Revisiting the putative role of heme as a trigger of inflammation

Abstract Activation of the innate immune system by free heme has been proposed as one of the principal consequences of cell‐free hemoglobin (Hb) exposure. Nonetheless, in the absence of infection, heme exposures within a hematoma, during hemolysis, or upon systemic administration of Hb (eg, as a Hb‐based oxygen carrier) are typically not accompanied by uncontrolled inflammation, challenging the assumption that heme is a major proinflammatory mediator in vivo. Because of its hydrophobic nature, heme liberated from oxidized hemoglobin is rapidly transferred to alternative protein‐binding sites (eg, albumin) or to hydrophobic lipid compartments minimizing protein‐free heme under in vivo equilibrium conditions. We demonstrate that the capacity of heme to activate human neutrophil granulocytes strictly depends on the availability of non protein‐associated heme. In human endothelial cells as well as in mouse macrophage cell cultures and in mouse models of local and systemic heme exposure, protein‐associated heme or Hb do not induce inflammatory gene expression over a broad range of exposure conditions. Only experiments in protein‐free culture medium demonstrated a weak capacity of heme‐solutions to induce toll‐like receptor‐(TLR4) dependent TNF‐alpha expression in macrophages. Our data suggests that the equilibrium‐state of free and protein‐associated heme critically determines the proinflammatory capacity of the metallo‐porphyrin. Based on these data it appears unlikely that inflammation‐promoting equilibrium conditions could ever occur in vivo.

extravascular Hb and heme release occurs when RBCs extravasate after a tissue trauma resulting in blood vessel injury and RBC leakage, causing either a hematoma or a contusion. Over time, these extravascular RBCs break down, releasing Hb and other cellular components. In contrast, systemic release of free Hb occurs during intravascular hemolysis, such as sickle cell disease or transfusion-associated hemolysis. 4 Extensive clinical experience suggests that hematomas, contusions, and systemic hemolysis are not usually associated with an exaggerated proinflammatory response. In contrast, cases of simple contusions presenting with cardinal signs of inflammation like edema, redness, and pain would raise concerns of an infection. 5 Therefore, it would be more intuitive to assume that the accumulation of Hb in damaged tissues promotes noninflammatory or even anti-inflammatory instead of proinflammatory effects. However, paradoxical observations have been reported in the literature over the last decade. Several studies have described proinflammatory signaling activities of heme via the activation of pattern recognition receptors such as TLR4 or the inflammasome signaling pathway in leukocytes and endothelial cells. 3,[6][7][8][9][10][11][12][13] An essential caveat of these opposing hypotheses regarding the function of heme in the regulation of inflammation may be determined by its macromolecular associations. Heme is a lipophilic molecule, which forms (mu-oxo) dimers, oligomers, and larger aggregates in aqueous solutions. [14][15][16] Therefore, heme exists almost exclusively bound within a hemoprotein (eg, Hb) in biological environments. Under certain conditions, such as after the oxidation of Hb to ferric metHb (HbFe 3+ ), the association of heme with the hemoprotein (eg, globin) weakens. 17,18 However, even under these conditions, the binding of the "released" metallo-porphyrin by other macromolecules such as albumin or lipids keeps the pool of "free" heme low. 14 By this mechanism, free heme accumulation in vivo remains orders of magnitude below the conditions under which protein-free heme-solutions have been tested for proinflammatory activities in vitro.
Here, we present data obtained from several in vitro and in vivo models that support the more intuitive hypothesis that protein-associated heme originating from RBC degradation does not induce inflammation.

| Preparation of heme-NaOH
Heme was dissolved in 10 mL NaOH (100 mmolÁL À1 ) at 37°C. The pH of the solution was adjusted to pH 7.8 using phosphoric acid.
The solution was sterile-filtered (0.22 lm) and used immediately.
The purified Hb solutions that we used were described previously. metHb (ferric) was generated by incubating oxyHb (ferrous) with a 5 9 molar excess of K3[Fe(CN)6] at room temperature for 10 minutes, followed by purification on a PD-10 column (GE Healthcare).
Hemoglobin concentrations and oxidation states were determined by spectral deconvolution, as described. 19 Hemoprotein and heme concentrations are indicated as molar concentrations of heme.

| Macrophage differentiation and stimulation
Mouse bone marrow-derived macrophages (BMDMs) were expanded and differentiated in RPMI medium supplemented with 10% FCS in the presence of recombinant mouse M-CSF (Peprotech) for 7 days.
At the end of the culture period, 100% of adherent cells were positive for F4/80 antigen.

| Neutrophil stimulation and flow cytometry
Fresh human donated blood was purchased from the Swiss Red Cross (SRK Blutspendedienst, Schlieren). The buffy coat was harvested in DPBS MgCl 2 /CaCL 2 (Gibco #14040) containing 500 IE/mL heparin by washing three times with DPBS MgCL 2 /CaCL2 and centrifuging at 400g for 10 minutes at 4°C. After the final wash, the supernatant was aspirated, and the cells were stimulated with heme at 37°C. Cells were put on ice immediately, and after centrifugation at 400g for 10 minutes at 4°, cells were stained with CD15-FITC antibody (Ab; Miltenyi Biotec #130-081-101), APC-CD14 (BD Biosciences #555399), and PE-CD62L (BD Biosciences #341012). Erythrocytes were lyzed with lyzing buffer (8.29 g ammonium chloride, 1 g potassium hydrogen carbonate, 0.037 g Na-EDTA per L) at room temperature for 1 minute.

| Western blotting
Samples were prepared with 5 lg of total protein per well in L€ ammli buffer. The western blot protocol was previously described and

| Bio-Plex cytokine assays
Concentrations of TNF-alpha, IL-6, and IL-8 were determined using Bio-Plex Cytokine Assays (Bio-Rad). The assays were analyzed with a Bio-Plex 200 system (Bio-Rad). Results were analyzed using Bio-Plex Data Pro software (Bio-Rad).

| Study approval
All experimental protocols were reviewed and approved by the Veterinary Office of the canton of Z€ urich. All animals were maintained at the animal facility of the University of Zurich and were treated in accordance with guidelines of the Swiss federal Veterinary Office.

| Solution characteristics determine neutrophil activation by free heme
Neutrophil granulocytes are known to be very sensitive to exposure to diverse inflammatory stimuli that ultimately trigger a conserved activation response, including the shedding of cell surface E-selectin (SELE).
Exposure of washed human peripheral blood leukocytes to proteinfree crystalline heme that has been dissolved in NaOH (100 mmolÁL À1 ) with subsequent adjustment to pH 7.8 resulted in a drastic reduction in cell surface CD62L expression on CD14 low CD15 high neutrophil granulocytes after 30 minutes of exposure ( Figure 1A). This effect was comparable to the effect induced by TNF-alpha. The effect of heme was dose-dependent with 100% shedding at micromolar concentrations, as demonstrated by the logistic regression model shown in Figure 1B (aggregated data from 5 healthy blood donors with heme exposure assessed at 10 concentrations from 10 À8 to 10 À1 mol/L, see also Figure S1). In the next set of experiments, we compared the neutrophil-activating effect of NaOH-dissolved heme with the effects of more physiological protein-associated heme, namely, heme-albumin and metHb (HbFe 3+ ) ( Figure 2C). Protein association of heme with albumin or Hb shifted the dose-response curve for neutrophil CD62L shedding to the right by several orders of magnitude. By modeling the equilibrium concentration in the experiments performed with heme-albumin we could show that the free heme is the active neutrophil-stimulating component, which is inactivated by protein association ( Figure S2). With metHb, significant shedding was only observed at heme concentrations above 1 mmolÁL À1 , which corresponds to >16 mg/mL metHb. The shedding of CD62L correlated with the nuclear condensation of granulocytes observed 2 hour after stimulation ( Figure 1D). It is important to note that these experiments were performed in serum-and protein-free conditions, meaning that there were no alternative heme-association sites available in the incubation medium.
F I G U R E 1 Effect of heme solution characteristics on granulocyte activation. (A) Granulocytes were identified by flow-cytometry as the CD14 low CD15 high population. Surface expression of CD62L was analyzed in response to a 30-minute exposure to TNF-alpha or NaOH-dissolved heme. Reduction of CD62L-signal indicates shedding. (B) Dose-effect relationship of NaOH-dissolved heme concentration and CD62L-shedding (fraction of CD62L-negative granulocytes). Shown is the mean AE SD of the logistic regression of data obtained from dose-response experiments performed with blood from five healthy donors (10 heme concentrations were tested per experiment). (C) CD62L shedding induced by various heme formulations: protein-free heme (NaOH-dissolved heme) in red, heme-albumin in blue, and metHb in brown. (D) DNA (DAPI) stains of nuclei of neutrophil granulocytes exposed to heme at the concentrations indicated by the dots in Figure 1C. While at 16 lmolÁL À1 , heme-albumin induces virtually no nuclear condensation, NaOH-dissolved heme exerts an apoptotic effect. The same can be observed at much higher concentrations (250 lmolÁL À1 ) when comparing heme-albumin to metHb. (E) Stopped-flow experiments of heme to hemopexin transfer. Heme transfer to hemopexin can be considered a sensitive method for detecting free (ie, solubilized) heme. The fast fraction represents free heme, the intermediate fraction represents heme only loosely bound to proteins or present in aggregates, and the slow fraction represents heme tightly bound to proteins. The distribution of heme to these three compartments is visualized by the bar plots and is distinct for all three heme formulations VALLELIAN ET AL.

| 5 of 15
To estimate concentrations of free heme or quasi-free heme (eg loosely protein-associated) in the different experimental heme solutions, we used stopped-flow spectrophotometry to measure the fraction of heme that was available for rapid hemopexin binding ( Figure 1E). In NaOH-dissolved heme, which induced CD62L shedding at micromolar concentrations, approximately 80% of the heme demonstrated an ultra-fast binding pattern (heme-hemopexin complex formation completed within <100 milliseconds). In contrast, the ultra-fast hemopexin-binding fraction of heme was reduced to less than 10% in heme-albumin. In solutions of metHb, the proportion of free heme available for ultra-rapid hemopexin binding was found to be very low. These observations suggest that in any physiological scenario in which heme might be present in extracellular spaces as a component of a natural hemoprotein, the concentration of free or quasi-free heme can be expected to be very low, possibly below the minimum range required to trigger granulocyte activation and other innate immunity responses.

Heme-albumin
Heme-NaOH Heme response  Figure 2A). In contrast, less ATP depletion was observed during exposure to heme-albumin, and no effect could be detected when either cell type was exposed to metHb or oxyHb at concentrations of up to 500 lmolÁL À1 , even in the absence of serum. When the identical experiments were performed in the presence of serum (2% FCS in endothelial cells and 10% FCS in macrophage experiments), the heme-induced effects were further reduced or completely eliminated.
We also confirmed the critical role of protein association in cellular heme effects in a metabolic experiment using a Seahorse extracellular flux analyzer ( Figure 2B). This experiment demonstrated that under serum-free conditions, 4 hours of exposure to NaOH-dissolved heme induced complete deterioration of mitochondrial respiration at concentrations as low as 10 lmolÁL À1 , while exposure to albumin-associated heme in the presence of 2% FCS stimulated respiratory capacity to levels above baseline.
The expression of heme oxygenase (HMOX1) can be considered a direct signal of intracellular heme translocation in healthy cells. In endothelial cells, we found that, in the absence of serum, heme-albumin triggered a dose-dependent HMOX1 protein expression signal in an exposure range of 0-175 lmolÁL À1 . In contrast, exposure to NaOH-dissolved heme resulted in virtually no signal, presumably as a result of the strong toxicity of the treatment. Heme translocation with subsequent HMOX1 induction was very limited upon treatment with metHb or oxyHb ( Figure 2C). This observation is consistent with the very low free heme fraction in these solutions. Heme-albumin is also a robust inducer of dose-dependent HMOX1 expression in BMDMs ( Figure 2D). HUVECs and from HUVECs exposed to either 150 lmolÁL À1 heme-albumin or IL-1b revealed a strongly dichotomous response pattern triggered by the two treatments. The group of genes regulated by the archetypical proinflammatory stimulus IL-1b displayed virtually no overlap with the genes that were regulated by heme-albumin. An enrichment analysis for gene functions revealed that IL-1b-regulated genes (IL-1b up, red label) belonged to categories such as immune response, chemotaxis, cell adhesion, and inflammation ( Figure 3B). In contrast, no significant inflammationrelated enrichment was observed for the heme-albumin-triggered response genes (heme-up, blue label), which were more characteristic of categories such as response to unfolded proteins. More specifically, as shown in Figure 3C, the top IL-1b-induced genes were not induced by heme-albumin at 25, 75, or 150 lmolÁL À1 .
Heme-albumin, in contrast, induced HMOX1 in a dose-dependent manner. The strictly dichotomous responses triggered by the archetypical inflammatory stimulator IL-1b and by heme-albumin were confirmed by RT-PCR analysis of the expression of SELE, ICAM1, VCAM1, CCL20, HMOX1, and the oxidative response gene ATF3 ( Figure 3D). In the presence of 2% FCS, none of the initially explored heme compounds (NaOH-dissolved heme, heme-albumin, metHb, or oxyHb) induced appreciable secretion of IL-8 or upregulation of cell-surface VCAM1 over an exposure range of 10-500 lmolÁL À1 for 6 hours ( Figure 3E).   response. In addition, a more detailed analysis reveals no significant induction of LPS-induced genes by heme-albumin ( Figure 4C). As shown in Figure 4D, RT-PCR gene expression data for TNF-alpha, IL-6, Spic, and Hmox1 from an independent set of experiments were consistent with these results. We have also repeated these studies, yielding identical results, with Hmox1-deficient macrophages. These experiments exclude the possibility that exaggerated expression of Hmox1 could explain the absence of inflammation ( Figure 4E).
We also measured the secretion of the inflammatory mediators TNF-alpha and IL-6 from mouse BMDMs into the culture medium during treatment with NaOH-dissolved heme, heme-albumin, oxyHb, and metHb for 18 hours (culture medium contained 10% FCS).
Again, compared to the archetypical proinflammatory stimulus LPS, we did not detect appreciable cytokine secretion under most conditions ( Figure 5A). However, a weak TNF-alpha signal appeared in macrophages that were treated with the highest concentration of NaOH-dissolved heme (500 lmolÁL À1 ). We therefore repeated this experiment in serum-free medium and found the TNF-alpha secretion response shifted to the left with a maximum signal at 50 lmolÁL À1 heme ( Figure 5B). At higher heme concentrations, TNFalpha secretion declined due to increasing toxicity, as indicated by a In the second model, we injected phenylhydrazine (PHZ) into mice to induce severe and acute hemolysis, with a drop in hematocrit from 48% to 31% within 48 hours ( Figure 6D). This treatment was accompanied by drastic increases in heme concentrations in F4/ 80 + liver macrophages ( Figure 6C). However, compared to treatment with LPS, systemic hemolysis did not provoke enhanced expression of TNF-alpha mRNA or IL-6 mRNA in F4/80 + Kupffer cells of the liver ( Figure 6E).

| DISCUSSION
The overall framework of Hb-mediated disease is based on three predominant hypotheses, which link specific biochemical properties of Hb and heme with pathophysiological processes that can occasionally be observed in patients with hemolysis or at sites of local Hb accumulation. 2 The first process links the nitric oxide (NO) leading to the accumulation of ferric metHb(Fe 3+ ) in tissue. 24 In subsequent reactions, metHb or metHb-derived heme participates in redox chain reactions that lead to the accumulation of modified lipids and proteins, as well as to heme degradation and to the release of free iron. 19,25 Ultimately, these processes can result in oxidative tissue injury, which may be accompanied by a tissue regenerative response or by secondary inflammation. 26 The third process suggests that free heme could be recognized by innate immunity receptors as an endogenous DAMP, which may directly activate inflammatory signaling pathways in leukocytes and endothelial cells.
This mechanism has been suggested to contribute to the chronic inflammatory state that is observed in sickle cell disease. 10 Furthermore, various studies in sickle cell disease mouse models have demonstrated that the injection of heme solutions can trigger endothelial hyperactivation, leading to vasoocclusion or an acute chest syndrome-like phenotype. 11,27 Purified heme was found to be an activator of TLR4 8,11,27 in some studies and of the inflammasome 9 in others, and these activities were considered to be the molecular mechanism behind the coexistence of inflammation and hemolysis.
Our current study focused on the third pathophysiological process and aimed to systematically address in different cell culture and animal models the putative proinflammatory activity of hemopro- Only under protein-free conditions did we observe a limited heme-induced TNF-alpha response in cultured macrophages, which was triggered via signaling of the classical TLR4-MyD88-TRIF pathway of NF-kB activation. 28 However, even this response was small compared to the LPS response observed in the same cells, and it occurred only at the boundary of heme-induced cytotoxicity. To further support the hypothesis that the extent and nature of macromolecular associations determine whether heme does or does not function as an activator of immune cells, we explored heme-and hemoprotein-triggered activation of neutrophil granulocytes in washed peripheral whole blood cells. In the absence of serum, this model appears to be extremely sensitive for the study of the doseresponse relationships of heme-triggered leukocyte activation. 6,29,30 Our experiments demonstrated that as free heme concentrations decline with the increasing availability and strength of heme-protein associations (protein-free heme ( heme-albumin < metHb), neutrophil CD62L shedding disappears. Our studies were not designed to structurally characterize the neutrophil-activating species in protein-free heme solutions. However, in the absence of macromolecular associations, such as in an aqueous solution of purified iron protoporphyrin IX, aggregates of heme are known to form. We also found by dynamic light scattering that protein-free aqueous heme solutions at neutral pH values contained aggregates with a particle size of up to 1000 nm (mean diameter, 320 nm; data not shown).
Such aggregates, which were not detectable in protein-associated heme solutions, are a strong candidate for an innate immunity receptor activator. Alternatively, the complete absence of other proteins may provide an environment for low-affinity interactions of mono-F I G U R E 6 Effects of heme on inflammatory pathways in vivo. (A) Subcutaneous Matrigel plugs were explanted 5 days after implantation and stained for Ly6 (granulocytes), F4/80 (macrophages), and iNOS. Plugs were enriched with LPS (1 lg/mL), metHb (1 mmolÁL À1 ), or hemealbumin (1 mmolÁL À1 ). (B) The cellular infiltrate from control, heme-albumin and LPS enriched plugs were isolated at day 5 after injection by laser-microdissection (LCM), and TNF-alpha mRNA expression was analyzed by RT-PCR. Relative expression levels were normalized for HPRT mRNA expression and are displayed as 2 ÀdCT (TNF À HPRT) . (C) FACS analysis of F4/80 expression on liver cells before and after purification with F4/80-conjugated magnetic beads. (D) Hematocrit levels of mice treated with saline (control) or phenylhydrazine (PHZ 90 mg/kg), determined 48 hours after treatment. (E) Analysis of heme content, TNF-alpha mRNA expression, and IL-6 mRNA expression in F4/80 + macrophages of the liver from control mice and mice treated with PHZ or LPS (0.5 mg/kg) for 4 hours VALLELIAN ET AL. | 13 of 15 or oligomeric heme with TLR4 or with components of the inflammasome in vitro.
The absence of a liver macrophage inflammatory response during severe intravascular hemolysis and the absence of inflammatory activation of invading macrophages in heme-enriched Matrigel plugs substantiated our principal findings in two in vivo models.
Our in vitro and in vivo studies suggest that physiological heme is likely not a bona fide activator of inflammation. However, it might still be possible that heme or Hb could alter inflammation in more complex models of sequential priming and activation processes or as a synergistic agonists with inflammatory activators. 31,32 Heme could, for example, enhance priming of leukocytes for subsequent activation by classical inflammatory stimuli. As an example of a synergistic model it has been observed that Hb could enhance macrophage activation by multiple TLR agonists by a so far undefined mechanism. 33