/ml collagenase IV and 0.1 mg/ml DNase I (Roche) with shaking at 37˚C for 30 min. Cell suspension was passed through a 100 mm cell strainer and washed twice with FACS buffer (PBS, 3% FCS, 2 mM EDTA). Cells were centrifuged (800 g, 7 min, 4˚C) and resuspended in 40% Percoll solution and layered on top of 80% Percoll solution. Gradient centrifugation was carried out (900 g, 20 min, 4˚C, no brake). Leukocytes were collected from the interphase, washed with FACS buffer and centrifuged (800 g, 7 min, 4˚C). Cells were finally resuspended in FACS buffer and further stained for flow cytometry.
Aliquots of 10 6 cells/100 mL of staining buffer per well were incubated each with 1 mg of purified anti-CD16/CD32 for 20 min at 4˚C in the dark, in order to block non-specific binding of antibodies to the FcgIII and FcgII receptors. Cell suspension was incubated with a fixable viability dye (AmCyan, eBioscience, #65-0866-14) diluted in DPBS during 20 min at 4˚C in the dark to exclude dead cells. Subsequently, these cells were separately stained with the surface markers for 20 min at 4˚C in the dark with 1 mg of primary antibodies. For cytokines and transcription factors, cells were stained with antibodies to surface antigens, fixed and permeabilised according to the manufacturer's instructions (Foxp3/Transcription Factor Staining Buffer Set; eBioscience, 00-5523-00). Fluorescently labelled anti-mouse used are summarised below (Supplementary file 1D). Cells were subsequently washed twice with, and re-suspended in FACS buffer. Finally, cell data were acquired on a LSR II SORP H271 (BD Biosciences). Flow cytometric analysis was done using FlowJo (Treestar). In all experiments, FSC-H versus FSC-A was used to gate on singlets with dead cells excluded using the fluorescencecoupled fixable viability dye. Murine inflammatory peritoneal cells were gated as follow: (1) SPMs: viable (AmCyan low ), CD45 high , CD3 and CD19 low , Ly6G low , F4/80 int and CD11b int (2) LPMs: viable (AmCyan low ), CD45 high , CD3 and CD19 low , Ly6G low , F4/80 high and CD11b high (3) Neutrophils: viable (AmCyan low ), CD45 high , CD3 and CD19 low , Ly6G high , F4/80 low and CD11b low (4) Dendritic cells: viable (AmCyan low ), CD45 high , CD3 and CD19 low , Ly6G high , F4/80 low , CD11b low and CD11c high . Liver macrophages were characterised as published previously (You et al., 2013). Inflammatory cells in the liver were gated as follow: AmCyan low , CD45 high , CD3/CD19 low .

Time-of-flight mass cytometry (CyTOF)
For time-of-flight mass cytometry (CyTOF) experiments, cells were isolated from the peritoneal cavity as previously describe (Ray and Dittel, 2010), 3 Â 10 6 cells were stained with lanthanide-conjugated antibodies as previously described (Becher et al., 2014). Antibodies were purchased already labelled or labelled using the MaxPar antibody labelling kit (Fluidigm, Supplementary file 1F). Raw signals were normalised to median bead intensity using four elemental calibration beads (Fluidigm) as previously described (Finck et al., 2013). The Live intact single events were gated for their DNA content (MaxPar Intercalator, Ir 193) and viability 198 Pt). CD45 neg cells were excluded. Analysis was done using t-SNE for dimensionality reduction on arc sinus transformed data (Hinton, 2008). On t-SNE plots the clusters were manually gated based on their marker expression.

Phagocytic assay
The phagocytic capacity of plated peritoneal macrophages was assessed as described in the manufacturer protocol (Cayman Chemical, #500290). Peritoneal macrophages were isolated from WT and MAC-CX43 KO mice as described above. 50'000 macrophages were plated per well on a chamber slide (Nunc Lab Tek II, #154534) and incubated for 12 hr at 37˚C and 5% CO 2 . Macrophages were primed or not with LPS 1 mg/ml for 4 hr and exposed or not to FITC latex beads coated with rabbit IgG diluted 1:200 for 2 hr (four groups in total). Negative controls were incubated on a separate slide chamber at 4˚C.

Compounds
Toll-like receptors 4 and 2 agonists were used to stimulate peritoneal macrophages in vitro: lipopolysaccharides (LPS) from E.coli O111:B4 (Sigma-Aldrich, # L4130) and Pam3CSK4 (Invivogen, # tlrlpms). In order to block ATP release via hemichannels, 18-alpha-glycyrrhetinic acid (Sigma-Aldrich, # G8503) was used as a nonspecific inhibitor of connexin channels and Gap27 (Tocris Bioscience, # 1476) was used as a specific CX43 inhibitor. 18-alpha-glycyrrhetinic acid was dissolved in DMSO to obtain a solution 20 mg/ml as described in supplier protocol. The final DMSO concentration in the supernatant was 0.25%.